Artemis Analytics Compares Anthropic to AWS, Forecasts High Margins and ARR Growth

Author: Artemis Analytics

Compiled by Deep潮 TechFlow

Shenchao Insight: While the market is still debating whether open-source models will turn cutting-edge AI into a commodity, Artemis offers a counterintuitive perspective: Anthropic’s true moat isn’t the model itself, but its enterprise-level lock-in capabilities akin to AWS. This analysis breaks down how to interpret Anthropic’s S-1 through three dimensions—compute power, gross margin, and channel structure—providing valuable insights for investors tracking how AI narratives translate into valuations for crypto and tech stocks.

In 2015, skeptics viewed AWS as a highly unprofitable cost center within Amazon, offering a commoditized product similar to electricity, with no clear understanding of the size of the cloud market. Worse still, the market feared that competitors like Google and Microsoft would undercut AWS’s pricing.

Fast-forward to 2026, Amazon has become the undisputed leader in cloud computing, building a powerful business far beyond its original S3 offering through a comprehensive product portfolio, economies of scale, and a thriving developer ecosystem.

By 2026, Anthropic faces similar skepticism: despite having raised and spent billions of dollars, their cutting-edge models may still be commoditized by open-source competitors.

We believe that, with its exceptional team, Anthropic will not only continue to push the boundaries of cutting-edge models but also build a highly cohesive suite of all-in-one AI products that enable enterprises to easily build, scale, and maintain AI systems for every vertical industry worldwide. These products include, but are not limited to, state-of-the-art domain-specific models (such as drug discovery models) and tools designed to ensure AI safety, compliance, and performance optimization.

Anthropic will create enterprise-grade lock-in, with long-term gross margins expected to exceed 60% or even higher, building a business as highly profitable as AWS, with ARR surpassing AWS by 2027.

Complete financial model available at: https://www.artemis.ai/anthropic-thesis

At its core, Anthropic boils down to three factors: 1) Whether it can attract top-tier researchers to develop cutting-edge models; 2) Whether it can acquire and secure computing power at low cost ($10 million to $15 million per megawatt); and 3) Whether it can charge a premium on tokens via its API (e.g., $50 million in annual recurring revenue per megawatt).

From key metrics:

• Demand side: We estimate monthly net new ARR to be between $10 billion and $15 billion. Even as token prices decline, token usage continues to rise, and enterprise demand for cutting-edge models and agent deployments has become significantly de-risked.
• Supply side: Total locked computing power (GW). According to reports, Anthropic has secured 15 to 16 GW of computing power by 2030, and at least 20 GW of training computing power is required to support a $1 trillion ARR—we believe this is achievable by 2030 (according to reports, OpenAI will secure at least 30 GW of total computing power by 2030).
• Gross margin: The difference between the ARR (in millions of dollars) Anthropic can charge per megawatt and the cost of computing power per megawatt (in millions of dollars) is key to determining Anthropic’s long-term profitability and whether it’s a good business.

In addition:

• Anthropic is a business focused on enterprise APIs. 90% of its ARR comes from APIs. The $10 trillion endpoint depends on Anthropic’s ability to sell models directly and through AWS Bedrock, Gemini Enterprise Agent Platform, and Microsoft Foundry. The consumer business is insignificant.
• Look at net ARR, not gross ARR. Investors need to deduct the 15–20% revenue share paid to AWS, Gemini, and Microsoft, and exclude revenue from Meta and from Chinese AI labs distilling Anthropic models.

Many things could go right.

In our base case, Anthropic could secure over 30.8 GW of computing power by 2030, supporting $1 trillion in ARR, with a total cost of $18 million per megawatt of computing power and $50 million in ARR per megawatt of inference.

This implies a steady-state gross profit margin of 66%, an EBIT profit margin of 30%, and total training and R&D expenses declining to approximately 25% of revenue.

In this state, Anthropic is highly profitable and growing rapidly as its agents continue to penetrate over 300 million businesses worldwide.

Sort by probability:

• High risk: Codex and OpenAI Astra are capturing market share from Anthropic. We’ve observed from engineers around us that they are beginning to shift from Claude Code to Codex. After testing Astra and Fable last weekend, we found the gap has narrowed significantly—customers may now migrate to OpenAI. Here are some direct customer quotes: Head of Blockchain Data at Artemis: “From my personal experience, I’ve been using Codex exclusively and prefer its communication style. Some people treat them as complementary—using Claude Code for planning and Codex for execution—but I see them as substitutes and use only Codex. I was previously a heavy user of Claude Code in my terminal, but now I prefer Codex within the ChatGPT app.”
• Artemis FinTech Analyst: "OpenAI’s Codex is incredibly powerful—it can visualize what the agent is doing, clearly designed for programming. Claude Code offers a solid first-stage experience in the CLI, but Codex’s GUI is now far superior for observing agent behavior. Claude Code created the first generation of programming experience in the CLI, but I believe the second generation will emerge in applications with a GUI."
• A PM from a $3 billion company: One of our friends is a staunch supporter of Claude Code and Claude Harness, and last weekend shared this: “Wow, bro, I have to say, Astra is incredibly powerful.”
• High risk: The switching cost between models is low. In Q2 2026, 25% of gross revenue flowed through third-party channels such as AWS Bedrock and the Gemini Enterprise Agent Platform. Through AWS Bedrock, APIs provided by Anthropic can be easily replaced by OpenAI’s Astra and other open-source weighted models. If Anthropic does not offer state-of-the-art models, enterprises can switch easily, potentially leading to significant revenue displacement.
• Medium risk: Harnesses like Grokbot and Instinct do not require Anthropic. Grokbot is built on x.ai’s cutting-edge models and directly challenges Claude Cowork and Claude’s harnesses. If Instinct and new AI applications are built on open-source or other cutting-edge models, this could force Anthropic to move up the stack by acquiring harnesses or applications, or be compelled to innovate similarly with products like Claude Code, Claude Design, and Claude Cowork.
• Lower risk: Computing power cannot be locked in. Anthropic has already locked in nearly 15 GW of computing power and requires an additional 15 GW to support $1 trillion in ARR. An IPO exceeding $100 billion would be sufficient for Anthropic to lock in computing capacity. That said, Amazon, NVIDIA, Google, and SpaceX are simultaneously suppliers, shareholders, and competitors (Amazon has Titan, SpaceX has Grokbot, and NVIDIA is moving up the stack through its partnership with Hugging Face). We believe the likelihood of this scenario occurring is extremely low.
• Lower risk: Anthropic has stopped releasing frontier models—either researchers have become so wealthy from IPOs that they no longer wish to continue, or they’ve hit technical bottlenecks. Currently, incentives are perfectly aligned; but what if the researchers’ equity is worth $150 million to $300 million? A $5 million to $10 million equity stake at a $10 billion to $20 billion valuation would be worth hundreds of millions at a $3 trillion valuation. We believe it is extremely unlikely that Anthropic will stop releasing its most advanced models.

