Anthropic Enters IPO Preparation with $650M Funding and S-1 Filing

TL;DR

Anthropic being traded on the same chart as other assets recently is not an isolated event, but a set of signals: a post-money valuation in the hundreds of billions, confidential S-1 filing, rapidly growing revenue run rate, and rumors surrounding Claude 5.

For investors, the meaning of this set of signals is straightforward. AI Frontier Lab is no longer just proving its value through papers, model rankings, and product reputation—it’s now speaking in terms the public markets can understand. Model capabilities, enterprise adoption, revenue quality, compute costs, and risk disclosures are being integrated into a single valuation framework.

There is currently no official announcement, product page, or model card from Anthropic confirming that Claude Fable 5 shares its underlying architecture with Mythos, features enhanced safety guardrails, or improved capabilities for long contexts and complex tasks. These claims should still be regarded as rumors or market expectations. What truly deserves discussion is not what Fable 5 has already demonstrated, but why an unconfirmed new model is being included in Anthropic’s IPO narrative ahead of official validation.

Anthropic’s timeline has been sufficiently packed. On May 28, the company announced the completion of a $65 billion Series H funding round, valuing it at $965 billion post-money, and stated that its run-rate revenue had already exceeded $47 billion earlier that month. On June 1, Anthropic confirmed it had confidentially submitted a draft Form S-1 to the SEC for an IPO, with the number of shares and price range yet to be determined; the listing remains subject to SEC review, market conditions, and other factors.

This changes Anthropic’s position in the market. It is no longer just the “safety-focused model company” outside of OpenAI, but a leading candidate for a super-large AI platform preparing for the public market. Private markets can pay for future possibilities, but public markets also pay for the future—only they require companies to break down those possibilities into more verifiable metrics.

These metrics include whether revenue sources are stable, whether customers are concentrated, whether hashing power costs are controllable, whether model leadership can be sustained, and whether regulatory risks are disclosable and manageable. For companies with frontier models, this transition is more difficult than for traditional software companies.

When traditional SaaS companies go public, investors typically examine ARR, net retention rate, gross margin, sales efficiency, and customer composition. Frontier model companies must also answer these same questions, but additionally face challenges around training and inference costs, model iteration speed, security incidents, cloud provider dependency, and chip cycles. The stronger the model, the greater the revenue potential—but also the heavier the costs and regulatory variables.

This is also what makes Anthropic unique at this moment. Its high valuation cannot be sustained solely by the claim that “Claude is smarter,” but requires a more complete narrative: continuous improvement in model capabilities, enterprise customers willing to pay, sufficient revenue traction, a security positioning that enables entry into high-value use cases, and an still-open window in the capital markets. The rumors surrounding Claude 5 have been amplified because it appears to be the next piece of this puzzle.

If Anthropic had simply rolled out a new model in a routine manner, the market likely wouldn’t have been this excited. The rumors surrounding Claude 5 were amplified precisely because they coincided with the critical IPO phase—funding, valuation, and the S-1 filing—layered with major model disclosures, perfectly crafting the narrative that capital markets love most: a company proving, just before going public, that it remains firmly at the forefront of capability advancement.

A prediction market has already opened betting on whether Claude 5 will be made publicly available before June 30, 2026. The current odds reflect the implied probability priced in by traders. Although public information has not yet fully confirmed the previously rumored timeline, it is clear that prediction markets are already pricing in a near-term release of Claude 5—and the pricing is not low.

This expectation carries inherent information. The market directly translates the “product cadence” into a “valuation narrative”: if Anthropic can indeed consistently release stronger models, its nearly trillion-dollar post-money valuation’s high revenue run rate will be interpreted as a natural outcome of rapid platform expansion; conversely, if the model release pace slows significantly, the nearly trillion-dollar valuation would have to rely more heavily on the quality and certainty of its current revenue to sustain itself.

This approach is not unfamiliar. Before going public, consumer internet companies emphasize user growth and retention, cloud companies highlight enterprise customers and net expansion rate, and semiconductor companies focus on orders and production capacity. AI model companies have not yet established a fully mature public market template, so the release of a model itself has become a visible signal—it serves as both a product update and a demonstration of capability, influencing not only developers and enterprise customers but also investors’ expectations for the next phase of revenue growth.

