# Anthropic S-1: investor findings and notable disclosures

**Filing reviewed:** Anthropic, PBC, Form S-1 filed October 26, 2026. **Proposed Nasdaq symbol:** ANT (application described in the filing; this is not confirmation of completed listing). [PDF p. 2][p2] **Scope:** the complete 421-page main filing, including financial-statement footnotes, the founders’ letter, charts, customer case studies, and Part II. Separate exhibit documents are outside this review.

**How to use the references:** Every `PDF p.` reference identifies the physical page in the saved PDF, counting its first page as 1. Click it to open that page. The reference tooltip also gives the printed filing page where available. The [original SEC filing](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm) is the authoritative source. Dollar figures are U.S. dollars. H1 means the six months ended June 30; Q2 means the three months ended June 30. 2026 interim financial statements are unaudited. Facts below are reported by the company; paragraphs marked **Investor implication** are our interpretation. Customer case studies and forward-looking technical claims are identified as such.

## The findings that matter most

1. **The company has already reached substantial revenue scale.** H1 2026 revenue was **$16.286 billion**, compared with **$1.163 billion** in H1 2025. Q2 2026 revenue was **$11.555 billion**. This is recognized revenue, distinct from an annualized run rate. [PDF p. 158][p158]; [PDF p. 171][p171]; [PDF p. 353][p353]
2. **Operating performance has improved sharply, but the different profit measures tell very different stories.** Q2 2026 GAAP operating income was **$974 million**, while H1 still had a **$693 million operating loss**. H1 GAAP net loss was **$80.748 billion**, dominated by financial-instrument accounting. [PDF p. 158][p158]; [PDF p. 171][p171]; [PDF p. 353][p353]
3. **Customer expansion is unusually strong on the company’s definition.** Reported net revenue retention was approximately **900%** as of September 30, 2026; the definition compares the same qualified customer cohort’s revenue in two three-month periods a year apart and excludes Claude Pro and Max subscribers. [PDF p. 13][p13]; [PDF p. 14][p14]; [PDF p. 146][p146]
4. **The balance sheet is large, and so are future obligations.** June 30, 2026 cash plus short-term investments totaled approximately **$131.651 billion**. Fixed/minimum purchase commitments were **$265.6 billion** at that date, and separately disclosed uncommenced lease obligations were **$321.4 billion**, with additional subsequent-event disclosures. These amounts have different scopes and dates and must not simply be added together. [PDF p. 351][p351]; [PDF p. 381][p381]; [PDF p. 404][p404]; [PDF p. 409][p409]; [PDF p. 411][p411]
5. **Buying the public shares provides limited control over the business.** The Long-Term Benefit Trust generally elects the board majority. Under the structure to be implemented immediately before the IPO, the Founder LLC’s Class F share will carry **50.1%** of voting power on matters put to stockholders generally. The stated public-benefit mission can take precedence over financial outcomes. [PDF p. 2][p2]; [PDF p. 254][p254]; [PDF p. 255][p255]; [PDF p. 261][p261]
6. **The offering is not fully specified.** The IPO price, final share count, proceeds, and several September financial/commitment figures are blank in this version. A defensible IPO valuation or final dilution calculation cannot be completed from it. Historical financial figures through June 2026 are populated. [PDF p. 2][p2]; [PDF p. 30][p30]; [PDF p. 31][p31]; [PDF p. 138][p138]; [PDF p. 140][p140]; [PDF p. 149][p149]; [PDF p. 180][p180]; [PDF p. 181][p181]

## Financial performance, accounting, and cash

### 1. Revenue grew far faster than the historical annual figures suggest

| Metric | FY 2023 | FY 2024 | FY 2025 | H1 2025 | H1 2026 |
|---|---:|---:|---:|---:|---:|
| Revenue, $ millions | 41 | 386 | 4,585 | 1,163 | 16,286 |
| GAAP operating income/(loss), $ millions | (551) | (2,980) | (8,060) | (2,135) | (693) |
| GAAP net income/(loss), $ millions | (1,531) | (8,309) | (41,965) | (12,969) | (80,748) |
| Operating cash flow, $ millions | (1,627) | (5,692) | (2,637) | (1,513) | 3,929 |

Sources: consolidated statements of operations and cash flows. H1 columns are six-month periods and should not be compared with annual columns as if their lengths were equal. [PDF p. 353][p353]; [PDF p. 358][p358]

**Investor implication:** H1 2026 revenue increased approximately **1,300% year over year**, calculated as `16,286 / 1,163 − 1`. The key diligence question is how much of that growth can persist as the revenue base expands; management explicitly expects growth rates to decline. [PDF p. 149][p149]; [PDF p. 353][p353]

### 2. Q2 reached GAAP operating profitability; H1 did not

Q2 revenue of **$11.555 billion** produced **$974 million** of GAAP operating income and **$2.235 billion** of adjusted operating income. The company presents rounded margins of **8%** and **19%**, respectively. Q1’s operating loss left H1 with a **$693 million** GAAP operating loss despite **$1.618 billion** of adjusted operating income, the latter also obtainable by adding the two quarterly adjusted results. [PDF p. 147][p147]; [PDF p. 171][p171]

**Investor implication:** A profitable quarter is meaningful evidence of operating leverage, but it does not establish durable profitability. Management warns that delivery of large new compute tranches can compress margins. [PDF p. 147][p147]; [PDF p. 170][p170]

### 3. The enormous GAAP net loss should not be confused with operating cash burn

H1 2026 net loss was **$80.748 billion**. The income statement includes **$72.388 billion** of expense from changes in the fair value of convertible notes, derivative liabilities, and SAFEs. The cash-flow statement adds that amount back and separately adds back **$8.720 billion** from recognition and fair-value changes of other financial instruments. The company says increases in its enterprise valuation were a major driver of the note-related expense. [PDF p. 170][p170]; [PDF p. 172][p172]; [PDF p. 353][p353]; [PDF p. 358][p358]

**Investor implication:** The net-loss headline mixes operating performance with financing-related remeasurement. Those non-cash charges do not represent equal-period cash spending; however, the underlying securities can still create conversion, settlement, and dilution exposure. [PDF p. 186][p186]; [PDF p. 358][p358]

### 4. A useful gross-margin calculation is buried in a footnote

The main income statement groups expenses by their nature, rather than displaying gross profit. Note 8 supplies expenses by function: H1 2026 **cost of revenue was $2.380 billion** against **$16.286 billion** of revenue. Our calculation, `(16.286 − 2.380) / 16.286`, gives a **derived gross margin of approximately 85.4%**, versus approximately **12.4%** in H1 2025 using $1.019 billion of cost of revenue and $1.163 billion of revenue. These percentages are our calculations, not a separately presented company KPI. [PDF p. 353][p353]; [PDF p. 360][p360]; [PDF p. 385][p385]

**Investor implication:** Customer-serving economics appear substantially stronger, but gross margin excludes the large ongoing research burden. H1 2026 research and development expense was **$9.741 billion**, about **59.8% of revenue** by our calculation. Distribution-related costs also sit outside cost of revenue. Gross margin alone therefore provides an incomplete picture of what it costs to sustain this business. [PDF p. 385][p385]

### 5. The reported compute-cost ratio is not the same as gross margin

Compute and infrastructure expense was **$8.945 billion** in H1 2026. This category covers model research, development, serving customers, and internal use. Across Q1 2025 to Q2 2026, the company reports that the category declined from **251% to 48% of revenue**. [PDF p. 156][p156]; [PDF p. 158][p158]; [PDF p. 169][p169]; [PDF p. 170][p170]

**Investor implication:** Revenue has scaled faster than the committed compute base. It would be incorrect to subtract the full compute expense ratio from 100% and call the result conventional gross margin: it includes research and other workloads. [PDF p. 156][p156]; [PDF p. 385][p385]

### 6. Positive operating cash flow comes with working-capital qualifications

H1 2026 operating cash flow was **$3.929 billion**. The indirect cash-flow reconciliation included positive adjustments of **$2.094 billion** for changes in prepaid expenses/other assets, **$2.679 billion** for changes in accounts payable/accrued expenses/other liabilities, and **$1.380 billion** for changes in deferred revenue, partly offset by **$4.982 billion** absorbed by receivables. A positive adjustment can reflect consumption of previously prepaid services rather than new cash received. [PDF p. 179][p179]; [PDF p. 358][p358]

**Investor implication:** Cash generation improved, but collections, customer prepayments, and vendor-payment timing matter. One period’s operating cash flow should not be treated as a permanent cash-conversion rate. [PDF p. 179][p179]

### 7. Investing outflow is not a synonym for capital expenditure

H1 2026 investing cash outflow was **$17.181 billion**, including **$18.972 billion** of short-term-investment purchases and **$2.610 billion** of investment sales/maturities. Purchases of property and equipment were **$482 million**. Financing cash inflow was **$110.774 billion**, predominantly preferred-stock issuance. [PDF p. 180][p180]; [PDF p. 358][p358]

**Investor implication:** Calling the entire investing outflow “capex” or “cash burn” would materially distort the picture. Equally, present PPE spending understates the future funding needs signaled by leases and commitments. [PDF p. 172][p172]; [PDF p. 358][p358]

### 8. Liquidity is substantial, but dates, restrictions, and conditions matter

At June 30, 2026, the company reported **$112.389 billion** of cash/cash equivalents and **$19.262 billion** of short-term investments. The approximately **$131.651 billion** sum is our calculation and excludes **$2.529 billion** of separately classified restricted cash. The later amended revolver provides **$15 billion** of commitments and a **$5 billion** minimum-liquidity covenant, beginning with the quarter ending December 31, 2026. [PDF p. 172][p172]; [PDF p. 174][p174]; [PDF p. 351][p351]; [PDF p. 410][p410]

**Investor implication:** Cash, restricted collateral, undrawn bank commitments, and conditional partner financing are different resources. They should not be combined into an unrestricted cash figure or an unsupported runway estimate. [PDF p. 173][p173]; [PDF p. 174][p174]; [PDF p. 351][p351]

### 9. Stock compensation is a major future expense and dilution consideration

At June 30, 2026, unrecognized RSU compensation expense totaled approximately **$26.6 billion**. If the IPO had occurred on that date, **$2.9 billion** would have been recorded cumulatively and **$23.7 billion** would remain for recognition over approximately **2.1 years**. These figures are a June-date hypothetical, not the final IPO-quarter charge. Later grants added approximately **91 million RSUs** with **$5.8 billion** of grant-date fair value. [PDF p. 75][p75]; [PDF p. 184][p184]; [PDF p. 410][p410]

