Anthropic IPO Prospectus: Key Takeaways and Valuation Insights

Anthropic, a prominent AI company, has filed its prospectus, a 416-page document that offers insights into its business model, financials, and future prospects. This extensive document, while typical for major IPOs of the last decade, presents a challenge for investors due to its sheer volume and the nature of its disclosures.

Navigating the Prospectus: A Glass Half Full or Half Empty?

The information within the prospectus can be interpreted in two ways: as good news (glass half full) or bad news (glass half empty). This duality applies to several key areas:

Revenue Growth

• Glass Half Full: Anthropic reports exponential revenue growth, with a 1300-1400% increase year-over-year and significant quarter-over-quarter growth. Revenues have surged from negligible amounts to over $16 billion in the last quarter, with annualized run rates potentially reaching $100 billion by year-end. The company boasts 11,900 customers, with approximately 1,900 spending over $1 million, and many repeat customers increasing their spending. This suggests a fast-growing company with sticky customers.
• Glass Half Empty: The prospectus omits detailed third-quarter numbers, which, according to investor meetings, showed a stalling of revenue growth. This slowdown is attributed to three factors:
  • Compute Power Constraints: Anthropic claims it couldn't grow as fast as desired due to limitations in compute power.
  • Scaling Challenges: Achieving 100% quarter-over-quarter growth becomes increasingly difficult as the company scales.
  • Increased Competition: OpenAI's competitive pricing is impacting Anthropic's revenue growth. Additionally, revenue concentration remains a concern, with the top five customers still accounting for 25% of revenues. A significant portion of Anthropic's reported gross revenues comes from partnerships with big tech companies, where 15-20% is paid back to the partner as a marketing expense. While not deceptive, this differs from OpenAI's net revenue reporting and can make direct comparisons challenging.

Profitability

• Glass Half Full: Expenses are growing at a slower rate than revenues, demonstrating economies of scale. Compute costs, a critical expense, have dropped from 250% of revenue six quarters ago to less than 50%. This trend has led to shrinking losses and, in the last reported quarter, a turn to operating profit. This trajectory suggests a path to sustained profitability.
• Glass Half Empty: Anthropic's gross margin is high, close to 80% (even after reclassifying partnership revenue portions as cost of goods sold). However, this is partly achieved by moving compute costs from a variable to a fixed cost category. While this improves unit economics, it means the company needs to maintain significant growth to keep operating profits positive. The single profitable quarter is highly dependent on continued growth. Stock-based compensation is also a substantial and growing expense, which could become a drain if revenue growth falters.

Reinvestment

• Glass Half Full: While traditional capital expenditure appears small, Anthropic is making significant investments in compute power through contractual commitments. These commitments, totaling over $450 billion (and collective commitments exceeding $600 billion), primarily with big tech companies like Broadcom, are crucial for growth in the AI business. This strategy allows Anthropic to lock in essential compute power, playing catch-up with competitors like OpenAI. The potential for chips to become cheaper over time and built-in flexibility in some commitments are also positive signs.
• Glass Half Empty: The $600 billion in commitments represents a massive fixed cost for a young, money-losing company. If growth doesn't materialize, these commitments could become a significant drag. There's also uncertainty about whether compute power will indeed get cheaper, as data centers are becoming more expensive to build and chip costs might not decline as rapidly as expected, leading to continuous and incrementally draining investments.

Risk

• Glass Half Full: Anthropic's customers show strong loyalty and repeat business, suggesting sticky, potentially non-discretionary revenues. The company has accumulated a substantial cash balance of $114 billion (potentially $200 billion after the IPO), providing a buffer against risks and enabling it to meet near-term compute commitments. Regulatory risks, while present, could also create a competitive moat, keeping out cheaper competition.
• Glass Half Empty: Significant operating risks exist due to high fixed costs that require sustained growth to meet. Financial risk is substantial, as compute commitments are contractual and akin to debt; failure to meet them could lead to business failure. Regulatory actions could also significantly impact operating income and margins, potentially jeopardizing the business if stringent enough.

Summary of Prospectus Insights

• Revenue Growth: Fast growth and strong customer engagement are positives, but slowing growth, increasing competition, and compute power constraints are negatives. The circularity of revenue within the AI ecosystem (companies buying services from each other) adds opacity.
• Profitability: Expenses growing slower than revenues and improving margins are positive. However, profitability is a bet on continued growth, as large compute cost commitments will hit income statements if growth falters.
• Reinvestment: Catching up on compute power and improving unit economics by fixing variable costs are positives. The downside is creating a much larger fixed cost base that necessitates sustained growth.
• Risk: A large cash buffer and the company's potential "too big to fail" status are positives. Operating, financial, and failure risks remain significant, with a 10% probability of failure (though not necessarily a complete loss of value).

Valuation: Looking to the Future

Valuation requires moving beyond historical data to make judgments about the future.

