Insights Investing
Analyzing Anthropic’s IPO Prospectus: Evelyn's Lens on AI Capital Cycles and Balance Sheet Risks
Podcast · Listen (11 min)
Anthropic's IPO: $900B Committed, $13.8B Guaranteed | Investor Playbook
In brief
DeepValues Research, “Analyzing Anthropic’s IPO Prospectus: Evelyn's Lens on AI Capital Cycles and Balance Sheet Risks” (Nov 20, 2026): An analysis of the prospective IPO filing breakdown published by Evelyn's Lens demonstrates that while artificial intelligence represents a secular technological shift, Anthropic's underlying financial disclosures reveal significant structural vulnerabilities.
Source: https://research.deepvalues.ai/insights/2026-11-20-evelyns-lens-the-biggest-ipo-in-history-should-you-buy-the-ai-bubble/ · Markdown
Contents
- Who wrote it
- What was written
- The Three Capital Classifications (Red, Blue, Purple Money)
- Historical Analogies & Bubble Typology
- What the documents show
- The $900 Billion Long-Term Commitment
- The Severe Duration Mismatch
- Margin Volatility and Operational Leverage Decay
- Financial Disclosure vs. Structural Reality
- The Buffett reading
- What is not known
- The lesson for a patient investor
- Key numbers
An analysis of the prospective IPO filing breakdown published by Evelyn's Lens demonstrates that while artificial intelligence represents a secular technological shift, Anthropic's underlying financial disclosures reveal significant structural vulnerabilities. Despite rapid top-line growth and temporary margin improvements, the company's ledger exhibits severe capital duration mismatches, substantial exposure to venture-capital-subsidized end demand, and circular financing structures with major cloud infrastructure providers. Anthropic has committed to hundreds of billions in long-term, non-cancelable compute lease liabilities while relying primarily on short-term, pay-as-you-go usage revenue. For value-minded investors evaluating market front-runners, distinguishing between secular technological adoption and sustainable equity performance requires inspecting corporate balance sheets, capital recycling flows, and fixed-cost structures.
Who wrote it#
The financial breakdown of Anthropic’s filing was presented by Evelyn's Lens, an independent financial analyst and creator behind the YouTube channel Evelyn爱武林. Her professional background includes training as a Wharton graduate, earning the CFA designation, and working as an institutional investor at Two Sigma, accumulating over ten years of professional investing experience across public markets and macroeconomic analysis.
Her analytical methodology relies on unpacking fundamental microeconomics, primary regulatory filings, corporate balance sheets, and overarching macroeconomic trends. Whether analyzing major corporate credit issuers like American Express, evaluating manufacturing supply chains in Vietnam, or dissecting pre-IPO ledgers, her commentary incorporates logic derived directly from official financial disclosures and primary research.
For value investors, this framework offers valuable signal by filtering market narrative through balance sheet realities. Rather than focusing on technological momentum or top-line revenue growth in isolation, her approach assesses capital allocation durability, revenue quality, duration risk, and underlying cash flows.
What was written#
The analysis published by Evelyn's Lens breaks down Anthropic's prospective public filing by examining the quality and origin of its revenue streams, its long-term liabilities, and historical parallels from past capital cycles.
The Three Capital Classifications (Red, Blue, Purple Money)#
To evaluate the durability of Anthropic’s cash flows, the commentary categorizes incoming capital into three distinct buckets:
- Red Money (Circular Capital): This category represents capital looping between major cloud providers (Amazon, Google, and Microsoft) and Anthropic. Cloud giants and institutional entities have injected over $140 billion into Anthropic. In turn, Anthropic has committed to paying nearly $270 billion over a 10-year period in minimum compute lease obligations back to these same cloud providers. Capital enters Anthropic as equity or debt financing and flows back to its investors as high-margin cloud infrastructure revenue, artificially inflating host cloud growth metrics while locking Anthropic into heavy fixed overhead.
- Blue Money (Real Commercial Demand): This bucket reflects organic, self-sustaining end-user demand from cash-generative enterprises. Positive operational metrics highlight genuine commercial adoption: large enterprise accounts spending over $100,000 annually grew from under 1,000 to nearly 12,000, with over 50% of the Fortune 500 adopting Claude. Anthropic operates an 80% usage-based pricing model. Furthermore, compute expenses as a percentage of revenue fell from 251% in 2025 down to 48%, allowing the company to generate a brief positive operating margin of 8.4% in the second quarter.
- Purple Money (Venture-Backed Capital): This classification denotes revenue sourced from early-stage AI startups that are spending external venture capital funding rather than organic operating cash flows. Citing third-party transaction data from Ramp covering major Large Language Model (LLM) providers (combining OpenAI and Anthropic), the analysis highlights that approximately 80% of top-tier enterprise LLM spending originates from the top 1% of enterprise spenders, and roughly 80% of those top spenders are VC-backed startups. This creates an indirect funding loop: institutional capital flows into venture funds, which capitalize AI startups, which then spend lavishly on API tokens. This dynamic mirrors Yahoo’s performance during the 1999 dot-com bubble, where advertising revenue doubled to $1.1 billion primarily because VC-funded startups were burning capital on ad inventory before vanishing when venture funding dried up in 2001.