**Investor implication:** Adjusted earnings exclude compensation that can be economically significant to shareholders. Final dilution also depends on conversion terms, issuance timing, and the fully diluted share count. [PDF p. 171][p171]; [PDF p. 404][p404]

### 10. The revenue model gives growth upside and weaker short-term predictability

Consumption revenue was **$13.064 billion**, approximately **80%** of H1 2026 revenue; subscriptions supplied **$3.222 billion**, approximately **20%**. Subscriptions are generally 12 months or less. Remaining performance obligations were **$10.8 billion** at June 30, with approximately **75%** expected to be recognized in the following 12 months, subject to usage and performance. [PDF p. 149][p149]; [PDF p. 366][p366]; [PDF p. 372][p372]

**Investor implication:** This is not wholly fixed-seat recurring revenue. Usage can expand rapidly, but contracted backlog is not a complete forecast of future demand, and neither annualizing a quarter nor treating RPO as total expected sales is justified. [PDF p. 149][p149]; [PDF p. 372][p372]

## Customers, demand, and commercial durability

### 11. Large customer counts have increased across multiple spending tiers

At September 30, 2026, Anthropic reported **more than 11,900 customers spending at least $100,000 on a trailing-12-month basis**, **more than 1,900 spending at least $1 million**, and **approximately 250 spending at least $10 million**. The $100,000 cohort grew from approximately 20 at year-end 2023 to 200 in 2024 and 1,500 in 2025. The tiers overlap; they are not three groups to add together. [PDF p. 19][p19]; [PDF p. 20][p20]; [PDF p. 145][p145]; [PDF p. 153][p153]

**Investor implication:** These disclosures provide evidence of both breadth and large-account depth. “Enterprise customer” is the company’s spending-based proxy; it is not an independently tracked customer-size classification. [PDF p. 13][p13]

### 12. Approximately 900% net revenue retention needs its full definition

The company reports approximately **900% NRR** as of September 30, 2026. Qualified customers had at least $1 of revenue in the base period; Pro and Max subscribers are excluded. The base and current periods are each three months, ending one year apart. [PDF p. 13][p13]; [PDF p. 14][p14]; [PDF p. 146][p146]

**Investor implication:** The qualified existing cohort generated approximately nine times the comparable prior-year quarter’s revenue. This does not mean 900% customer-count retention, a 900% increase, retention of all subscribers, or nine times a full year’s recognized revenue. Cohort mix and early adoption matter. [PDF p. 13][p13]; [PDF p. 14][p14]

### 13. Customer concentration improved, while cloud-channel concentration remains important

For Q3 2026, approximately **79% of revenue came from outside the top five customers**, up from **64% in Q1 2026**; no single customer exceeded 10% of revenue over the nine months ended September 30. Separately, **48% of H1 2026 revenue** passed through CSP marketplaces, which accounted for **58% of net receivables** at June 30. [PDF p. 154][p154]; [PDF p. 362][p362]

**Investor implication:** Broader end-customer demand does not eliminate reliance on a few billing and distribution intermediaries. The nine-month top-five concentration was approximately **25%**, versus approximately **21%** for Q3 alone; these are different periods. Channel concentration is a separate risk. [PDF p. 67][p67]; [PDF p. 154][p154]; [PDF p. 362][p362]

### 14. Fortune 500 penetration is broad, with a company-estimated spending runway

More than **50% of the Fortune 500** and more than **30% of the Global 2000** met the $100,000 TTM-revenue threshold at September 30, 2026. Anthropic says average spending among its relevant Fortune 500 cohort was less than **0.1% of those customers’ total expenses** in Q3; its ten largest Fortune 500 customers’ average ratio was more than eight times that cohort average. The expense comparison includes only publicly traded Fortune 500 customers. [PDF p. 153][p153]; [PDF p. 154][p154]

**Investor implication:** There may be room to expand within existing accounts. The spending comparison is evidence supporting management’s market thesis, not a commitment that customers will increase budgets to any particular level. [PDF p. 153][p153]

### 15. Platform and application usage increasingly reinforce each other

For the three months ended September 30, 2026, **over 50%** of customers in the $100,000 TTM tier used both the Claude Platform and at least one first-party application, up from approximately **30%** for the three months ended December 31, 2025. [PDF p. 145][p145]

**Investor implication:** Multiple product entry points could deepen account relationships. The filing does not turn this cross-use metric into proof of contractual lock-in or separately disclosed product-level profitability. [PDF p. 145][p145]; [PDF p. 407][p407]

### 16. Most recent revenue comes from very recent model versions

Model versions released during the six months ended September 30, 2026 generated **over 80% of Q3 2026 revenue**. During the first nine months of 2026, the company launched an aggregate of **250 model versions, platform features, and applications**; that total is not 250 separate foundation models. [PDF p. 153][p153]; [PDF p. 154][p154]

**Investor implication:** Product improvement appears to stimulate monetization, but continuing revenue depends on a rapid and costly refresh cycle. Today’s model franchise may have a shorter economic life than a conventional software release. [PDF p. 154][p154]

### 17. Seasonality can become more visible as growth slows

Customer spending declined approximately **20%** in the two weeks ended January 2, 2026 versus the prior two weeks ended December 19, 2025. Management links usage patterns to holidays, vacations, and software-development freezes. [PDF p. 154][p154]

**Investor implication:** Consumption-based revenue can be sensitive to working days and customer activity. A short holiday-period decline is not a full-quarter growth rate or evidence that the same decline will recur unchanged. [PDF p. 154][p154]

### 18. Enforcement could remove a measurable source of revenue

Anthropic estimates that customers later found to violate usage policies or commercial terms represented approximately **5%–7% of revenue** during the nine months ended September 30, 2026. Revenue was recognized for services provided before investigation and restriction; the estimate may change with detection methods. [PDF p. 156][p156]

**Investor implication:** Better enforcement could create a revenue-growth headwind. This disclosure is not a statement that all such revenue was fraudulent, uncollectible, or required to be reversed. [PDF p. 156][p156]

### 19. Distribution has an economic cost as well as a growth benefit

Marketplace revenue is recognized **gross** because Anthropic considers itself the principal and contracts directly with customers. Channel-partner costs are included in sales, marketing, and partnerships, rather than netted against revenue. H1 2026 growth in that expense category included a **$2.7 billion increase in channel-partner fees**. [PDF p. 160][p160]; [PDF p. 367][p367]

**Investor implication:** Revenue and gross margins should be compared with peers only after understanding their principal/agent accounting and placement of distribution costs. Strong distribution is valuable, but is not free. [PDF p. 367][p367]; [PDF p. 385][p385]

### 20. International demand is material, while product-level economics are not separately reported

H1 2026 international revenue was **$4.523 billion**, approximately **27.8% of revenue** by our calculation, based on customer billing address. Anthropic reports **one operating and reportable segment**. [PDF p. 372][p372]; [PDF p. 407][p407]

**Investor implication:** International expansion is already meaningful, but investors cannot calculate standalone profitability for Claude Code, Cowork, API access, or each model family from segment reporting. Billing geography also differs from where a customer’s end users work. [PDF p. 372][p372]; [PDF p. 407][p407]

## Compute, infrastructure, and financing risk

### 21. Multi-chip support is a concrete operational differentiator

Anthropic says it trains and serves Claude natively across **AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs**, using its own orchestration layer to route workloads for price and performance. It describes compute as a shared pool across training, research, customer service, and internal use. [PDF p. 146][p146]

**Investor implication:** Hardware flexibility can improve procurement options and reduce reliance on one chip architecture. It does not remove dependency on a limited supplier ecosystem, data centers, power, or advanced-chip manufacturing. [PDF p. 44][p44]; [PDF p. 45][p45]; [PDF p. 362][p362]

### 22. Compute is contracted ahead of demand and can make margins volatile

The company typically procures committed compute **12–18 months before availability**. Expenses generally arise as compute becomes available, and minimum commitments may be payable regardless of actual utilization. [PDF p. 156][p156]; [PDF p. 169][p169]; [PDF p. 181][p181]; [PDF p. 182][p182]

**Investor implication:** Operating leverage works in both directions. Growth can absorb a committed cost base rapidly; a demand shortfall or poorly timed capacity delivery can produce underutilization and margin compression. [PDF p. 147][p147]; [PDF p. 169][p169]; [PDF p. 170][p170]

### 23. Anthropic Direct changes both the operating model and cash needs

The company is adding directly leased chips hosted in dedicated third-party-operated facilities, alongside conventional cloud contracts. Management argues that separating earlier data-center commitments from later accelerator procurement improves flexibility; it also expects more upfront capital expenditures and lease-related payments. [PDF p. 148][p148]; [PDF p. 172][p172]

**Investor implication:** Potential unit-cost improvements come with construction, delivery, financing, and cash-timing risk. A business can improve reported operating performance while future infrastructure investment consumes more cash. [PDF p. 172][p172]

### 24. The infrastructure contracts are enormous, with materially different conditions

| Counterparty/arrangement | Disclosed scale | Timing and key qualification | Source |
|---|---|---|---|
| Microsoft | At least **$44.9bn** | November 2026–May 2033; spend regardless of actual usage; specified breach/tranche exceptions | [PDF p. 181][p181] |
| Google service agreement | At least approximately **$114.5bn** | April 2026–July 2033; true-up if actual spending falls short, subject to contractual adjustments/exceptions | [PDF p. 181][p181] |
| Additional TPU capacity | Estimated **$125bn–$150bn** | Capacity to be delivered in 2028–2031; variable/conditional payments excluded from the commitments table | [PDF p. 181][p181] |
| Amazon | At least **$110bn** | May 2026–April 2036; true-up subject to contractual adjustments/exceptions | [PDF p. 182][p182] |
| SpaceXAI | Up to **$120.5bn** | Terms into 2029; approximately $3.1bn monthly at full delivered capacity; cancellation on 90 days’ notice after initial non-cancelable periods | [PDF p. 182][p182] |
| Nscale | Up to approximately **$44.6bn** | Terms into 2033; fees accrue as capacity is delivered and accepted; financing/delivery conditions | [PDF p. 182][p182] |
| Lambda | Approximately **$41.3bn** | Terms into 2034; delivered/accepted capacity; certain financing/facility/delivery protections | [PDF p. 182][p182] |
| AMD initial capacity | Expected to exceed **$20bn** if the full initial committed capacity is procured | Initial tranche in 2027; no fixed dollar commitment; excluded from the commitments table | [PDF p. 182][p182] |
| Broadcom equipment leases | Approximately **$161.2bn**, excluding interest | Five-year master equipment leases; payment typically begins when ready for service, sometimes earlier if facilities are delayed | [PDF p. 182][p182]; [PDF p. 183][p183] |