Target Revenues

The total addressable market for AI depends on several factors: * AI as a tool vs. employee replacement: Replacement implies a much larger market. * Targeting highly paid vs. all employees: Targeting all employees expands the market. * Subset vs. all businesses: Broader application means a larger market. * US/Europe vs. global reach: Global reach significantly increases market size.

The belief is that AI will disrupt some industries, targeting white-collar and highly paid workers, leading to a total addressable market of $5-10 trillion. Anthropic is positioned as a lead player in the premium segment, with 80% of its revenue from usage-based enterprise models. A plausible target revenue for Anthropic in 10 years is $700-800 billion, potentially reaching $1 trillion if it also becomes a direct player in certain businesses (e.g., pharmaceuticals) rather than just a supplier.

Profitability (Target Margins)

The AI market can be divided into: * Mass Market: Less powerful, commoditized products with margins closer to business services (around 10%). * Premium Segment: Powerful agents with high margins (closer to enterprise software, 40%), driven by client data and stickiness.

Anthropic is expected to focus on the premium segment (75%) but also participate in the mass market (25%), leading to an unadjusted operating margin of 32.5%. This margin will be adjusted later for regulatory risks.

Reinvestment

Compute power investments age quickly, requiring continuous upgrades. While chips may get cheaper, the overall power, real estate, energy, and water needs for compute power will increase. The assumption is that the current compute power crunch will ease, leading to a steady-state reinvestment ratio of $1 of capital for every $3 of additional revenue.

Risk Story

• Operating Risk: The cyclicality of demand for AI agents is unknown. Anthropic carries the risk of large technology companies with the burden of compute commitments. A cost of capital of 9.74% is used, reflecting its current risk profile, converging to the median US company cost of capital (9.13%) over time.
• Financing Risk: Contractual compute commitments are like debt. Anthropic's current bond rating is BB+, below investment grade. However, a large cash balance (estimated $200 billion post-IPO) and its "too big to fail" status mitigate failure risk. A 10% probability of failure is assigned, where value would be half if it occurs.
• Regulatory Risk: Anthropic has invited government regulation, which, while potentially creating a competitive moat, will likely increase costs and reduce margins. A 5% reduction in operating margins (from 32.5% to 27.5%) is assumed due to regulatory effects.

Valuation Outcome

Based on these assumptions: * Target revenues of $1 trillion in 10 years. * Adjusted margins of 27.5%. * Reinvestment efficiency of $3 revenue per $1 capital. * Cost of capital starting at 9.74% and converging to 9.13%. * 10% failure risk.

The value of Anthropic's operating assets is estimated at $1.7 trillion. Adding a cash balance of $220 billion (including IPO proceeds) and subtracting debt (present value of contractual commitments, estimated at $473 billion), the equity value is $1.48 trillion. This is close to the lower end of rumored offering prices. Based on an estimated share count, the per-share value is $447.

Best and Worst Case Scenarios

• Worst Case: AI becomes a mass market with lower margins, revenue growth reflects AI as merely a tool, and the cost of capital rises. This could lead to a negative equity value, implying bankruptcy.
• Best Case: AI replaces people across businesses, leading to $1.5 trillion in revenues, premium market margins of 40%, cheap compute power, and a lower cost of capital (first quartile). This could result in an equity value of $4 trillion.

Corporate Governance

Anthropic's corporate governance structure is highly complex and raises concerns about shareholder power. It features five classes of shares: * Class A: Public shares, one vote per share. * Employee Shares: Voting rights limited to benefit trust changes. * Class N: Corporate shares (Alphabet, Amazon), no voting rights, but share in profits. * Class T: Held by a benefit trust, which will elect the majority of the board. * Class F: Held by Anthropic's seven co-founders, with 50.1% of voting rights.

Anthropic is also a public benefit corporation under Delaware law, allowing it to balance the interests of multiple stakeholders. This provides legal cover to make decisions that may not maximize shareholder value, potentially protecting the company from lawsuits. The benefit trust, composed of founder-picked individuals, further centralizes power. This structure is described as "dictator-level corporate governance," significantly reducing shareholders' ability to influence the company.

IPO and Trading

The IPO offering price is expected to be between $1.5 trillion and $2 trillion. The stock will likely be traded based on mood and momentum rather than fundamental valuation initially. Unlike most IPOs, insiders (excluding founders and key figures) will be allowed to sell up to 10% of their shares on day one, with remaining shares unlocking in tranches over 180 days. Founders and key figures can trade after 365 days. While this might concern traders, historical data suggests the impact of unlockings is often overstated as markets anticipate these events.

Given the current valuation of $1.48 trillion based on an upbeat story, if Anthropic is priced significantly higher, it would be considered too richly priced for investment. However, as the story evolves and news emerges, the price will fluctuate. If the price drops sufficiently below the estimated value, especially considering the governance structure, it could become an attractive investment opportunity in the future.