Anthropic Deep Value Investment Research

Hunting ten-year five-baggers among great growth stocks — pressing the upside question: "Can it get much bigger?"

• QHow large is its market ceiling? Is it expanding an existing pie, or creating an entirely new market?7/10

  A

  The ceiling is extremely high, and Anthropic is both “expanding an existing pie” and “opening up a new one.” The real growth-investing question, however, is how much of this enormous pie it can actually take and keep. That is far less certain than the ceiling itself.

  Start with the size of the pie. This is a trillion-level market that is still expanding at close to 50% growth: Gartner expects global AI spending to reach 2.52 trillion dollars in 2026, up about 44% year over year, with the overwhelming majority in AI infrastructure and the model layer only one piece of it. The report’s framing is consistent with that, rating industry attractiveness at 4/5 while adding an important caveat honestly: the McKinsey 2025 global survey shows that 88% of companies now use AI regularly in at least one function, but only about 39% of respondents say AI has affected EBIT at the enterprise level. The demand trend is huge, but large-scale profitability across the industry has not yet broadly landed. So the high ceiling is real; “this pie has already proved it can be monetized steadily” is not yet real.

  As for whether Anthropic is expanding an existing pie or creating a new market, the answer is both, but the current revenue engine is mainly the former. It is taking existing workloads such as “writing code / running agents / enterprise knowledge work,” rebuilding them with LLMs, and making them better. That means taking share from existing spend and migrating software and labor budgets onto tokens. The evidence is that its revenue mix is heavily enterprise-oriented: the report cites Reuters in saying growth is mainly driven by enterprises, especially code generation, while Claude’s consumer traffic is only about 2% of ChatGPT’s. External data supports the same point: about 80% of Anthropic’s revenue comes from enterprise customers, and enterprise API accounts for about 70% of total revenue. The part with true “new market creation” color is products such as Claude Code: it reached more than 2.5 billion dollars of annualized revenue roughly 18 months after public launch and has doubled since early 2026. This “AI engineer in the terminal” format has genuinely created some demand from scratch.

  Measured strictly by Baillie Gifford LTGG standards, however, two limits should not be overstated. First, expanding an existing pie is not the kind of “defining a brand-new category from zero” story Baillie Gifford values most. It looks more like competing for share in a large known demand pool through performance and workflow embedding, with many competitors and fast followers. Second, a high ceiling does not mean Anthropic can own it alone. The report repeatedly stresses that the profit pool may not remain in the model layer: it has committed to spending more than 100 billion dollars with AWS over the next ten years to lock in up to 5GW of compute, so a meaningful portion of the industry “rent” is taken by the compute layer first.

  Conclusion: on market ceiling, Anthropic is a genuinely strong story: a trillion-level, high-growth market where it has already proved rapid penetration. But it is mainly taking share in a huge existing pie rather than uniquely creating a new category, and the profit from that pie has to be shared with hyperscale cloud providers. By growth-investing standards, an upper-middle score is reasonable here, but a large TAM alone should not justify a perfect score.

  Jun 11, 2026
• QCan its revenue at least double over the next five years? Will growth mainly be driven by volume, price, or new businesses?8/10

  A

  It can almost certainly double, and most likely do far more than double. This is Anthropic’s strongest and least controversial dimension. But the honest caveat is that “whether it can double over the next five years” is not really the question. The real question is how long this pace can last, and what exactly is driving the growth.

  First, consider how little suspense there is around doubling. Anthropic’s run-rate revenue trajectory is rare in recent years: about 87 million dollars in January 2024 -> about 1 billion by the end of 2024 -> about 9 billion by the end of 2025 -> 14 billion in February 2026 -> more than 30 billion dollars in April. By Series H, the company disclosed that run-rate had crossed 47 billion dollars. In other words, its annualized revenue grew roughly 10x over the past 12 months. With that base momentum, asking whether revenue can double in five years is almost like asking whether an accelerating train will be faster five years from now. The report’s judgment that revenue growth is “extremely high, and rare” is fully valid; no discount is needed here.

  The key is the second question: is growth driven by volume, price, or new businesses? All three are present, but structurally “volume” is the dominant engine, “price” is a headwind, and “new businesses” are the amplifier.