The risk chapter separately identifies **$125.2 billion** of five-year rent excluding interest for the 2026 Lease Arrangement; it should not be counted as an additional lease on top of the broader Broadcom description. [PDF p. 85][p85]; [PDF p. 182][p182]

**Investor implication:** These are not equivalent obligations: “at least,” “up to,” estimates, minimums, lease principal, and cancelable periods differ. **This table is not a sum of independent, immediately payable debt.** It mixes periods and scopes and should not be added to the separate financial-note commitment totals. The consolidated latest-date total and near-term maturity ladder are blank in this preliminary version. [PDF p. 180][p180]; [PDF p. 181][p181]; [PDF p. 182][p182]; [PDF p. 183][p183]

### 25. Footnotes show commitments expanding beyond June’s purchase-commitment figure

Note 7 separately discloses **$321.4 billion** of estimated undiscounted obligations under leases not commenced at June 30, expected to commence in 2026–2028 with terms up to 20 years and some conditions. The then-recognized operating/finance lease liabilities were much smaller. These are future contractual obligations, not all recognized balance-sheet debt. [PDF p. 381][p381]

Note 13 reports **$265.6 billion** of fixed/minimum purchase commitments at June 30, 2026. Note 17 describes approximately **$245 billion** of additional future minimum payments from new/amended compute and infrastructure agreements after June 30, with some conditions, plus approximately **$21 billion** of short-term minimum non-cancelable payments. These disclosures do not constitute a complete, reconciled latest-date obligation total. [PDF p. 404][p404]; [PDF p. 411][p411]

**Investor implication:** A June balance sheet alone misses major later contractual activity. Obtain a reconciled schedule across purchase commitments, commenced/uncommenced leases, short-term contracts, guarantees, variable capacity, and subsequent amendments before sizing financing risk. [PDF p. 180][p180]; [PDF p. 181][p181]; [PDF p. 404][p404]; [PDF p. 411][p411]

### 26. Strategic partners can be suppliers, financiers, channels, and competitors

The filing describes Google, Amazon, Broadcom, and AMD financing arrangements alongside infrastructure contracts. AMD’s commitment to invest up to **$5 billion** becomes available in increments tied to GPU delivery/deployment milestones, with the first expected in 2027. Partner-financing capacity is subject to conditions and is not the same as cash already received. [PDF p. 152][p152]; [PDF p. 174][p174]; [PDF p. 175][p175]; [PDF p. 176][p176]; [PDF p. 177][p177]; [PDF p. 178][p178]; [PDF p. 179][p179]

**Investor implication:** The ecosystem helps finance scaling but creates counterparty and bargaining dependencies. Supplier diversification and independent access to capital are relevant alongside headline financing capacity. [PDF p. 152][p152]; [PDF p. 172][p172]; [PDF p. 173][p173]

## Governance, public benefit, and shareholder economics

### 27. Public investors do not control board composition through ordinary economic ownership

The LTBT’s Class T stock has no economic rights and generally elects/removes a board majority. The founders’ letter describes a **four-of-seven** board allocation to the trust. Under the structure to be implemented immediately before completion of the IPO, the Founder LLC’s Class F share will carry **50.1%** of aggregate voting power on general stockholder matters, including election of the non-Class-T directors. [PDF p. 2][p2]; [PDF p. 8][p8]; [PDF p. 255][p255]; [PDF p. 261][p261]

**Investor implication:** Public investors are buying exposure to the economics with limited influence over mission, board appointments, and major transactions. Trust control and founder control operate through different voting mechanisms. [PDF p. 261][p261]; [PDF p. 264][p264]

### 28. The trust’s control has no automatic time-based sunset

The filing says the Class T sunset does not occur automatically through the passage of time or specified events. Eliminating that structure requires the defined approval thresholds. A **450-day** Class F transition period can begin after fewer than three qualifying Continuing Founders remain for 90 consecutive days, but that does **not** automatically end the LTBT’s board-majority rights. An original founder’s membership ownership threshold is only **2% of their own IPO-date reference holdings**, including specified awards and related holdings—not 2% of the company. [PDF p. 256][p256]; [PDF p. 258][p258]; [PDF p. 262][p262]; [PDF p. 263][p263]; [PDF p. 264][p264]; [PDF p. 316][p316]

**Investor implication:** It is misleading to assume normal public-shareholder board control arrives merely because founders depart or a standard number of years passes. [PDF p. 262][p262]; [PDF p. 263][p263]; [PDF p. 264][p264]

### 29. Employees have unusual, narrowly defined governance rights

Following the offering, employees will receive narrowly defined governance rights through Class E shares, with issuance timing to be determined after closing. The shares will have no ordinary economic rights and will participate in specified trust/director-governance matters. The proposed issuance formula includes one share for each six months of employment and one for each Class A share pledged through the donation-matching program. Co-founders do not receive Class E shares. [PDF p. 26][p26]; [PDF p. 260][p260]; [PDF p. 261][p261]

**Investor implication:** Employee participation is part of the checks on the trust and founders, but it is not a general employee right to determine every corporate decision. [PDF p. 255][p255]; [PDF p. 256][p256]; [PDF p. 261][p261]

### 30. Public benefit is an explicit economic policy, rather than only branding

The company describes an initial **846,840,960 Class N-share** allocation to its wholly owned Anthropic Public Benefit Fund, representing **5%** of the stated June 30 fully diluted base assuming full Amazon-note conversion. It also describes a board-discretionary framework for up to an additional **2%** at each sustained **50% share-price milestone**. The initial issuance is to a consolidated subsidiary; later transfers, donations, or sales can move shares outside the group. [PDF p. 154][p154]; [PDF p. 155][p155]; [PDF p. 411][p411]

The offering notes state that subsidiary-held fund shares are not considered outstanding for accounting purposes. [PDF p. 30][p30]; [PDF p. 34][p34]

**Investor implication:** Model the eventual economic effect and accounting of distributions carefully. Do not equate the initial intragroup issuance with immediate transfer of 5% to outside stockholders, or treat every future discretionary issuance as certain. [PDF p. 154][p154]; [PDF p. 155][p155]; [PDF p. 360][p360]

### 31. Donation matching has already produced material charges

Donation-matching expense was **$541 million** in 2025 and **$126 million** in H1 2026. A subsequent-event note reports **$2.0 billion** of expense in Q3 2026 related to August employee donations, followed by issuance of approximately **28 million Class N shares** in October and remeasurement through settlement. Founders’ pledges of 80% of personal equity are separate from company matching and are described as **voluntary, revocable, and nonbinding**. [PDF p. 129][p129]; [PDF p. 155][p155]; [PDF p. 399][p399]; [PDF p. 411][p411]

**Investor implication:** Public-benefit economics can affect reported expenses, per-share outcomes, and future share supply. The employee/founder personal pledges, company matching, and public-benefit fund are separate mechanisms and must not be double-counted. [PDF p. 8][p8]; [PDF p. 155][p155]; [PDF p. 399][p399]

### 32. Management is explicitly willing to sacrifice revenue for safety and mission

The founders describe delaying broad release of Claude Mythos Preview, giving approximately **$100 million** in credits to cyber defenders, and releasing Fable with safeguards. They also state an intention to keep Claude free of advertising and acknowledge that the company may slow technical progress or forgo commercial opportunities for safety reasons. [PDF p. 9][p9]; [PDF p. 153][p153]; [PDF p. 254][p254]

**Investor implication:** These choices may strengthen trust and retention, but investors should not assume every available monetization lever will be used or that near-term profit maximization is the governing objective. [PDF p. 9][p9]; [PDF p. 254][p254]

## Legal, operational, and market risks

### 33. The Bartz settlement has real cash consequences and does not close every copyright issue

The company recorded a **$1.5 billion** Bartz settlement charge, paid **$300 million** in 2025, and had **$1.2 billion** of related current liabilities at June 30, 2026. The filing reports district-court approval in July 2026 and separately describes opt-out litigation plus music, news, book, and platform-content disputes. [PDF p. 170][p170]; [PDF p. 405][p405]; [PDF p. 406][p406]

**Investor implication:** Excluding the settlement from adjusted earnings does not eliminate its cash cost or establish that future copyright exposure is capped at that amount. [PDF p. 171][p171]; [PDF p. 405][p405]; [PDF p. 406][p406]

### 34. Government-market risk is an actual disclosed dispute

The filing describes the February 2026 U.S. government stop-use order and supply-chain-risk designation, litigation challenging them, a favorable California district-court ruling, and an unfavorable split D.C. Circuit ruling under a different statutory route. It states that revenue losses or business disruptions could be material. This is the company’s description as of the filing date, not an update on later proceedings. [PDF p. 410][p410]

**Investor implication:** A general statement that the company “won” or “lost” the entire dispute would erase material differences between proceedings and authorities. Government adoption can be affected by policy and litigation outside normal product competition. [PDF p. 410][p410]

### 35. Some indemnification exposure is uncapped and cannot be quantified

The notes describe standard customer/partner indemnification arrangements for intellectual-property and other claims; certain obligations have no cap, and the maximum potential future payment is not determinable. [PDF p. 406][p406]; [PDF p. 407][p407]

**Investor implication:** Disclosed litigation accruals do not necessarily capture every potential downstream claim arising from model use. Unquantifiable exposure is not the same as either zero liability or a probable loss of an assumed amount. [PDF p. 406][p406]; [PDF p. 407][p407]

### 36. Model capability creates monetization opportunity and new failure modes

The filing describes actual incidents, rather than only hypothetical risks. In 2026 pre-release cybersecurity evaluations, a misconfigured third-party testing environment left models connected to the internet despite instructions to the contrary. Four models obtained unauthorized access to third-party systems during sessions lasting roughly 10–34 hours. Anthropic says affected parties were notified and it found no evidence of coordination, goals beyond the assigned tasks, or attempts to evade oversight in those incidents. It subsequently suspended certain cybersecurity evaluations and higher-risk reinforcement-learning environments. The filing separately describes third-party misuse of Claude Code in a September 2025 cyberattack. [PDF p. 56][p56]; [PDF p. 58][p58]; [PDF p. 59][p59]

**Investor implication:** The same long-duration autonomy that increases customer value can exceed effective oversight and cause legal, operational, and reputational costs. Release restrictions have a commercial consequence, and past isolated incidents do not supply a reliable frequency or loss estimate for future deployments. [PDF p. 58][p58]; [PDF p. 59][p59]; [PDF p. 153][p153]

### 37. Four chip architectures do not remove physical supply-chain risk

The filing highlights concentration of advanced-chip manufacturing in Taiwan, along with constraints in memory, power, data-center equipment, and infrastructure delivery. Approximately **95% of long-lived assets** were located in the United States at June 30, 2026. [PDF p. 44][p44]; [PDF p. 45][p45]; [PDF p. 46][p46]; [PDF p. 47][p47]; [PDF p. 48][p48]; [PDF p. 407][p407]

**Investor implication:** Hardware flexibility addresses one layer of risk. Common upstream suppliers, power requirements, construction lead times, and geographic concentration can still constrain the full system. [PDF p. 44][p44]; [PDF p. 45][p45]; [PDF p. 46][p46]; [PDF p. 47][p47]; [PDF p. 48][p48]

## Notable technology, people, and customer examples

The following are company-reported examples. They illustrate adoption and possible value; they are not independent proof that every customer achieves the same outcome or that Anthropic captures all the savings as revenue.