Historical Analogies & Bubble Typology#
The commentary draws a deliberate distinction between different historical asset bubbles:
- 1999 Internet Bubble ("Good Bubble"): While equity valuations were severely overextended, the massive over-investment yielded permanent, deflationary physical infrastructure, including extensive fiber-optic networks, cheap bandwidth, and lower server costs that dramatically reduced operational expenditures for future businesses. Although equities suffered catastrophic drawdowns—Amazon dropped 94%, Cisco fell 90%, and NetEase plummeted 97%—the underlying physical assets paved the way for the modern digital economy.
- 2008 Subprime Crisis ("Bad Bubble"): Built on excessive financial leverage, unbacked debt, and speculative housing inventory, this crisis failed to produce productive capital or technological advancements, leaving behind non-performing debts, illiquid assets, and systemic financial contagion.
The primary takeaway for investors is that a secular technology can succeed and transform the economy while its initial equity front-runners suffer permanent or long-term capital destruction. Recognizing a valid technological shift does not protect an investor from buying into an over-capitalized asset at peak valuation.
What the documents show#
An inspection of the regulatory disclosures cited in the analysis reveals severe structural balance sheet risks beneath Anthropic's top-line expansion.
The $900 Billion Long-Term Commitment#
Beyond the 10-year, $270 billion baseline compute commitment, regulatory notes after page 166 of the filing reveal total cloud compute procurement and lease obligations reaching nearly $900 billion through 2045. Crucially, over three-quarters of these contractual liabilities contain no provisions for early termination or cost adjustment.
The Severe Duration Mismatch#
The core balance sheet risk lies in the structural mismatch between fixed, long-term liabilities and volatile, uncommitted revenues:
- Total Obligations vs. Peak Revenue: The ~$900 billion long-term commitment represents more than 13 times Anthropic's highest annualized quarterly revenue.
- 12-Month Outflows vs. Guaranteed Inflows: Over the next 12 months, Anthropic is contractually obligated to pay approximately $70 billion in fixed compute and facility leases. However, guaranteed, contracted annual revenue stands at only $13.8 billion.
- The Uncovered Gap: The resulting ~$60 billion annual shortfall must be covered entirely by pay-as-you-go usage revenue. Because 80% of revenue is non-binding and usage-based, cash inflows are highly sensitive to enterprise budget cuts, VC funding pullbacks ("purple money"), or reductions in internal usage by cloud partners ("red money").
Margin Volatility and Operational Leverage Decay#
The fragility of Anthropic's business model manifested in its third-quarter financial results. Despite top-line volume growth, model price reductions slashed revenue by $600 million. Simultaneously, compute expenses surged 77% quarter-over-quarter. This violent spike in operating costs occurred because Anthropic was forced to scramble for high-priced, short-term compute capacity to satisfy demand spikes while locked into rigid baseline lease commitments. As a result of this extreme negative operating leverage, the positive 8.4% operating margin achieved in Q2 reversed back into an operating loss in Q3.
Financial Disclosure vs. Structural Reality#
| Metric / Filing Disclosure | Structural / Analytical Reality |
|---|---|
| Next 12-Month Rigid Commitments (~$70B) vs. Guaranteed Contracted Revenue ($13.8B) | Creates an immediate ~$60 billion annual revenue gap that must be filled by non-guaranteed, pay-as-you-go usage revenue. |
| Total Long-Term Obligations (~$900B through 2045) vs. Peak Annualized Revenue | Total non-cancelable compute liabilities exceed 13x peak annualized revenue, locking in multi-decade fixed costs against highly volatile software demand. |
| Q2 Operating Profit Margin (8.4%) vs. Q3 Margin Reversal (Negative) | Operating leverage is non-linear and unstable. Price reductions cut Q3 top-line by $600M while short-term compute procurement expenses rose 77% QoQ, flushing margins back into negative territory. |
The financial disclosures show that Anthropic's current operating model requires constant, uninterrupted demand growth to satisfy fixed infrastructure obligations. If compute demand growth decelerates or token prices drop, fixed lease costs remain static, triggering sharp negative operating leverage.
The Buffett reading#
Evaluating Anthropic through classical value investing principles—such as those articulated by Benjamin Graham, David Dodd, and Warren Buffett—highlights significant fundamental concerns regarding business quality, balance sheet solvency, and margin of safety.