  • Volume is the main engine. The driver is an explosion in enterprise usage and customer count: large customers spending more than 1 million dollars annually increased from 500+ to 1000+ in two months, and the number of 100,000-dollar-level customers grew about 7x over the past year; eight of the Fortune top ten are already customers. At base, workloads (token calls, seats, agent tasks) are scaling.
  • Price is a headwind, not a tailwind. The report states clearly that Anthropic has only limited pricing power, while Google, OpenAI, and open-source models are all pushing down the unit cost of intelligence; its pricing page is highly transparent and comparable. In other words, unit token prices trend downward over the long run, and growth comes from “selling more tokens,” not from “charging more per token.”
  • New businesses are a real amplifier. Claude Code has already exceeded 2.5 billion dollars in annualized revenue, with enterprise usage accounting for more than half. Cowork, the MCP ecosystem, and industry verticals (finance/healthcare/life sciences) are also contributing incremental growth. This is both volume and “new category,” the hand that pushes the doubling story from “highly likely” toward “far beyond that.”

  Growth investing cannot look only at the numerator. Two implicit premises need to be added, and the report repeatedly flags both. First, run-rate is an extrapolation from a recent period and is unusually sensitive to timing and accounting definitions. The report specifically notes that Reuters mentioned the two companies do not treat cloud revenue sharing in exactly the same way, so simply comparing “annualized revenue multiples” is not rigorous. Any growth discussion based on run-rate should recognize that it may overstate sustainable real revenue. Second, the growth rate itself must inevitably converge downward from an unsustainable “10x/year” pace. Doubling over five years is not the problem, but the market has already priced in expectations far beyond doubling (see the valuation and implied expectations question).

  Conclusion: a five-year revenue doubling is almost certain, with growth led by enterprise usage expansion, amplified by new products (Claude Code/agents), while price is a headwind. This is a textbook high-quality growth engine and deserves a high score. The only restraint is that the “quality” of growth, meaning whether it becomes real cash and whether the accounting holds up under scrutiny, must be examined strictly in the unit economics and valuation questions. A beautiful top line should not make the whole case true by default.

  Jun 11, 2026
• QFive years from now, what will take over as the next growth engine? Does this “second curve” exist today?5/10

  A

  The second curve already exists today, and there is more than one. This is Anthropic’s real advantage over many single-hit product companies. But the honest distinction is that its “second curves” currently look more like natural extensions and broadening of the same main curve (the enterprise AI platform) than a fully independent new growth pole that can take over after the core business peaks. Under Baillie Gifford’s standard, that deserves credit, but it is not yet in the “verified second growth engine” tier.

  Look first at the candidates that could take over, all of which have landed and are not PPT:

  First, Claude Code and agentic workflows. Claude Code reached more than 2.5 billion dollars of annualized revenue roughly a year and a half after public launch, and enterprise usage already accounts for more than half. This moves Anthropic from “selling model tokens” toward “selling AI engineers/agents that can do work,” a migration from chat interface to productivity system with higher unit value.

  Second, the MCP (Model Context Protocol) ecosystem and connectors. The report cites Google’s official disclosure that MCP is now compatible with the Google Gemini API/SDK, meaning Anthropic is not only a model supplier but also shaping the interface standard for the agent era. The report notes that the May 2026 acquisition of Stainless strengthened SDK and MCP server tooling, which fits directly with this line. If MCP truly becomes an industry standard, it could be a more durable asset than the model itself.

  Third, enterprise verticals and multi-cloud distribution. The report lists expansion into professional scenarios such as finance, healthcare, life sciences, cybersecurity, and the Japanese market, with enterprise reach through AWS, Google Cloud Vertex AI, and Microsoft Azure Foundry. This replicates the same capability horizontally across industries and channels.

  But under the strict Baillie Gifford question of “what takes over five years from now,” two non-inflated caveats are necessary.

  First, these “second curves” share the same root as the main curve and are not independent new poles. Claude Code, Cowork, MCP, and vertical industries are all different faces of “better model capability + workflow embedding,” relying on the same underlying model moat and the same compute base. If the core business, meaning frontier model leadership, is matched, these extensions will probably come under pressure together rather than countercyclically taking over like a true second curve when the main business peaks. The report’s long-term moat judgment of “unproven” points exactly to this.

  Second, they are still in a cash-burning phase and have not yet proved they can independently generate cash. The report states clearly that the company has had large historical cash burn, and management had expected not to stop burning cash until 2027. A group of businesses that has not yet funded itself deserves a discount when discussed as a “second growth engine”: it has proved the direction, not the sustainable economics.

  Conclusion: second curves do exist today, and they are multiple real businesses that have landed and are scaling (Claude Code, MCP/agent standards, enterprise verticals, multi-cloud distribution). This gives Anthropic clearly better “growth regenerativity” than single-product companies. But these curves share the same root, compute, and moat as the core business; they are more a widening of the main curve than independent successors, and their economics remain unverified while losses persist. An upper-middle score is reasonable for this dimension, but it should not be scored as if the company already owns a mature second engine.