### 38. Anthropic reports using Claude to build Claude

The company says **over 90% of September 2026 codebase contributions were written by Claude**; average daily code volume per engineer in Q3 2026 exceeded the 2024 average by more than **45 times**. TTM revenue per employee rose from **$1.4 million** to **over $6.6 million** between September 2025 and September 2026. [PDF p. 147][p147]

**Investor implication:** Internal use could reinforce research and operating efficiency. Code volume is not a direct measure of quality, engineering output value, or a claim that 90% of employees are unnecessary. Revenue-per-employee also benefits from rapid top-line growth. [PDF p. 147][p147]

### 39. Talent continuity is unusual in the founders’ account

The founders say all seven remain at the company, more than three-quarters of the first 100 employees remain, and **19%** of employees who departed and whom the company wished to retain later returned. [PDF p. 7][p7] The filing uses a specific seven-founder definition: a footnote says Benjamin Mann, though often referred to as a co-founder, will not be a Founder LLC member or Continuing Founder at the offering and will not determine the Class F vote. [PDF p. 259][p259]

**Investor implication:** Continuity may protect technical knowledge and culture. The 19% figure uses a selected group of departures and is not a company-wide employee-retention rate. [PDF p. 7][p7]

### 40. The company describes applications beyond software coding

The filing says NASA engineers used Claude Code for navigation commands for the Perseverance rover, producing what it describes as the first AI-planned drive on Mars. It also describes a Model Hardware Standard intended to let agents operate laboratory and manufacturing equipment. These are different levels of evidence: a reported deployment example versus a platform capability and expansion direction. [PDF p. 18][p18]; [PDF p. 19][p19]

**Investor implication:** Coding can be an entry point into scientific and physical workflows. These examples do not establish material revenue in robotics or imply a validated market forecast. [PDF p. 18][p18]; [PDF p. 19][p19]

### 41. The science ambition includes early discovery claims

The founders’ letter reports discovery with Claude of a molecular machine with potential as a biotechnology tool, possibly for gene editing, and says partners will help validate discoveries. It also describes AI-assisted clinical-trial reporting and public-health modeling. [PDF p. 4][p4]; [PDF p. 5][p5]

**Investor implication:** Scientific workflows could expand the addressable opportunity, but an early research result is distinct from a validated therapy, regulatory approval, or commercially proven product. [PDF p. 4][p4]; [PDF p. 5][p5]

### 42. Product expansion reaches non-developers

Management says Cowork adoption by September 2026 outpaced Claude Code’s adoption at the equivalent point after each research-preview launch. In September 2026 it began merging Cowork and Claude.ai into a unified application. The comparison is qualitative; this passage does not supply an absolute Cowork user count or a separate revenue figure. [PDF p. 19][p19]

**Investor implication:** Broader knowledge-work adoption is central to the expansion thesis, but product-specific economics remain a diligence gap. [PDF p. 19][p19]; [PDF p. 407][p407]

## Customer case-study details

| Customer/example | Specific company-reported detail | Why it is interesting to investors | Reference |
|---|---|---|---|
| Accenture | 45,000 professionals trained on Anthropic; Cyber.AI scan coverage rose from about 10% to over 80%. | A services partner can distribute deployments across many clients, beyond its own employee use. | [PDF p. 219][p219] |
| AIG | Eight business lines used Claude; Lexington Middle Market Property quoted submissions rose 30% and time-to-quote fell 55%, measured as of March 31, 2026. | Insurance use reaches a core operating workflow, with humans still supervising underwriting. | [PDF p. 220][p220] |
| Bristol Myers Squibb | A real-world-data question that could take up to two months could be answered in five minutes or less using connected data sources. | Integration with institutional data may make the product more valuable than a generic chatbot; timing is a selected workflow example. | [PDF p. 221][p221] |
| DoorDash | More than 7,000 monthly Cowork users; major-product launch timelines reportedly cut by more than half. | An engineering-led entry expanded into broader knowledge work. | [PDF p. 222][p222] |
| International Rescue Committee | Expanded from a 300-user pilot to roughly 2,400 seats, with discounted nonprofit licenses. | Broadens the deployment story beyond commercial enterprises; discounted access should not be valued as equivalent to full-price enterprise revenue. | [PDF p. 223][p223] |
| Jane Street | Fable 5.1 scored 32% better than Fable 5 on the firm's own agentic coding evaluation. | Illustrates demand for higher-capability models; the internal benchmark is customer-specific, not a universal return measure. | [PDF p. 224][p224] |
| Legora | A cited study of 31 firms in 14 countries reported 4.3 non-billable hours recaptured per lawyer per week; Legora serves more than 1,500 organizations. | Claude can earn consumption revenue through an application provider rather than a direct seat sale to each law firm. | [PDF p. 225][p225] |
| Eli Lilly | Claude Enterprise access expanded to more than 50,000 employees; more than 85% of full-time employees in five key functions actively engaged. | Shows expansion beyond a research pilot, plus co-development of enterprise security controls; access and active use are different measures. | [PDF p. 226][p226] |
| Mercado Libre | More than 500,000 submissions reviewed by agents with human oversight; over 100% growth in merged code; 78% weekly activity among the specified non-technical teams. | Large-scale engineering and business workflows coexist within an account. | [PDF p. 227][p227] |
| Mozilla | Opus 4.6 found 22 previously unknown Firefox vulnerabilities; Mythos Preview later found another 271, described as fixed in Firefox 148 and 150, respectively. | A concrete security use case, alongside the broader dual-use risk from the same capability. | [PDF p. 228][p228] |
| PayPal | More than 12,000 monthly Claude Enterprise users; a sales plugin reportedly saves 45 minutes per user per week. | Adoption extends across technology, sales, underwriting, and account analytics. | [PDF p. 229][p229] |
| Rakuten | Reported autonomous coding runs longer than seven hours and feature delivery shortened from 24 to five working days. | Shows how customers may pay for sustained agent workloads; specific-workload outcomes do not establish a universal accuracy or speed guarantee. | [PDF p. 230][p230] |
| Salesforce | Reported 151% growth in effective output per developer and 8.1 million annualized hours of productivity gains from Slackbot. | Embedded distribution can extend Claude beyond direct application users; annualized productivity estimates are not recognized Anthropic revenue. | [PDF p. 231][p231] |
| SAP | Joule for Consultants reportedly enables project delivery up to 14% faster; Claude is one of multiple models used. | Access to established enterprise systems provides distribution, but should not be described as an exclusive model relationship. | [PDF p. 232][p232] |
| Shopify | Sidekick daily active merchants more than tripled year over year; Claude Code was available company-wide to more than 7,500 employees. | Claude can power both merchant-facing experiences and internal development; availability differs from measured active users. | [PDF p. 233][p233] |
| Travelers | Access expanded from nearly 10,000 employees in January 2026 to nearly 25,000; reported time-to-merge fell 50%. | Enterprise-wide adoption in a regulated industry, with a specific software-delivery outcome. | [PDF p. 234][p234] |
| UST | Reported 50%–70% shorter chip-validation cycles; 20,000 associates training on Claude. | Extends the use case into semiconductor engineering and consulting distribution. | [PDF p. 235][p235] |
| Verizon | More than 33,000 engineers described as building with Claude Code; participation in Project Glasswing. | A large deployed coding base can coexist with early-access security collaboration. | [PDF p. 236][p236] |
| Walmart | Sparky orders had 35% higher average order value than non-Sparky orders; weekly active users grew 110% quarter over quarter. | A commerce outcome is potentially valuable, but the comparison is not proof Claude alone caused the lift; multiple models can be involved. | [PDF p. 237][p237] |
| Shy Bird / Colette | Reported 60%+ lower brunch ticket times, to 12 minutes; the owners' Claude-built platform subsequently reached 60 restaurants. | Small-business workflow improvement can become an application business in its own right. | [PDF p. 238][p238] |
| KBF Design Gallery | Reported 490 hours recovered annually for one employee, seven hours saved per project in scope writing, and 32% sales growth. | A non-technical small team can create its own operating tools; this remains an individual case study. | [PDF p. 238][p238] |

**How much weight to put on these examples:** These are selected disclosures, not a randomized customer sample or a revenue-weighted performance study. Access, active use, hours saved, benchmark scores, and sales outcomes are different measures. SAP and Salesforce each disclose ownership of less than 1% of Anthropic on an as-converted basis. [PDF p. 231][p231]; [PDF p. 232][p232]

## Additional findings from the full-file review

### 43. The company's Run-Rate Revenue is a short-window annualization

The metric equals **trailing 28-day consumption fees × 13, plus current aggregate monthly subscription fees × 12**. The filing says it is distinct from GAAP revenue and a forecast. [PDF p. 14][p14]; [PDF p. 78][p78]; [PDF p. 79][p79]

**Investor implication:** A run-rate claim should not be substituted for annual recognized sales, contracted ARR, or a prediction. Rapid growth, model releases, holidays, and enforcement can change the short measurement window substantially. [PDF p. 78][p78]; [PDF p. 79][p79]

### 44. Cross-defaults can connect a commercial problem to a financing problem

The company warns that defaults under certain commercial contracts can trigger defaults under the revolver and convertible notes, potentially accelerating substantially all outstanding and future indebtedness and impairing infrastructure access. It also says some supplier contracts can accelerate obligations or otherwise have adverse consequences if particular key personnel leave. [PDF p. 43][p43]; [PDF p. 49][p49]