Even among high-growth software and AI comparable companies, Anthropic appears undervalued at a $2 trillion valuation based on EV/ARR and EV/NTM revenue, and even at a $3 trillion valuation, it remains reasonable.

We believe Anthropic will go public in mid-to-late October at a $2 trillion valuation, with retail and institutional demand pushing it to $3 to $4 trillion.

To be honest, Anthropic might be worth buying regardless.

• Anthropic's reported net ARR of $65 billion as of July 2026 represents a 15-fold year-over-year growth, and our estimated ARR of $90 billion by the end of September 2026 also reflects a 15-fold year-over-year increase—this is generational growth.
• In addition, our more conservative forecast—compared to SemiAnalysis’s $300 billion ARR for 2027—is $125 billion ARR by the end of 2026 and $275 billion ARR by the end of 2027, representing 20 times the expected ARR for 2026 and 7.2 times the expected ARR for 2027.
• It’s hard not to go long on $ANTHR, given it’s one of the fastest-growing public companies of all time and a clear leader in enterprise AI.

How do you value a company that might develop AGI and sell it to over 300 million enterprises worldwide—each eager to replace or augment white-collar workers? A rough calculation: 300 million enterprises × one white-collar worker replaced per company at $120 ACV equals a global potential AI spend of $36 trillion. What if Anthropic captures 20% of that ($7.2 trillion in ARR)? Or 30% ($10.8 trillion in ARR)?

Ramp estimates that 56% of U.S. businesses have spending on AI, but the median is only $12 per month. We believe this reflects businesses paying for individual subscriptions to ChatGPT or Claude.

Our long-term view is that agents—rather than chatting with agents via Claude or ChatGPT—are the key to accelerating global and broader AI spending, yet current agent adoption in enterprises remains below 1%.

In short, our base case of $1 trillion in ARR by 2030 may still be too conservative.

If superintelligence arrives, could Anthropic use its own models to build models, significantly reducing research costs and expanding profit margins? If superintelligence is only accessible through Anthropic, would willingness to pay surge, making $50 million in ARR per megawatt potentially too conservative?

Time will tell—these cutting-edge models are drawing ever closer to superintelligence.

Finally, I recalled these words from Anthropic’s Chief Financial Officer: “Most humans think in linear, incremental ways. I’ve been at [Anthropic] for two years. This is the mindset I had to break for myself. Stop thinking linearly and start thinking exponentially.”

Congratulations to Anthropic, looking forward to the S-1 and exponential growth on the path to superintelligence.

This is the personal opinion of Jon Ma, published solely for informational and educational purposes. It does not constitute investment, financial, legal, or tax advice, nor does it constitute an offer or solicitation to buy, sell, or hold any securities (including any equity in Anthropic).

Artemis Analytics is not a registered investment advisor or broker-dealer. The author may hold a position in Anthropic through an SPV. Please conduct your own research and consult a licensed professional before making any investment decisions.

Anthropic is a private company. Its shares are not registered and are not publicly traded; equity in the secondary market lacks liquidity and can only be sold through restricted channels. The third-party "marks" or secondary market prices referenced in this document (including any implied valuations) may not reflect prices at which any actual transaction could occur. Any references to a potential future initial public offering, its timing, or pricing are speculative; there is no assurance that an IPO will occur or at what terms.

Conflict of interest. The authors and Artemis Analytics directly and/or through pooled investment vehicles hold economic interests in Anthropic and stand to benefit if Anthropic’s valuation increases. We may buy or sell these interests at any time without updating this page. Artemis Analytics is also a customer of Anthropic, and our products are partially built on Anthropic’s models; Artemis employees are cited in this article as users of Anthropic’s products. Readers should assume the authors are not neutral observers.

Forward-looking statements and estimates. This page contains estimates, projections, scenarios, and price targets (including ARR, gross margin, computing capacity, and 2030 valuation data), which represent the author’s own opinions or sources from third parties, including internal estimates from Artemis and published research from institutions such as SemiAnalysis. These data are not financial results reported by Anthropic, have not been verified or endorsed by Anthropic, and are inherently uncertain. Actual results may differ materially. Bull, base, and bear cases are illustrative scenarios and not predictions or guarantees.

Information from third-party sources is considered reliable but has not been independently verified, and we make no representations regarding its accuracy or completeness. All investments involve risk, including the possible loss of the entire investment amount; past performance is not indicative of future results. Before making any investment decision, please conduct your own research and consult a licensed financial advisor, lawyer, or tax professional. Anthropic is not affiliated with Artemis Analytics and has not reviewed or approved this content.