  Jun 11, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?5/10

  A

  The core competitive advantage is a combination of “frontier model capability + workflow embedding + enterprise-grade safety and governance trust + multi-cloud distribution,” not any single hard moat. The three-to-five-year judgment is widening in the short to medium term, still unproven over the long term. That is exactly why the report scores moat strength only 3/5, and I agree with that restraint.

  First, where the advantage is specific and evidence-based:

  • Scale and customer embedding are widening. Large customers spending more than 1 million dollars annually rose from 500+ to 1000+ in two months. Once enterprises connect Claude to codebases, document repositories, M365, Slack, and internal search, migration costs rise. This is a real but “moderate” switching cost.
  • Channel is a structural advantage. Being available through all three major clouds, AWS, Google Cloud Vertex AI, and Azure Foundry, is very friendly to enterprise procurement, and the report rates it “medium-strong.”
  • An interface-standard network effect is forming. MCP is already compatible with the Google Gemini SDK. If it becomes the de facto standard for the agent era, it could create more durable lock-in than the model itself.
  • Culture and safety governance are real strengths. The report lists “culture and operating capability” as a standalone strength; safety-first, research-driven behavior has trust value in regulated industries.

  But under strict Baillie Gifford standards, this moat has three hard flaws that must not be overstated:

  First, the most central moat, “model performance leadership,” is a depreciating moat, not an appreciating one. The report is blunt: in base tokens, general models, and the broad chat market, it is hard for Anthropic to have Visa/Moody's-like pricing power, while Google, OpenAI, and open-source models keep pushing down the unit cost of intelligence. Once performance gaps converge, the share and price supported by performance come under pressure together. A frontier model’s “past lead” proves the past, not leadership three to five years from now.

  Second, there is no sustainable cost advantage; cost is instead a disadvantage. The report judges “cost advantage” as weak: compute is a strategic, capital-heavy cost that may swallow the profit pool. Anthropic has committed more than 100 billion dollars to AWS over ten years and has locked in multi-gigawatt TPU capacity with Google/Broadcom. A moat with such heavy external dependence is less pure.

  Third, the barrier to replication is high, but not insurmountable. The report honestly acknowledges that replicating “today’s Anthropic” would require years, top-tier teams, and billions to tens of billions of dollars in capital. That is the road Anthropic itself just traveled. The barrier is “very high,” not “impossible to cross,” and OpenAI and Google both have the capital and compute.

  Looking ahead three to five years, which is the target of the moat question, the evidence for widening is concentrated in deeper workflow embedding (Claude Code with enterprise accounting for more than half), MCP standardization, deeper multi-cloud penetration, and accumulated safety trust in regulated industries. The risks of narrowing are concentrated in model capability commoditization, price wars, improving open-source alternatives, and customers viewing the model layer as a replaceable component (the report’s “risk of business model disruption”). Which force wins is not yet settled.

  Conclusion: Anthropic has a real but “combinational, medium-strength” moat. Its strengths are scale/channel/workflow embedding/safety culture; its weaknesses are cost and the depreciating nature of model performance. It is genuinely widening in the short to medium term, but whether it can hold over the long run depends on whether workflow lock-in and interface standards can settle into true institutional barriers. That has not yet been proved. A middle score here (consistent with the report’s 3/5) is fair, and fast growth should not automatically lift the moat score as well.

  Jun 11, 2026
• QIf the core business is disrupted, does it have the genes for self-reinvention? How does it handle mistakes and bad news?5/10

  A

  Anthropic shows a genuine strength in “how it handles mistakes and bad news”: it is transparent, willing to discuss risks publicly, and responds to crises with reasoned resistance rather than concealment. As for “self-reinvention genes if the core business is disrupted,” it has the two most important prerequisites, first-class research regeneration and a capital and compute base it is willing to fund at any cost. But this gene has not yet been tested by a true “life-or-death” cycle, so it should be described as “promising, unverified.”

  Start with its handling of mistakes and bad news, where the evidence is fairly hard and the assessment can be high. The report lists “culture and operating capability” as a strength, citing safety-first, research-driven, restrained organization. Two current facts support the view that it faces bad news directly:

  First, copyright litigation. Anthropic did not avoid the issue, but moved toward settlement: the 1.5 billion-dollar settlement in Bartz v. Anthropic is one of the largest copyright settlements in U.S. history; in that case, at the May 14, 2026 hearing, the judge had not yet granted final approval and requested more details on attorney fees and lead plaintiff allocation, so it remains in progress. Facing the issue rather than delaying is a positive signal.