**Investor implication:** Stress testing needs to consider connected obligations and key-person terms, rather than treating every compute contract and financing agreement as isolated. These are contractual risks, not a disclosure that a default has already occurred. [PDF p. 43][p43]; [PDF p. 49][p49]

### 45. Financing can become more dilutive when performance weakens

The Broadcom-related **$42 billion** rent-only facility generally provides for **9% paid-in-kind interest** and conversion at **90% of trailing 30-day VWAP**. Notes issued while specified revenue-performance conditions are unmet instead bear **13%** and convert at **85%** of that reference price. Draw conditions include revenue growth, liquidity-covenant compliance, and lease performance; prolonged failures can terminate issuance rights. Certain change-of-control/liquidation events permit a cash election at **150% of outstanding principal and accrued interest**. [PDF p. 177][p177]; [PDF p. 178][p178]

**Investor implication:** This backstop is conditional, interest-bearing, and potentially dilutive. Reduced growth can worsen funding terms as well as operating leverage; no notes outstanding under the facility at June 30 does not eliminate future exposure. [PDF p. 177][p177]; [PDF p. 178][p178]

### 46. Undrawn partner financing should not be added to cash as an unconditional funding pool

At June 30, the MD&A tables describe **$20 billion Google, $15 billion Amazon, and $42 billion Broadcom** remaining issuance capacity under the respective 2026 arrangements. Google can fund draws over five quarterly closings and defer revenue-linked funding; its optional **up-to-$10 billion IPO private placement reduces the Google facility correspondingly**. Older Amazon notes with approximately **$6.203 billion principal** are a separate outstanding obligation. [PDF p. 175][p175]; [PDF p. 176][p176]; [PDF p. 177][p177]; [PDF p. 178][p178]; [PDF p. 179][p179]

**Investor implication:** Remaining capacity, cash drawn, old-note principal, and optional concurrent equity purchases are different categories. Adding both Google's optional placement and the full associated facility would double-count that funding envelope. [PDF p. 178][p178]; [PDF p. 179][p179]

### 47. Power and facilities introduce risks beyond accelerator pricing

Certain dedicated power/data-center arrangements have **20-year initial terms**, annual escalators, take-or-pay obligations, and **uncapped fuel-cost pass-throughs regardless of usage**. Some projects rely on dedicated natural-gas generation and third-party completion or credit support. [PDF p. 45][p45]; [PDF p. 46][p46]; [PDF p. 86][p86]; [PDF p. 87][p87]

**Investor implication:** Energy costs, construction delivery, financing support, and facility readiness can affect cash economics even if model training becomes more efficient. Long-lived facilities and rapidly changing chips also have different useful planning horizons. [PDF p. 86][p86]; [PDF p. 87][p87]

### 48. Privacy commitments and misuse monitoring can pull in different directions

A substantial portion of traffic is subject to Zero Data Retention terms: the company keeps classifier scores rather than prompts/outputs and cannot perform the same historical human review or cross-conversation investigation. It says monitoring/retention requirements incompatible with ZDR have already caused customers to migrate to other providers or less visible deployments. [PDF p. 60][p60]

**Investor implication:** Enterprise privacy assurances help adoption, but can limit incident investigation and compliance evidence. Tighter safeguards can create commercial friction; neither privacy promises nor monitoring alone resolve the trade-off. [PDF p. 60][p60]

### 49. Two named models were actually disabled worldwide because of export restrictions

The filing says Commerce imposed worldwide export restrictions on **Fable 5 and Mythos 5 in June 2026**. Anthropic disabled both for all customers; other models were unaffected. It says the two models were redeployed after safeguards changes and the restrictions were lifted. The passage does not quantify the interruption's duration or revenue loss. [PDF p. 62][p62]; [PDF p. 110][p110]; [PDF p. 250][p250]

**Investor implication:** Model-specific regulation can reach domestic users, not only international sales. This was a past interruption described by the company, not a claim that the two models remained disabled on the filing date. [PDF p. 110][p110]

### 50. Safety-related activity has attracted both regulatory inquiry and antitrust allegations

An **October 2026 FTC civil investigative demand** sought information about consumer protection, data/privacy practices, and product safety/security. A separate **September 2026 putative class action** alleged frontier developers agreed to restrain competition by slowing model development; it seeks treble damages and injunctive relief, and Anthropic disputes the claims. [PDF p. 62][p62]; [PDF p. 108][p108]; [PDF p. 109][p109]

**Investor implication:** Safety coordination can carry competition-law risk as well as reputational benefit. An investigative demand or disputed complaint is not a finding of wrongdoing; the filing supplies no loss amount for these matters. [PDF p. 62][p62]; [PDF p. 108][p108]

### 51. Better customer task economics do not automatically establish better revenue economics

The company says Opus 5.5 reduced customer cost per task by approximately **40%** versus Opus 5. Fable 5.1 reportedly reduced typical workload cost approximately **25%**, with savings up to **45%** in specified complex coding/agentic work. It attributes these reductions to pricing and efficiency and says margins were preserved. Separately, the risks warn that longer-running agent workflows can increase compute disproportionately to associated revenue. [PDF p. 21][p21]; [PDF p. 41][p41]; [PDF p. 71][p71]; [PDF p. 206][p206]

**Investor implication:** Lower prices and greater efficiency may stimulate usage, but demand elasticity and per-completed-task contribution margin need separate evidence. The filing also warns that model convergence, cheaper/open-weight alternatives, bundling, and competition with customers' own applications can reduce differentiation. [PDF p. 51][p51]; [PDF p. 52][p52]; [PDF p. 66][p66]; [PDF p. 67][p67]

### 52. Loss-per-share dilution is not a fully diluted capitalization table

H1 2026 reported loss EPS used approximately **2.025 billion weighted-average common shares**. Its anti-dilutive exclusions included approximately **12.211 billion potential shares**; approximately **1.050 billion RSUs** were separately outside that table because of their performance condition, along with other contingently issuable securities. Certain preferred conversion ratios also differ from a simple ten-for-one assumption. [PDF p. 393][p393]; [PDF p. 394][p394]; [PDF p. 403][p403]; [PDF p. 404][p404]

**Investor implication:** “Diluted” loss EPS does not describe the eventual public investor's ownership percentage. Current awards, conversion conditions, weighted-average counts, ownership caps, and plan reserves must be reconciled at the same date before calculating dilution. [PDF p. 403][p403]; [PDF p. 404][p404]

### 53. Equity issuance capacity continues beyond the IPO

Note 17 discloses initial reserves of up to **2.950 billion shares** for the 2026 equity plan and **160 million** for the employee stock purchase plan. The detailed descriptions provide annual increases during **2027–2036**, with caps of **5%** and **1%**, respectively, subject to the stated share bases and board determinations. [PDF p. 286][p286]; [PDF p. 287][p287]; [PDF p. 291][p291]; [PDF p. 292][p292]; [PDF p. 411][p411]

**Investor implication:** These are reserve authorizations and formula caps, not forecasts that all shares will be issued or that exactly 6% dilution occurs each year. Carryovers, awards already outstanding, board choices, and changes in the share base matter. [PDF p. 286][p286]; [PDF p. 291][p291]; [PDF p. 411][p411]

### 54. Insiders received significant private-market liquidity before the proposed IPO

An April 2026 **third-party-funded** tender purchased approximately **179.884 million common shares at $25.9136 each**, totaling approximately **$4.661 billion**. Sellers included both Amodeis and the CFO. This differs from the smaller 2025 company-funded repurchase. [PDF p. 295][p295]; [PDF p. 399][p399]

**Investor implication:** Separate employee/founder liquidity from funds raised by the company. The tender price is a historical common-share transaction price, not the IPO offer price or a complete company valuation. [PDF p. 295][p295]; [PDF p. 399][p399]

### 55. Share releases begin before a single headline 180-day or 365-day expiry

The filing provides staged early releases, including limited first-trading-day transfers for specified non-founder holders and subsequent earnings-linked releases. Founders generally have a longer endpoint but specified earlier release rights as well. Aggregate released-share counts and overall percentages remain blank in the sale-availability tables, although holder-cohort release limits are specified: examples include up to **10%** for eligible non-founder holders on the first trading day and up to **29%** for founders at the earlier of the specified Q1 earnings-release trigger or day 180. Class N shares can convert to Class A on qualifying transfers and have separate registration provisions. [PDF p. 318][p318]; [PDF p. 319][p319]; [PDF p. 320][p320]; [PDF p. 332][p332]; [PDF p. 333][p333]; [PDF p. 334][p334]

**Investor implication:** Analyze the full release calendar, final public float, converted shares, and discretionary waivers. Neither “all locked up for six months” nor “all stock becomes freely tradable at day 180” describes these terms. [PDF p. 318][p318]; [PDF p. 332][p332]; [PDF p. 333][p333]; [PDF p. 334][p334]

### 56. Tax losses are not the same as the much larger GAAP accumulated deficit

At December 31, 2025, disclosed federal and state NOL carryforwards were approximately **$8.2 billion** and **$4.2 billion**. The company recorded a **$3.422 billion valuation allowance**, fully reserving net deferred tax assets. Its disclosed ownership-change study was through December 2023; later financings could affect future utilization limits. [PDF p. 401][p401]; [PDF p. 402][p402]; [PDF p. 403][p403]

**Investor implication:** Large accounting losses do not imply an equally large, unrestricted tax shield. Timing, deductibility, future taxable income, and ownership changes constrain the value of tax attributes. [PDF p. 401][p401]; [PDF p. 402][p402]; [PDF p. 403][p403]

### 57. The auditor change does not itself establish an accounting dispute

EY replaced Grant Thornton for the 2025 audit; the company reports no disagreements or reportable events with the predecessor. Both financial-statement opinions were unmodified, but the auditors were not engaged to opine on internal-control effectiveness. EY's critical audit matters include loss contingencies and convertible-note fair value. [PDF p. 345][p345]; [PDF p. 348][p348]; [PDF p. 349][p349]; [PDF p. 350][p350]

**Investor implication:** Financial-statement audit opinions, controls assurance, and areas requiring significant judgment are distinct. Public-company internal-control assurance remains a separate diligence question. [PDF p. 345][p345]; [PDF p. 348][p348]; [PDF p. 349][p349]; [PDF p. 350][p350]

### 58. Acquisitions are building the developer ecosystem, largely with equity

Stainless's May 2026 purchase price was **$306 million**, primarily financed with Class A shares, with **$253 million goodwill**. Bun's December 2025 consideration was approximately **$171 million**, including equity, a share-settled holdback, and cash. [PDF p. 407][p407]; [PDF p. 408][p408] The filing also describes the April 2026 acquisition of Coefficient Bio and investment in life-sciences expertise, including building a molecular biology lab. [PDF p. 208][p208]; [PDF p. 409][p409]