However, the impact of model releases on valuation is not linear. A stronger model can lead to higher API usage, greater enterprise contracts, and stronger customer retention, but it may also result in higher inference costs, more complex security reviews, and heavier infrastructure investments. The public market will ultimately not only ask, “Is it the strongest?” but also, “How much compute is consumed per dollar earned?” “Can gross margins improve?” and “Will security boundaries limit commercialization speed?”

The more differentiated aspect of Anthropic’s current narrative is not “just another chat model,” but the controlled frontier capabilities represented by Mythos and Project Glasswing.

According to Anthropic’s official disclosure, Claude Mythos Preview is a general-purpose, unreleased state-of-the-art model that will not be publicly released. Project Glasswing is designed for defensive security work, with controlled access granted to partners for use cases such as critical software security, zero-day vulnerability discovery, and remediation. Anthropic also disclosed that the project has already identified numerous zero-day vulnerabilities and has committed to providing up to $100 million in usage credits and $4 million in open-source security donations.

This gives Anthropic a valuation narrative distinct from that of a consumer-facing chatbot. It can position itself as a foundational model provider for complex tasks, high-value enterprise workflows, and safety-critical applications. For public markets, this is easier to align with enterprise budgets and easier to explain why businesses are willing to pay a premium for more reliable and secure models.

If there is indeed a technical connection between Fable 5 and Mythos in the future, and it is designed with stricter safeguards for a broader user base, it would create a narrative path: cutting-edge capabilities are first validated in controlled environments before being partially productized in a safer form. This path aligns with Anthropic’s long-standing emphasis on safety and meets enterprise customers’ demand for controllable AI.

But the same thing could also subject Anthropic to greater regulatory pressure. Cybersecurity capabilities have dual-use properties—models capable of identifying and patching vulnerabilities could also be misused in attack chains. Mythos’s controlled release has already demonstrated that such capabilities cannot be simply fully unleashed. If future, more general-purpose models are perceived by the market as “downscaling Mythos’s capabilities,” the company will need to clearly articulate its safety safeguards, access restrictions, abuse monitoring, and boundaries of responsibility.

These disclosures will be included in the S-1 risk factors. Public market investors are not only concerned with how powerful the model is, but also whether such strength leads to additional regulatory costs, national security reviews, reputational risk, and potential liability. For an AI company with a post-money valuation nearing $1 trillion, a major security incident may not merely be a product failure, but a systemic factor affecting the timing of the IPO and valuation multiples.

For Anthropic, what is currently most overvalued by the market is not the model's capabilities, but the relationship between those capabilities and its valuation.

Discussions around Fable 5, Claude 5, and even Mythos do indeed help boost market sentiment. Stronger models mean greater customer attention, stronger developer interest, and allow Anthropic to maintain its presence in the competition for cutting-edge models. However, these factors remain fundamentally growth expectations—they can explain why the market is willing to trade in anticipation of the future, but they are insufficient on their own to validate a valuation of nearly a trillion dollars.

Once truly entering the public market, investors’ concerns will quickly become specific. Of the $47 billion+ annualized revenue, how much will ultimately convert into sustainable, auditable income? Is enterprise customer growth driven by long-term deployments or phased trials? What proportion of revenue comes from top customers? What roles do strategic partners and cloud channels like Amazon and Alphabet play? More importantly, as model usage surges, can training and inference costs continue to decline, thereby supporting improved profit margins?

These metrics ultimately determine not just the rate of growth, but how the market defines Anthropic.

If the company can consistently reduce unit costs, increase customer retention, and build a broader software ecosystem around its model, investors are more likely to view it as a next-generation AI software platform. If revenue growth is consistently accompanied by massive compute investments and ongoing capital expenditures, the market is more likely to classify it as a high-growth, high-consumption AI infrastructure company. Both narratives can support high valuations, but they come with different valuation multiples and levels of risk tolerance.

The existence of OpenAI makes this comparison more apparent, but the two are not necessarily engaged in a simple zero-sum competition. OpenAI has stronger consumer access and ecosystem influence, while Anthropic has built a more distinct narrative around enterprise, security, and governance. What the public markets will truly compare in the future is not who releases a model first, but who can more consistently translate model capabilities into revenue growth and explain the future to investors with lower uncertainty.