  Second, the Pentagon dispute. After the Department of Defense listed it as a “supply chain risk,” Anthropic sued, publicly asserted its red lines (no mass surveillance, no autonomous weapons without human oversight), and candidly disclosed the financial impact. Management estimated in litigation that the event could affect “several billion dollars” of 2026 revenue, even though a federal appeals court rejected its request to temporarily halt the blacklist and only agreed to expedite the case. Openly stating bad news that could hurt several billion in revenue, while bearing commercial cost for principles, is exactly the kind of “sacrificing short-term interests for long-term value” honesty that Baillie Gifford likes.

  Now on self-reinvention genes, restraint is necessary, and both sides should be made clear.

  The prerequisites are real. First, first-class research regeneration: Anthropic’s core identity is research-driven, and it has reportedly been significantly more training-efficient (overtaking OpenAI’s revenue with about 4x lower training spend). That ability to produce frontier models with fewer resources is fundamental if a disruption requires rapid iteration. Second, a base it is willing to fund at any cost: it has locked in more than 100 billion dollars of AWS compute over ten years and raised 65 billion dollars in Series H at a valuation of about 96.5 billion dollars, meaning that even if one technical path is disrupted, it has the ammunition to fight again.

  The unverified flaws are equally real. First, Anthropic is too young (founded in 2023) and has never lived through a complete cycle in which its own core business was disrupted and then rebuilt from the edge. Its “reinvention gene” is inferred, not historically demonstrated. The report’s repeated emphasis that the long-term moat is “unproven” has the same root. Second, its pressure point is exactly “commoditization of the model layer.” The report’s “risk of business model disruption” is clear: if enterprises view models as replaceable components, value will move down to the data and application layers. Whether Anthropic can then shift successfully from “selling models” to “selling workflows/agent platforms” looks directionally right today (Claude Code, MCP), but no one has seen it complete that turn under disruption.

  Conclusion: on handling mistakes and bad news, Anthropic performs very well and deserves a high assessment. On self-reinvention genes, it holds the two critical cards, research regeneration and a capital/compute base, but it has never faced a true disruption test, and its core business is exposed to the pressure point of commoditization. That supports an upper-middle “promising but unverified” assessment; crisis communication alone should not make the reinvention capability a perfect score.

  Jun 11, 2026
• QDoes management, especially the founders, have a long-term perspective, and are their interests deeply tied to the company? Are they willing to sacrifice current profits for five to ten years from now?5/10

  A

  The founders have a very strong long-term perspective and are clearly willing to sacrifice current profits for five to ten years from now. This is one of Anthropic’s high-scoring areas under the Baillie Gifford framework. But “interests deeply tied to the company” needs to be separated: the founders’ mission is deeply tied to the company (strong), while pure financial alignment with outside minority shareholders has been deliberately diluted by the governance structure (a double-edged sword, which the report rightly calls “an important fact rather than a footnote”).

  First, the evidence for long-term perspective and “willingness to sacrifice the present for the long term” is hard:

  • The governance design itself institutionalizes long-termism. Anthropic is a Public Benefit Corporation and has established the Long-Term Benefit Trust (LTBT); by April 2026, with Novartis CEO Vas Narasimhan joining the board, Trust-appointed directors held a majority of board seats. Writing “public benefit and long-term mission” into board control is a long-term commitment far beyond ordinary companies.
  • Willingness to sacrifice short-term profit is behavior, not a slogan. It has had large historical cash burn, with management previously expecting cash burn not to stop until 2027, while still committing more than 100 billion dollars to AWS over ten years to secure compute. That is the classic choice to sacrifice near-term cash flow for a platform position 5-10 years out.
  • It bears commercial cost for principles. In the Pentagon dispute, it would rather be blacklisted and absorb what management described as a “several billion dollars” 2026 revenue impact than abandon its red lines against mass surveillance/autonomous weapons. Short term, that loses money; long term, it protects brand and trust.

  But “interests deeply tied to the company” must be split honestly rather than scored high in a blanket way:

  First, mission alignment is strong, while financial alignment is weak from the outside shareholder’s perspective. The report’s point is accurate: the Delaware PBC + LTBT structure allows the board to legally consider public benefit, affected stakeholders, and shareholder interests together, and Trust members themselves have no financial interest in the company. That is a social strength, but for value investors who care only about per-share intrinsic value growth, it means the company is not run solely to maximize shareholder returns. This is a real cost.