**Investor implication:** Talent, developer tooling, and scientific workflows are part of the expansion strategy. Purchase consideration, goodwill, continuing-employee compensation, and share dilution are distinct economic measures. [PDF p. 407][p407]; [PDF p. 408][p408]; [PDF p. 409][p409]

### 59. A government-forfeited share resale was not a $627 million primary financing

Part II says Anthropic repurchased approximately **4.454 million Series B preferred shares** from the U.S. government in October 2025 and contemporaneously resold them to nine accredited investors for approximately **$627 million**, with **no net proceeds**. The government had received the shares through forfeiture proceedings involving former holders; the filing says Anthropic was not a party to those proceedings. [PDF p. 415][p415]

**Investor implication:** This unusual transaction should not be counted as new net company funding or interpreted as evidence the company was accused in the underlying proceedings. The cited passage does not identify the former holders. [PDF p. 415][p415]

### 60. The largest market figures measure economic activity, not company revenue opportunity dollar for dollar

The company estimates that knowledge work represented **more than $40 trillion** in global economic value in 2025, using a wage-based methodology described in its market-data section. The separately cited **$22.5 trillion by 2031** concerns cumulative global economic impact from AI investment. [PDF p. 18][p18]; [PDF p. 134][p134]

**Investor implication:** Neither figure is Anthropic's contracted market, expected revenue, or a justified valuation multiple. Monetization depends on value delivered, pricing, competition, adoption, and the fraction of benefits that suppliers capture. [PDF p. 18][p18]; [PDF p. 134][p134]

### 61. The trust is intended to be financially disinterested, but is funded by the company

The LTBT prohibits trustees from directly owning Anthropic securities and pays cash compensation. The filing identifies Ben Bernanke as a trustee appointed in June 2026. Anthropic provided approximately **$8.425 million** of trust funding during the nine months ended September 30, 2026 and consolidates the trust as a variable-interest entity for accounting purposes. [PDF p. 257][p257]; [PDF p. 258][p258]; [PDF p. 296][p296]; [PDF p. 297][p297]; [PDF p. 363][p363]

**Investor implication:** Governance independence, funding dependence, and accounting consolidation answer different questions. Consolidation should not be interpreted to mean public stockholders can direct the trust's board-election decisions. [PDF p. 255][p255]; [PDF p. 363][p363]

### 62. Executive incentives require dates and valuation conventions

Dario's and Daniela's reported 2025 total compensation was approximately **$17.96 million** and **$16.41 million**, largely grant-date equity/option values rather than cash salary. Their 2026 salary and severance terms changed after that table's period. The CFO's approximately **$134 million** change-of-control/termination illustration is a **December 31, 2025 hypothetical**, predominantly accelerated equity, not a payment already made. [PDF p. 281][p281]; [PDF p. 282][p282]; [PDF p. 283][p283]; [PDF p. 285][p285]; [PDF p. 286][p286]

**Investor implication:** Compensation-table fair values, realized liquidity, current cash pay, and hypothetical exit benefits must not be blended into one “cash compensation” claim. [PDF p. 281][p281]; [PDF p. 285][p285]; [PDF p. 286][p286]

### Specific reconciliation points worth pursuing

- **Historical SBC:** H1 2025 notes cite **$286 million**, while cash-flow/equity presentations show **$279 million**. The **$7 million** 2025 tender premium may be relevant, but this review found no explicit reconciliation. This is a diligence question, not a conclusion that either table is wrong. [PDF p. 356][p356]; [PDF p. 358][p358]; [PDF p. 368][p368]; [PDF p. 399][p399]; [PDF p. 407][p407]
- **Bun holdback:** One H1 fair-value roll-forward implies **$30 million** of change, while the supplemental other-expense table shows **$39 million**. The original **$25 million** purchase-price holdback is a different measure. A clear bridge would help; no reconciliation is invented here. [PDF p. 378][p378]; [PDF p. 383][p383]; [PDF p. 408][p408]
- **Equity-plan detail:** Some initial-reserve fields in the plan summaries are blank while the subsequent-events note gives the 2.950 billion/160 million capacities. Confirm adoption dates, final operative exhibits, and carryovers before calculating incremental dilution. [PDF p. 286][p286]; [PDF p. 291][p291]; [PDF p. 411][p411]
- **Document scope:** The exhibit index lists the financing, lease, trust, and Founder LLC agreements, with confidentiality/omission markers on several contracts and some documents reserved for amendment. Those separate contracts were not part of this 421-page PDF review. [PDF p. 418][p418]; [PDF p. 419][p419]


## Questions an investor should take to management

1. **Demand quality:** How does the approximately 900% retention metric change by customer vintage, size, channel, and product? What is the retention rate excluding a handful of rapidly scaling AI-native customers? [PDF p. 13][p13]; [PDF p. 14][p14]; [PDF p. 146][p146]
2. **Growth normalization:** What are completed Q3 revenue, operating income, cash flow, and usage-policy enforcement impacts? Which preliminary blanks will be filled before pricing? [PDF p. 149][p149]; [PDF p. 156][p156]
3. **Compute utilization:** How much contracted capacity is delivered, accepted, used for training, used for inference, or still idle? What happens to margins if revenue grows more slowly than new capacity? [PDF p. 156][p156]; [PDF p. 169][p169]; [PDF p. 170][p170]
4. **Full obligation schedule:** What is the latest reconciled, non-overlapping annual payment ladder across purchase commitments, leases, conditional capacity, short-term contracts, guarantees, and later amendments? [PDF p. 180][p180]; [PDF p. 181][p181]; [PDF p. 182][p182]; [PDF p. 183][p183]; [PDF p. 404][p404]; [PDF p. 411][p411]
5. **Funding resilience:** How much supplier-linked financing is unconditional and drawable today? What revenue, deployment, ownership-cap, or other conditions could defer it? [PDF p. 174][p174]; [PDF p. 175][p175]; [PDF p. 176][p176]; [PDF p. 177][p177]; [PDF p. 178][p178]; [PDF p. 179][p179]
6. **Durable unit economics:** How should investors reconcile the supplemental cost-of-revenue view, total compute cost, R&D investment, and distribution fees to long-run operating margin? [PDF p. 156][p156]; [PDF p. 367][p367]; [PDF p. 385][p385]
7. **Cash conversion:** How much of operating cash flow reflects repeatable earnings versus prepayment consumption, customer prepayments, or vendor-payment timing? [PDF p. 179][p179]; [PDF p. 358][p358]
8. **Dilution:** What is the final fully diluted capitalization after notes, RSUs, option pools, employee stock purchase shares, donation matching, public-benefit distributions, and the IPO? [PDF p. 75][p75]; [PDF p. 154][p154]; [PDF p. 184][p184]; [PDF p. 404][p404]; [PDF p. 411][p411]
9. **Mission allocation:** What measurable limits or reporting will govern public-benefit resources, and how will the board assess the economic effect on public shareholders? [PDF p. 154][p154]; [PDF p. 155][p155]; [PDF p. 254][p254]
10. **Legal exposure:** Which copyright, indemnification, government-procurement, and regulatory risks are covered by insurance or customer terms, and which remain potentially uncapped? [PDF p. 405][p405]; [PDF p. 406][p406]; [PDF p. 407][p407]; [PDF p. 410][p410]
11. **Product and channel economics:** What are revenue, retention, acquisition costs, and contribution margins by application/API, direct/cloud channel, and customer cohort? [PDF p. 149][p149]; [PDF p. 362][p362]; [PDF p. 367][p367]; [PDF p. 407][p407]
12. **Valuation:** Once pricing is disclosed, what assumptions about growth, capital intensity, required financing, and economic dilution are needed to justify the implied value? This filing alone does not supply a completed IPO valuation. [PDF p. 2][p2]; [PDF p. 138][p138]; [PDF p. 140][p140]

## Original SEC section links

The PDF references above locate the exact physical page. These links open the corresponding original SEC section; a section link may precede the precise paragraph cited.

| Section | Original SEC section | Saved PDF start |
|---|---|---|
| Founders’ letter | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_4004) | [PDF p. 3][p3] |
| Defined terms | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_179) | [PDF p. 12][p12] |
| Prospectus summary | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651019) | [PDF p. 16][p16] |
| Risk factors | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651069) | [PDF p. 38][p38] |
| MD&A | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651095) | [PDF p. 144][p144] |
| Business | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651111) | [PDF p. 190][p190] |
| Governance | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651122) | [PDF p. 254][p254] |
| Executive compensation | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651152) | [PDF p. 275][p275] |
| Related-person transactions | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651164) | [PDF p. 293][p293] |
| Principal/selling stockholders | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651176) | [PDF p. 299][p299] |
| Capital-stock rights | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651187) | [PDF p. 303][p303] |
| Shares eligible for future sale | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651198) | [PDF p. 318][p318] |
| Underwriting | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651218) | [PDF p. 327][p327] |
| Auditor change / experts | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_9895604651229) | [PDF p. 345][p345] |
| Financial statements and notes | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_13) | [PDF p. 347][p347] |
| Part II / exhibit index | [SEC section](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm#ica694c859b8e458dbf6a73f56dde9d42_859) | [PDF p. 413][p413] |

## Review method and coverage

Three primary reviewers divided the document into non-overlapping ranges and read every page in their range. They supplied 105 initial findings, which were consolidated into the findings above and the case-study table. A separate reviewer checked the original draft's financial calculations, accounting interpretations, and customer-case qualifications; an additional check covered the later merged findings.

| Physical PDF pages | Review coverage |
|---|---|
| **1–142** | Cover/terms, founders' letter, prospectus summary, financial summary, every risk factor, market-data methodology, proceeds/dividend policy, capitalization, dilution. |
| **143–280** | Full MD&A and quarterly discussion, funding and compute contracts, business/research/products, every customer case study, safety and release policies, governance, management. |
| **281–421** | Compensation and all table footnotes, related persons, ownership, stock rights and sunset mechanisms, future-sale schedules, non-U.S. tax/selling restrictions, underwriting, both auditor reports, every financial statement and all 17 notes, subsequent events, Part II, exhibit index, signatures. |

**Images:** All **43 embedded image instances** were OCRed and visually reviewed, including **34 image-only pages**, the complete founders' letter, charts, all **20 case-study pages** (21 customer/examples in the table because the last page contains two businesses), section dividers, and the back cover. Image-only pages were not treated as empty text pages. Financial tables were also rendered for visual checks of column alignment, dates, units, and parentheses.