- Business Quality & Economic Moat: Anthropic's core product output—API access to LLM tokens—operates as a commoditized output sold against capital-intensive fixed inputs (leased compute capacity). With 80% of revenue generated on a flexible, usage-based model and short contract durations, enterprise switching costs are minimal. The $600 million top-line reduction in Q3 driven by competitive price cuts demonstrates an absolute absence of pricing power. Without pricing power or structural lock-in, the business lacks an economic moat to defend returns on capital.
- Capital Allocation & Balance Sheet Risk: From a Graham & Dodd perspective, entering into $900 billion in non-cancelable lease obligations creates an alarming Fixed Charge Coverage Ratio shortfall. Fixed charge coverage measures a firm's ability to cover fixed operational expenses with earnings before fixed charges and taxes. With $70 billion in mandatory 12-month lease commitments set against only $13.8 billion in guaranteed contracted revenue, the firm relies entirely on uncommitted usage volume to remain solvent, introducing unprecedented financial leverage.
- Price versus Value: Seeking a prospective trillion-dollar scale valuation for an enterprise experiencing operating margin flushes and negative operating leverage violates core margin-of-safety tenets. Paying a peak valuation multiple for a company whose operating profit can turn negative in a single quarter due to compute expense spikes leaves equity investors exposed to severe capital impairment.
What is not known#
A full quantitative assessment of Anthropic's financial condition is constrained by several undisclosed parameters in the public disclosures:
- Internal Recycling / Self-Dealing Volume: The exact breakdown of how much reported "blue" commercial revenue actually represents cloud host partners (Amazon, Google, Microsoft) purchasing Claude services for their own internal operations remains unquantified in the filing.
- Granular Startup Exposure: While aggregate market data from Ramp highlights VC concentration across top LLM spenders, Anthropic does not disclose the precise percentage of its own customer ledger represented by non-viable startups dependent on venture subsidies.
- Net Customer Retention and Churn: The disclosures lack granular net revenue retention (NRR) and churn cohort metrics under pay-as-you-go enterprise pricing, preventing an precise evaluation of customer lifetime value if VC funding contracts or switching costs decline.
The lesson for a patient investor#
The analysis of Anthropic’s balance sheet offers vital strategic lessons for disciplined, long-term investors navigating high-growth technology cycles:
- Decouple Technological Adoption from Equity Returns: A technology can transform society while the early market leaders fail to generate long-term equity value. Historical precedents demonstrate that even surviving market leaders can take decades to recover from valuation peaks: Cisco took 25 years to regain its 2000 high, Microsoft required 16 years, Amazon experienced a 94% drawdown over two years (taking a decade to break even), and SoftBank saw its shares drop 99% during the dot-com collapse.
- Beware Duration Mismatches in High-Fixed-Cost Businesses: Capital-intensive software businesses that lock in long-term, non-cancelable infrastructure liabilities while selling uncommitted, pay-as-you-go services face asymmetric downside risk. When top-line growth slows or token prices fall, fixed operational liabilities consume liquidity rapidly.
- Capture the Infrastructure Surplus Without Equity Risk: The most prudent way to capitalize on an over-funded technological boom is often to utilize the resulting deflationary infrastructure. As capital cycles build excess compute, power grids, and software tools, businesses and individuals can adopt these advanced capabilities to enhance productivity, capturing value directly without overpaying for speculative front-runner equities.
Key numbers#
- $140+ Billion: Total capital injected by cloud providers and institutional investors into Anthropic.
- ~$270 Billion: Anthropic's minimum 10-year compute payment obligation back to its host cloud providers.
- ~$900 Billion: Total cloud compute procurement and lease commitments extending through 2045 (over 3/4 non-cancelable).
- 13x+: Ratio of total long-term compute commitments relative to Anthropic's peak annualized quarterly revenue.
- ~$70 Billion: Mandatory compute and lease expenditure due within the next 12 months.
- $13.8 Billion: Total fixed, guaranteed incoming revenue contracted with customers.
- 80%: Share of incoming revenue derived from usage-based pricing without long-term binding contracts.
- ~12,000: Number of enterprise customers spending over $100,000 annually (up from <1,000 in the prior year).
- 251% down to 48%: Drop in compute costs as a percentage of revenue from 2025 to Q2.
- 8.4%: Single-quarter positive operating profit margin achieved in Q2 (reversed back to an operating loss in Q3).
- $600 Million: Direct revenue reduction in Q3 caused by model price reductions.
- 77%: Quarter-over-quarter increase in compute expenses during Q3 driven by short-term compute procurement commitments.
- 80% / 80% (Ramp Data Sample): Approximately 80% of top LLM software spend comes from the top 1% of enterprise spenders; roughly 80% of those top spenders are VC-backed startups across the combined OpenAI and Anthropic sample.
- 94%: Maximum stock drawdown experienced by Amazon after the 1999 dot-com peak before recovering 10 years later.
Watch
Source links
- DEEPVALUES MONITOR BRIEF
- 史上最大IPO,买不买?我从Anthropic账本中看到的AI泡沫