  Second, the founders’ personal ownership cannot be rigorously verified. Public information indicates that Dario and Daniela Amodei and other co-founders retain meaningful but diluted equity, though the exact percentages are not publicly disclosed. So “founder interests are deeply tied to the company” is true at the mission/control level, but unproven at the level of verifiable financial alignment. After the large G and H rounds (Series H raised 65 billion dollars at a valuation of about 96.5 billion dollars), founder ownership has probably been materially diluted. The report also states clearly that equity incentives and dilution cannot yet be rigorously assessed because the company is private and lacks full capitalization disclosure.

  Third, board quality is a positive. The combination of Dario, Daniela, Reed Hastings, Chris Liddell, Vas Narasimhan, and others brings technology, enterprise, and public-governance experience. The broad capital allocation direction, research/compute/small acquisitions close to the main path such as the May 2026 acquisition of Stainless, is not unreasonable. But the report also honestly says this is still one full cycle away from “proven excellent capital allocation,” and there is no dividend or buyback record.

  Conclusion: the founders’ long-term perspective and willingness to sacrifice the present for the long term are among Anthropic’s most testable strengths and deserve a high assessment. But “deep interest alignment” must be layered: mission and control alignment are strong, while financial alignment with outside minority shareholders is deliberately ceded by the PBC/LTBT structure, and founder personal ownership has been diluted through multiple financings and cannot be verified. The combined assessment is upper-middle to strong, but the “public-benefit, not purely shareholder-oriented” governance tilt must be recorded as a clear caveat, not treated as an unqualified positive.

  Jun 11, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth sustainable and not dependent on harming society or regulatory arbitrage?6/10

  A

  If Anthropic disappeared tomorrow, enterprise customers, especially heavy Claude Code users and those that have embedded Claude into core workflows, would “miss it a lot, but find substitutes.” It is already close to a “critical supplier” in enterprise/coding scenarios, but it is far from the tier of “irreplaceable infrastructure.” On whether its growth is sustainable and not dependent on harming society or regulation, this is one of its cleanest and most commendable advantages relative to peers, while it still carries two real regulatory/legal tails. These two layers need to be discussed separately and honestly.

  First, “how much would they miss it” (indispensability). The evidence points to “strong demand, moderate lock-in”:

  • The intensity of missing it is real. Large customers spending more than 1 million dollars annually already exceed 1000, and eight of the Fortune top ten are customers; Claude Code has more than 2.5 billion dollars in annualized revenue, with enterprise accounting for more than half. Enterprises that have connected Claude to codebases/document repositories/M365/Slack do face real migration costs. The report rates switching costs as “moderate.”
  • But substitutes are ready. The report is blunt: on the consumer side, Claude’s traffic is only about 2% of ChatGPT’s, so it is not the first mass-market entry point; on the enterprise side, competitors (OpenAI with about 25 billion dollars in annualized revenue, Google, open-source models) can step in. The report’s “risk of business model disruption” states that if enterprises view the model layer as a replaceable component, value will move down to the data and application layers. So “missing it” is more about switching pain than sole-source indispensability where operations stop without it. By Baillie Gifford’s high bar for true indispensability, this can only be upper-middle, not perfect.

  Now on “whether growth is sustainable and not dependent on harming society or regulation.” This is the implicit double premise of the question (indispensability + social/regulatory sustainability), and it is also Anthropic’s relative bright spot:

  Positive side (sustainable, even trading restraint for trust): it is a company that writes safety governance into the business model: PBC + LTBT structure, with Trust-appointed directors already holding a majority of the board. In the Pentagon dispute, it would rather be blacklisted than do mass surveillance or autonomous weapons without human oversight. Its growth is not being fattened by regulatory arbitrage or user harm, which is a real long-term positive in YMYL/regulated industries.

  But two tails must be stated honestly, not hidden behind the good story:

  First, the “original sin” of copyright has not been resolved. The 1.5 billion-dollar settlement in Bartz v. Anthropic is one of the largest copyright settlements in U.S. history, but as of May 14, 2026 the judge had not granted final approval, and more than 25 authors rejected the settlement and filed another suit. This shows that the legality of its early training data remains historically disputed, a real reservation on whether the growth method is fully clean.

  Second, government/policy boundaries are real risks, not paper risks. The Pentagon dispute has already been described by the company as possibly affecting “several billion dollars” of 2026 revenue, and the appeals court also did not halt the blacklist. Holding principles deserves respect, but the objective result is pressure on some government revenue. Sustainability has friction costs in its tension with regulation.