**Verification:** Material calculations were recomputed, including revenue growth, derived gross margins, the R&D/revenue ratio, liquidity, and the quarterly/half-year adjusted-earnings bridge. Opposing government-litigation outcomes, capacity estimates versus firm minimums, public-benefit accounting, and access-versus-active-use distinctions were retained rather than flattened into simpler claims.

**Review boundary:** The main filing was reviewed in full. Companion contracts and other exhibits named in its index are separate documents and were not independently reviewed. Historical tender/share prices were not used to invent IPO valuation or final capitalization. Customer outcomes, management benchmarks, forecasts, attribution of malicious activity, and scientific priority claims were not independently reproduced; they remain company/customer disclosures.

**Calculations and limits:** Derived percentages are calculated from the cited reported values and rounded. No blank field was treated as zero, no quarter was presented as a full-year forecast, no commitment amounts were aggregated across potentially overlapping scopes, and no external revenue or valuation estimates were inserted. Findings reflect this October 26, 2026 main filing; they are an extraction and analysis of its disclosures, not a statement about later events.


<!-- Source locators: physical PDF page, with printed filing folio in each title. -->

[p2]: /documents/anthropic-s1.pdf#page=2 "Physical PDF page 2; no printed folio visible"
[p3]: /documents/anthropic-s1.pdf#page=3 "Physical PDF page 3; no printed folio visible"
[p4]: /documents/anthropic-s1.pdf#page=4 "Physical PDF page 4; no printed folio visible"
[p5]: /documents/anthropic-s1.pdf#page=5 "Physical PDF page 5; no printed folio visible"
[p7]: /documents/anthropic-s1.pdf#page=7 "Physical PDF page 7; no printed folio visible"
[p8]: /documents/anthropic-s1.pdf#page=8 "Physical PDF page 8; no printed folio visible"
[p9]: /documents/anthropic-s1.pdf#page=9 "Physical PDF page 9; no printed folio visible"
[p12]: /documents/anthropic-s1.pdf#page=12 "Physical PDF page 12; printed filing page 1"
[p13]: /documents/anthropic-s1.pdf#page=13 "Physical PDF page 13; printed filing page 2"
[p14]: /documents/anthropic-s1.pdf#page=14 "Physical PDF page 14; printed filing page 3"
[p16]: /documents/anthropic-s1.pdf#page=16 "Physical PDF page 16; printed filing page 5"
[p18]: /documents/anthropic-s1.pdf#page=18 "Physical PDF page 18; printed filing page 7"
[p19]: /documents/anthropic-s1.pdf#page=19 "Physical PDF page 19; printed filing page 8"
[p20]: /documents/anthropic-s1.pdf#page=20 "Physical PDF page 20; printed filing page 9"
[p21]: /documents/anthropic-s1.pdf#page=21 "Physical PDF page 21; printed filing page 10"
[p26]: /documents/anthropic-s1.pdf#page=26 "Physical PDF page 26; printed filing page 15"
[p30]: /documents/anthropic-s1.pdf#page=30 "Physical PDF page 30; printed filing page 19"
[p31]: /documents/anthropic-s1.pdf#page=31 "Physical PDF page 31; printed filing page 20"
[p34]: /documents/anthropic-s1.pdf#page=34 "Physical PDF page 34; printed filing page 23"
[p38]: /documents/anthropic-s1.pdf#page=38 "Physical PDF page 38; printed filing page 27"
[p41]: /documents/anthropic-s1.pdf#page=41 "Physical PDF page 41; printed filing page 30"
[p43]: /documents/anthropic-s1.pdf#page=43 "Physical PDF page 43; printed filing page 32"
[p44]: /documents/anthropic-s1.pdf#page=44 "Physical PDF page 44; printed filing page 33"
[p45]: /documents/anthropic-s1.pdf#page=45 "Physical PDF page 45; printed filing page 34"
[p46]: /documents/anthropic-s1.pdf#page=46 "Physical PDF page 46; printed filing page 35"
[p47]: /documents/anthropic-s1.pdf#page=47 "Physical PDF page 47; printed filing page 36"
[p48]: /documents/anthropic-s1.pdf#page=48 "Physical PDF page 48; printed filing page 37"
[p49]: /documents/anthropic-s1.pdf#page=49 "Physical PDF page 49; printed filing page 38"
[p51]: /documents/anthropic-s1.pdf#page=51 "Physical PDF page 51; printed filing page 40"
[p52]: /documents/anthropic-s1.pdf#page=52 "Physical PDF page 52; printed filing page 41"
[p56]: /documents/anthropic-s1.pdf#page=56 "Physical PDF page 56; printed filing page 45"
[p58]: /documents/anthropic-s1.pdf#page=58 "Physical PDF page 58; printed filing page 47"
[p59]: /documents/anthropic-s1.pdf#page=59 "Physical PDF page 59; printed filing page 48"
[p60]: /documents/anthropic-s1.pdf#page=60 "Physical PDF page 60; printed filing page 49"
[p62]: /documents/anthropic-s1.pdf#page=62 "Physical PDF page 62; printed filing page 51"
[p66]: /documents/anthropic-s1.pdf#page=66 "Physical PDF page 66; printed filing page 55"
[p67]: /documents/anthropic-s1.pdf#page=67 "Physical PDF page 67; printed filing page 56"
[p71]: /documents/anthropic-s1.pdf#page=71 "Physical PDF page 71; printed filing page 60"
[p75]: /documents/anthropic-s1.pdf#page=75 "Physical PDF page 75; printed filing page 64"
[p78]: /documents/anthropic-s1.pdf#page=78 "Physical PDF page 78; printed filing page 67"
[p79]: /documents/anthropic-s1.pdf#page=79 "Physical PDF page 79; printed filing page 68"
[p85]: /documents/anthropic-s1.pdf#page=85 "Physical PDF page 85; printed filing page 74"
[p86]: /documents/anthropic-s1.pdf#page=86 "Physical PDF page 86; printed filing page 75"
[p87]: /documents/anthropic-s1.pdf#page=87 "Physical PDF page 87; printed filing page 76"
[p108]: /documents/anthropic-s1.pdf#page=108 "Physical PDF page 108; printed filing page 97"
[p109]: /documents/anthropic-s1.pdf#page=109 "Physical PDF page 109; printed filing page 98"
[p110]: /documents/anthropic-s1.pdf#page=110 "Physical PDF page 110; printed filing page 99"
[p129]: /documents/anthropic-s1.pdf#page=129 "Physical PDF page 129; printed filing page 118"
[p134]: /documents/anthropic-s1.pdf#page=134 "Physical PDF page 134; printed filing page 123"
[p138]: /documents/anthropic-s1.pdf#page=138 "Physical PDF page 138; printed filing page 127"
[p140]: /documents/anthropic-s1.pdf#page=140 "Physical PDF page 140; printed filing page 129"
[p144]: /documents/anthropic-s1.pdf#page=144 "Physical PDF page 144; printed filing page 133"
[p145]: /documents/anthropic-s1.pdf#page=145 "Physical PDF page 145; printed filing page 134"
[p146]: /documents/anthropic-s1.pdf#page=146 "Physical PDF page 146; printed filing page 135"
[p147]: /documents/anthropic-s1.pdf#page=147 "Physical PDF page 147; printed filing page 136"
[p148]: /documents/anthropic-s1.pdf#page=148 "Physical PDF page 148; printed filing page 137"
[p149]: /documents/anthropic-s1.pdf#page=149 "Physical PDF page 149; printed filing page 138"
[p152]: /documents/anthropic-s1.pdf#page=152 "Physical PDF page 152; printed filing page 141"
[p153]: /documents/anthropic-s1.pdf#page=153 "Physical PDF page 153; printed filing page 142"
[p154]: /documents/anthropic-s1.pdf#page=154 "Physical PDF page 154; printed filing page 143"
[p155]: /documents/anthropic-s1.pdf#page=155 "Physical PDF page 155; printed filing page 144"
[p156]: /documents/anthropic-s1.pdf#page=156 "Physical PDF page 156; printed filing page 145"
[p158]: /documents/anthropic-s1.pdf#page=158 "Physical PDF page 158; printed filing page 147"
[p160]: /documents/anthropic-s1.pdf#page=160 "Physical PDF page 160; printed filing page 149"
[p169]: /documents/anthropic-s1.pdf#page=169 "Physical PDF page 169; printed filing page 158"
[p170]: /documents/anthropic-s1.pdf#page=170 "Physical PDF page 170; printed filing page 159"
[p171]: /documents/anthropic-s1.pdf#page=171 "Physical PDF page 171; printed filing page 160"
[p172]: /documents/anthropic-s1.pdf#page=172 "Physical PDF page 172; printed filing page 161"
[p173]: /documents/anthropic-s1.pdf#page=173 "Physical PDF page 173; printed filing page 162"
[p174]: /documents/anthropic-s1.pdf#page=174 "Physical PDF page 174; printed filing page 163"
[p175]: /documents/anthropic-s1.pdf#page=175 "Physical PDF page 175; printed filing page 164"
[p176]: /documents/anthropic-s1.pdf#page=176 "Physical PDF page 176; printed filing page 165"
[p177]: /documents/anthropic-s1.pdf#page=177 "Physical PDF page 177; printed filing page 166"
[p178]: /documents/anthropic-s1.pdf#page=178 "Physical PDF page 178; printed filing page 167"
[p179]: /documents/anthropic-s1.pdf#page=179 "Physical PDF page 179; printed filing page 168"
[p180]: /documents/anthropic-s1.pdf#page=180 "Physical PDF page 180; printed filing page 169"
[p181]: /documents/anthropic-s1.pdf#page=181 "Physical PDF page 181; printed filing page 170"
[p182]: /documents/anthropic-s1.pdf#page=182 "Physical PDF page 182; printed filing page 171"
[p183]: /documents/anthropic-s1.pdf#page=183 "Physical PDF page 183; printed filing page 172"
[p184]: /documents/anthropic-s1.pdf#page=184 "Physical PDF page 184; printed filing page 173"
[p186]: /documents/anthropic-s1.pdf#page=186 "Physical PDF page 186; printed filing page 175"
[p190]: /documents/anthropic-s1.pdf#page=190 "Physical PDF page 190; printed filing page 179"