  Conclusion: on indispensability, Anthropic is “strong demand + moderate lock-in”; enterprise customers would miss it a lot but could replace it, so under Baillie Gifford standards it gets an upper-middle rather than full score. On social/regulatory sustainability, its safety-governance DNA is among the cleanest and most commendable in the peer group, but the copyright settlement is not final and government boundaries remain contested. Those two tails must be recorded. Overall, the assessment is clearly positive, and the company’s choice to trade restraint for long-term trust is exactly the kind of trait long-term investors like Baillie Gifford would reward.

  Jun 11, 2026
• QWhat are the unit economics of this business (gross margin, incremental returns)? Do they improve or deteriorate with scale? Where does the money it earns go?4/10

  A

  Unit economics are improving rapidly from “disastrous” to “respectable but still far from top-tier software,” and gross margin is improving as scale grows. This is the most substantive positive change over the past year. But the report is right to remain conservative overall because there is no public-company-grade disclosure. Nearly all money goes back into research, compute, and distribution, with zero dividends or buybacks. This question needs to hold “excellent direction” and “still uncomfortable absolute level” together.

  First, correct one area where the report was conservative because of timing, and supplement it with first-hand data. When the report was written (based on 2024 burn and early metrics), it marked gross margin/operating margin as “unknown,” inferred from cash burn that Owner Earnings were likely still negative, and conservatively estimated about -3 billion dollars/year. That caution was responsible. But as of mid-2026, there is now a more concrete gross-margin trajectory to reference: Anthropic’s 2024 gross margin was about -94%, recovered to about 40% in 2025 (revised down from an internal 50% target because inference costs on Google/Amazon clouds were about 23% higher than expected), and management’s 2028 gross-margin target is about 77%; separately, according to SemiAnalysis, its inference gross margin rose from about 38% to about 70% within a year.

  These numbers have three implications for unit economics:

  First, incremental returns (scale effects) are genuinely positive. Moving from -94% to about 40% gross margin, and inference margin to about 70%, shows that as usage scales and model/inference efficiency improves, the marginal economics of each additional unit of token sold are improving materially. This is evidence that “scale makes it better,” and the report’s concern about being “swallowed by compute” is easing at the margin.

  Second, the absolute level is still clearly below mature software, and the accounting is fragile. Overall gross margin of about 40% is far below classic SaaS at 70-80%; the report repeatedly emphasizes that there is no public 10-K/10-Q, and there is no public-company-grade disclosure for gross margin, operating cash flow, capitalization/expensing policies, or SBC dilution. Reuters also noted that accounting treatment for cloud revenue sharing among model companies is not consistent. So although the direction of improvement is credible, the precise figures should not be treated as audited conclusions. The report’s judgment that “the business is understandable, but the economics are not transparent” still holds.

  Third, scaling also magnifies absolute costs. When revenue doubles every six weeks, compute spending also doubles; even with a 70% inference margin, the absolute cost numbers remain enormous. This is the core concern behind the report’s statement that “compute is strategic, capital-heavy, and may swallow the profit pool.” Better unit economics do not mean the overall company is already profitable.

  As for “where the money it earns goes,” the answer is that there is basically no “earned money” available for distribution; the money is mainly raised capital, and it is all reinvested in the main path. The report and first-hand information are consistent: cash goes into research and products, compute and channels, and small acquisitions close to the main path (the May 2026 acquisition of Stainless to strengthen SDK/MCP); there has been large historical cash burn, with management once expecting not to stop burning cash until 2027; there are no dividends and no buybacks. The largest capital destination is the compute commitment itself: more than 100 billion dollars to AWS over ten years. This is a machine that puts all fuel into the expansion engine, not a cash cow.

  Conclusion: on unit economics, Anthropic’s story is “excellent direction, still shallow absolute level.” Gross margin has moved from -94% to about 40%, inference margin to about 70%, and scale effects are genuinely improving incremental economics, which is more positive than the picture when the report was written. But it is still far below top-tier software, accounting is opaque, absolute costs expand along with scale, the company is not yet profitable overall, and all money goes into R&D and compute with zero shareholder returns. This dimension deserves a lower-middle-to-middle score recognizing the improvement trend, but beautiful run-rate revenue must not make unit economics look like a verified top-tier business. It is getting better; it has not yet arrived.

  Jun 11, 2026
• QFor it to rise fivefold in ten years, what conditions must all hold at the same time? Are those conditions realistic? What expectations are implied by today’s price?2/10

  A

  For Anthropic to rise fivefold from today’s valuation over ten years, a series of difficult conditions must hold at the same time, and today’s private-market valuation has already priced in a substantial portion of those victories upfront. This is the core of the report’s “good company, bad price” judgment. On the latest metrics, the price is even tighter than it was when the report was written. This question must first make clear what today’s price implies, then revisit whether the conditions are realistic.