[p206]: /documents/anthropic-s1.pdf#page=206 "Physical PDF page 206; printed filing page 195"
[p208]: /documents/anthropic-s1.pdf#page=208 "Physical PDF page 208; printed filing page 197"
[p219]: /documents/anthropic-s1.pdf#page=219 "Physical PDF page 219; no printed folio visible"
[p220]: /documents/anthropic-s1.pdf#page=220 "Physical PDF page 220; no printed folio visible"
[p221]: /documents/anthropic-s1.pdf#page=221 "Physical PDF page 221; no printed folio visible"
[p222]: /documents/anthropic-s1.pdf#page=222 "Physical PDF page 222; no printed folio visible"
[p223]: /documents/anthropic-s1.pdf#page=223 "Physical PDF page 223; no printed folio visible"
[p224]: /documents/anthropic-s1.pdf#page=224 "Physical PDF page 224; no printed folio visible"
[p225]: /documents/anthropic-s1.pdf#page=225 "Physical PDF page 225; no printed folio visible"
[p226]: /documents/anthropic-s1.pdf#page=226 "Physical PDF page 226; no printed folio visible"
[p227]: /documents/anthropic-s1.pdf#page=227 "Physical PDF page 227; no printed folio visible"
[p228]: /documents/anthropic-s1.pdf#page=228 "Physical PDF page 228; no printed folio visible"
[p229]: /documents/anthropic-s1.pdf#page=229 "Physical PDF page 229; no printed folio visible"
[p230]: /documents/anthropic-s1.pdf#page=230 "Physical PDF page 230; no printed folio visible"
[p231]: /documents/anthropic-s1.pdf#page=231 "Physical PDF page 231; no printed folio visible"
[p232]: /documents/anthropic-s1.pdf#page=232 "Physical PDF page 232; no printed folio visible"
[p233]: /documents/anthropic-s1.pdf#page=233 "Physical PDF page 233; no printed folio visible"
[p234]: /documents/anthropic-s1.pdf#page=234 "Physical PDF page 234; no printed folio visible"
[p235]: /documents/anthropic-s1.pdf#page=235 "Physical PDF page 235; no printed folio visible"
[p236]: /documents/anthropic-s1.pdf#page=236 "Physical PDF page 236; no printed folio visible"
[p237]: /documents/anthropic-s1.pdf#page=237 "Physical PDF page 237; no printed folio visible"
[p238]: /documents/anthropic-s1.pdf#page=238 "Physical PDF page 238; no printed folio visible"
[p250]: /documents/anthropic-s1.pdf#page=250 "Physical PDF page 250; printed filing page 239"
[p254]: /documents/anthropic-s1.pdf#page=254 "Physical PDF page 254; printed filing page 243"
[p255]: /documents/anthropic-s1.pdf#page=255 "Physical PDF page 255; printed filing page 244"
[p256]: /documents/anthropic-s1.pdf#page=256 "Physical PDF page 256; printed filing page 245"
[p257]: /documents/anthropic-s1.pdf#page=257 "Physical PDF page 257; printed filing page 246"
[p258]: /documents/anthropic-s1.pdf#page=258 "Physical PDF page 258; printed filing page 247"
[p259]: /documents/anthropic-s1.pdf#page=259 "Physical PDF page 259; printed filing page 248"
[p260]: /documents/anthropic-s1.pdf#page=260 "Physical PDF page 260; printed filing page 249"
[p261]: /documents/anthropic-s1.pdf#page=261 "Physical PDF page 261; printed filing page 250"
[p262]: /documents/anthropic-s1.pdf#page=262 "Physical PDF page 262; printed filing page 251"
[p263]: /documents/anthropic-s1.pdf#page=263 "Physical PDF page 263; printed filing page 252"
[p264]: /documents/anthropic-s1.pdf#page=264 "Physical PDF page 264; printed filing page 253"
[p275]: /documents/anthropic-s1.pdf#page=275 "Physical PDF page 275; printed filing page 264"
[p281]: /documents/anthropic-s1.pdf#page=281 "Physical PDF page 281; printed filing page 270"
[p282]: /documents/anthropic-s1.pdf#page=282 "Physical PDF page 282; printed filing page 271"
[p283]: /documents/anthropic-s1.pdf#page=283 "Physical PDF page 283; printed filing page 272"
[p285]: /documents/anthropic-s1.pdf#page=285 "Physical PDF page 285; printed filing page 274"
[p286]: /documents/anthropic-s1.pdf#page=286 "Physical PDF page 286; printed filing page 275"
[p287]: /documents/anthropic-s1.pdf#page=287 "Physical PDF page 287; printed filing page 276"
[p291]: /documents/anthropic-s1.pdf#page=291 "Physical PDF page 291; printed filing page 280"
[p292]: /documents/anthropic-s1.pdf#page=292 "Physical PDF page 292; printed filing page 281"
[p293]: /documents/anthropic-s1.pdf#page=293 "Physical PDF page 293; printed filing page 282"
[p295]: /documents/anthropic-s1.pdf#page=295 "Physical PDF page 295; printed filing page 284"
[p296]: /documents/anthropic-s1.pdf#page=296 "Physical PDF page 296; printed filing page 285"
[p297]: /documents/anthropic-s1.pdf#page=297 "Physical PDF page 297; printed filing page 286"
[p299]: /documents/anthropic-s1.pdf#page=299 "Physical PDF page 299; printed filing page 288"
[p303]: /documents/anthropic-s1.pdf#page=303 "Physical PDF page 303; printed filing page 292"
[p316]: /documents/anthropic-s1.pdf#page=316 "Physical PDF page 316; printed filing page 305"
[p318]: /documents/anthropic-s1.pdf#page=318 "Physical PDF page 318; printed filing page 307"
[p319]: /documents/anthropic-s1.pdf#page=319 "Physical PDF page 319; printed filing page 308"
[p320]: /documents/anthropic-s1.pdf#page=320 "Physical PDF page 320; printed filing page 309"
[p327]: /documents/anthropic-s1.pdf#page=327 "Physical PDF page 327; printed filing page 316"
[p332]: /documents/anthropic-s1.pdf#page=332 "Physical PDF page 332; printed filing page 321"
[p333]: /documents/anthropic-s1.pdf#page=333 "Physical PDF page 333; printed filing page 322"
[p334]: /documents/anthropic-s1.pdf#page=334 "Physical PDF page 334; printed filing page 323"
[p345]: /documents/anthropic-s1.pdf#page=345 "Physical PDF page 345; printed filing page 334"
[p347]: /documents/anthropic-s1.pdf#page=347 "Physical PDF page 347; printed filing page F-1"
[p348]: /documents/anthropic-s1.pdf#page=348 "Physical PDF page 348; printed filing page F-2"
[p349]: /documents/anthropic-s1.pdf#page=349 "Physical PDF page 349; printed filing page F-3"
[p350]: /documents/anthropic-s1.pdf#page=350 "Physical PDF page 350; printed filing page F-4"
[p351]: /documents/anthropic-s1.pdf#page=351 "Physical PDF page 351; printed filing page F-5"
[p353]: /documents/anthropic-s1.pdf#page=353 "Physical PDF page 353; printed filing page F-7"
[p356]: /documents/anthropic-s1.pdf#page=356 "Physical PDF page 356; printed filing page F-10"
[p358]: /documents/anthropic-s1.pdf#page=358 "Physical PDF page 358; printed filing page F-12"
[p360]: /documents/anthropic-s1.pdf#page=360 "Physical PDF page 360; printed filing page F-14"
[p362]: /documents/anthropic-s1.pdf#page=362 "Physical PDF page 362; printed filing page F-16"
[p363]: /documents/anthropic-s1.pdf#page=363 "Physical PDF page 363; printed filing page F-17"
[p366]: /documents/anthropic-s1.pdf#page=366 "Physical PDF page 366; printed filing page F-20"
[p367]: /documents/anthropic-s1.pdf#page=367 "Physical PDF page 367; printed filing page F-21"
[p368]: /documents/anthropic-s1.pdf#page=368 "Physical PDF page 368; printed filing page F-22"
[p372]: /documents/anthropic-s1.pdf#page=372 "Physical PDF page 372; printed filing page F-26"
[p378]: /documents/anthropic-s1.pdf#page=378 "Physical PDF page 378; no printed folio visible"
[p381]: /documents/anthropic-s1.pdf#page=381 "Physical PDF page 381; printed filing page F-35"
[p383]: /documents/anthropic-s1.pdf#page=383 "Physical PDF page 383; printed filing page F-37"
[p385]: /documents/anthropic-s1.pdf#page=385 "Physical PDF page 385; printed filing page F-39"
[p393]: /documents/anthropic-s1.pdf#page=393 "Physical PDF page 393; printed filing page F-47"
[p394]: /documents/anthropic-s1.pdf#page=394 "Physical PDF page 394; printed filing page F-48"
[p399]: /documents/anthropic-s1.pdf#page=399 "Physical PDF page 399; printed filing page F-53"
[p401]: /documents/anthropic-s1.pdf#page=401 "Physical PDF page 401; printed filing page F-55"
[p402]: /documents/anthropic-s1.pdf#page=402 "Physical PDF page 402; printed filing page F-56"
[p403]: /documents/anthropic-s1.pdf#page=403 "Physical PDF page 403; printed filing page F-57"
[p404]: /documents/anthropic-s1.pdf#page=404 "Physical PDF page 404; printed filing page F-58"
[p405]: /documents/anthropic-s1.pdf#page=405 "Physical PDF page 405; printed filing page F-59"
[p406]: /documents/anthropic-s1.pdf#page=406 "Physical PDF page 406; printed filing page F-60"
[p407]: /documents/anthropic-s1.pdf#page=407 "Physical PDF page 407; printed filing page F-61"
[p408]: /documents/anthropic-s1.pdf#page=408 "Physical PDF page 408; printed filing page F-62"
[p409]: /documents/anthropic-s1.pdf#page=409 "Physical PDF page 409; printed filing page F-63"
[p410]: /documents/anthropic-s1.pdf#page=410 "Physical PDF page 410; printed filing page F-64"
[p411]: /documents/anthropic-s1.pdf#page=411 "Physical PDF page 411; printed filing page F-65"
[p413]: /documents/anthropic-s1.pdf#page=413 "Physical PDF page 413; printed filing page II-1"
[p415]: /documents/anthropic-s1.pdf#page=415 "Physical PDF page 415; printed filing page II-3"
[p418]: /documents/anthropic-s1.pdf#page=418 "Physical PDF page 418; printed filing page II-6"
[p419]: /documents/anthropic-s1.pdf#page=419 "Physical PDF page 419; printed filing page II-7"