  First, update the valuation anchor to the current point in time. The report was written on 2026-05-19, using the confirmed 350-380 billion-dollar range and treating 900 billion dollars as an unconfirmed FT rumor. But the facts have since landed: at the end of May 2026, Anthropic completed Series H, raising 65 billion dollars at a post-money valuation of about 965 billion dollars, while officially disclosing that run-rate had crossed 47 billion dollars; on June 1, 2026 it confidentially filed an S-1 with the SEC. In other words, the “900 billion dollars” the report prudently excluded is no longer rumor, but confirmed reality. That moves the report’s conclusion from “expensive” to “more expensive.”

  So what does today’s valuation of about 965 billion dollars imply? A sober breakdown:

  • The multiple itself. 965 billion / 47 billion ≈ 20.5x run-rate revenue. For comparison, OpenAI’s financing at the end of March 2026 valued it at about 852 billion dollars against about 25 billion dollars of annualized revenue (about 34x). Anthropic’s revenue multiple is actually lower than OpenAI’s, but that depends on the “47 billion run-rate is real and sustainable” assumption. Run-rate is an extrapolation from a recent period, and the report specifically warns that it is unusually sensitive to timing/accounting definitions and that the two companies use inconsistent accounting for cloud revenue sharing, so simple multiple comparisons are not rigorous.
  • The arithmetic of a ten-year fivefold increase. Moving from ~965 billion to ~4.8 trillion dollars requires about 17%/year compound growth in equity value, and only works if there is no further large dilution. Yet it has just completed two huge G and H financing rounds in the past year. In other words, the market requires not only continued high growth, but also conversion of growth into real profit and an end to trading equity for compute.

  The checklist of conditions required for a fivefold outcome, consistent with the report’s “key assumptions,” looks like this:

  1. Revenue keeps growing fast and the base does not collapse. This is the most likely condition to hold. The growth engine is extremely strong (see the revenue-doubling question), but a 10x/year rate must converge, and the key is whether “sufficiently high” growth can continue on a trillion-level base.
  2. Run-rate turns into real, sustainable free cash flow. This is the most uncertain condition. Gross margin has improved from -94% to about 40%, inference margin is about 70%, and the 2028 target is 77%; the direction is right, but the company remains loss-making, management had once expected cash burn not to stop until 2027, and compute costs scale with the business.
  3. Compute/channel revenue sharing does not permanently suppress mature-period margin. Anthropic has committed more than 100 billion dollars to AWS over ten years, so the risk that cloud vendors take part of the profit pool is structural.
  4. The moat hardens into an institutional barrier (MCP becomes a standard, workflows are truly locked in), rather than being eroded by model commoditization. The report judges the long-term moat “unproven.”
  5. Regulatory/copyright/government conflicts do not become a persistent cash-flow hit. The 1.5 billion-dollar copyright settlement has not yet received final approval, and the Pentagon dispute was described by the company as potentially affecting several billion dollars of revenue.

  For these five conditions to hold at the same time is difficult. If any one of them, especially condition 2 or 4, breaks, multiple compression alone could cause permanent capital loss. The report’s intrinsic value range (reasonable 120-220 billion, optimistic 250-360 billion dollars) is far below today’s roughly 965 billion, meaning a buyer today is betting at a price “well above the optimistic case.” The report’s own threshold for “clearly overvalued” was above 300 billion dollars, and today’s price is more than three times that line.

  Conclusion: a ten-year fivefold outcome is not impossible, but it requires five things to be delivered together: high growth, real positive free cash flow, margin not swallowed by compute, an institutionalized moat, and a regulatory soft landing. None is easy. Today’s valuation of about 965 billion dollars, about 20.5x run-rate, already implies that Anthropic becomes infrastructure-level platform and captures most of the victory. The margin of safety is not merely insufficient; it is negative. This dimension (valuation and implied expectations) should clearly receive a low score. That has nothing to do with company quality; it is simply that price has pulled too much future into the present, exactly the kind of situation where Baillie Gifford would press “do not buy, only watch.”

  Jun 11, 2026
• QWhy has the market not realized all of this yet? Is it because investors do not understand it, look down on it, or fail to look far enough? What will become the “narrative inflection point”?3/10

  A

  For Anthropic, this question reverses: the market has not “failed to realize it”; the market has already fully, even excessively, realized its greatness. It is one of the most sought-after private assets in the world, and pricing has pulled future victories far forward. So Baillie Gifford’s usual question, “why has the market not realized it yet,” barely applies here. The more accurate framing is that the market understands the growth, but may not have fully priced “price risk, accounting fragility, and governance tilt”. The narrative inflection point is therefore two-sided: there are catalysts that could keep the story moving up, and events that could puncture the premium.

  First, prove that the market does not fail to understand it or look down on it. An undervalued target does not attract capital like this: Series G in February 2026 raised 30 billion dollars at a 380 billion valuation; in April, Google announced an investment of up to 40 billion dollars, lifting the valuation to 350 billion and locking in 5GW of TPU capacity; by the end of May, Series H raised 65 billion at about a 965 billion-dollar valuation, surpassing OpenAI to become the most valuable AI startup. Top institutions, including many sovereign wealth funds and growth funds, are fighting to get in. So the three classic forms of underappreciation, “not understood, looked down on, or not looked far enough,” basically do not apply to Anthropic. It is the kind of asset people are “looking too far ahead and believing too fully.”

  So what might the market “not fully realize”? Not the good parts, but three price-related risks covered by the optimistic narrative:

  First, the fragility of run-rate metrics is being downplayed. The market uses about 47 billion dollars of run-rate to justify the valuation, but the report repeatedly warns that run-rate is an extrapolation from a recent period, unusually sensitive to timing/accounting definitions, and Reuters notes that model companies do not use consistent accounting for cloud revenue sharing. Simple multiple comparisons are not rigorous. This uncertainty is easy to ignore in a frenzy.

  Second, the profit pool may not sit in the model layer. Anthropic has committed more than 100 billion dollars to AWS over ten years, so a meaningful portion of the economics goes to cloud vendors. If models are commoditized, value will move down into data and application layers. The market is pricing a “platform ruler,” but that identity has not yet been proved.

  Third, governance tilts toward public benefit rather than pure shareholder orientation. LTBT-appointed directors already hold a majority of board seats, and the report emphasizes that this is a real cost for investors who “only care about per-share intrinsic value.” During a growth chase, this point is often treated as background noise.

  Now answer the implicit premise: “what will become the narrative inflection point.” I divide it into upside catalysts and downside triggers, because for an already richly priced asset, the inflection point is more likely to “confirm/disconfirm the premium” than to “discover value”:

  Upside catalysts (letting fundamentals catch up with the high valuation, or pushing it higher):

  • IPO and disclosure. Anthropic confidentially filed an S-1 with the SEC on June 1, 2026; once a public S-1 appears, the market will for the first time see real gross margin, operating cash flow, and SBC dilution. The report correctly lists “whether IPO documents or more transparent disclosure appear” as a key tracking item. If financial quality is better than expected, it is a positive inflection point; if it reveals ugly profit/dilution details, it reverses. This is the most certain and most powerful inflection point.
  • Evidence of free cash flow turning positive. If the company clearly crosses into FCF positivity around 2027-2028, alongside gross margin that has risen from -94% to about 40%, inference margin around 70%, and the 77% 2028 target, the discount for “depending on future cash flow” would narrow.
  • MCP truly becomes the de facto standard for the agent era, and customers do not migrate even when alternative models are available.

  Downside triggers (puncturing the premium and causing multiple compression):

  • Large-customer growth slows materially, or cash burn worsens again.
  • Copyright/government conflicts escalate into persistent cash-flow hits: the 1.5 billion-dollar settlement has not yet received final approval, and the Pentagon blacklist was not halted while the company described an impact of several billion dollars of revenue.
  • Model capability is matched, and token price wars degrade revenue quality.
  • Market sentiment shifts from “platform ruler” back to “excellent but replaceable model company.” The report states that this narrative shift alone could pull valuation back to the 50-150 billion-dollar range, implying permanent capital loss of 60-85% relative to the current price.

  Conclusion: the honest answer for Anthropic is that the market has already realized, and even over-priced, the story. The Baillie Gifford-style “perception gap I can exploit” runs in the opposite direction here. This is not an overlooked gem, but a heavily chased, high-priced asset that has pulled a large amount of future victory into today. The only real perception gap may sit in the corners covered by frenzy: price risk, accounting fragility, and governance cost. The most likely narrative inflection points are public IPO disclosure (S-1 already filed, highest impact), whether FCF turns positive, and regulatory/competitive events; for a currently richly priced asset, downside triggers are more destructive than upside catalysts are additive. This dimension (perception gap/market pricing review) should receive a low score: the direction is negative, price has overdrawn the future, and that is the basis for the report’s view that “the most rational posture is not to chase, but to wait.”

  Jun 11, 2026