The $47 Billion Question: Anthropic’s Daniela Amodei Bets Big on AI’s ROI as IPO Looms

San Francisco has seen its share of speculative frenzies, but the latest twist in the city’s real estate market reveals everything about the current state of AI hype. Several Bay Area homeowners now list properties with an unusual offer: they’ll trade their home for a piece of Anthropic stock [3]. Not cash. Not Bitcoin. Equity in a company that, until recently, burned through capital at a rate that would make most venture capitalists blanch.

This is the backdrop against which Anthropic president Daniela Amodei is preparing to take the company public. And she’s not in a mood to entertain skeptics.

On Thursday, the company disclosed that its annualized revenue crossed $47 billion in May—up from roughly $9 billion at the end of 2025 [1]. Amodei essentially told the market to get used to the idea that AI’s returns are real, measurable, and accelerating. The numbers are staggering: a 5.2x revenue expansion in roughly five months. But the real story isn’t just top-line growth. It’s about what Anthropic is doing internally to justify that valuation, and whether the broader AI ecosystem can sustain the returns investors now demand.

The Self-Writing Codebase: Claude Eats Its Own Dog Food

The most revealing data point from Anthropic’s pre-IPO disclosure isn’t the revenue figure—it’s that the company has effectively become its own most important customer in a way few enterprise software firms have managed.

According to a report shared on the same day as the revenue announcement, more than 80% of the code merged into Anthropic’s production codebase in May wasn’t authored by humans, but by its own AI model, Claude [2]. This isn’t an experimental side project or a toy demo. This is the code that runs the company’s core infrastructure—the code that powers the very model sold to enterprises. The transformation triggered an 8x increase in the volume of code shipped per engineer [2].

Let that sink in. Anthropic uses Claude to build Claude. CEO Dario Amodei had been telegraphing this shift for months, but the sheer scale—80% of production code—represents a milestone even the most bullish AI optimists might have considered years away. The industry has started calling this “recursive self-improvement” [2], a term that sounds like science fiction until you see the deployment numbers.

The implications for the IPO narrative are profound. When investors ask whether AI can deliver meaningful productivity gains, Anthropic can point to its own engineering org as a controlled experiment. The company isn’t just selling shovels in a gold rush; it uses those shovels to dig its own mine, and the yield per engineer increased by an order of magnitude. If this pattern holds across the enterprise landscape, the ROI question starts to answer itself.

The Revenue Trajectory: Hockey Stick or Bubble?

Still, $47 billion in annualized revenue demands scrutiny. The jump from $9 billion at the end of 2025 to $47 billion in May 2026 represents a growth rate extraordinary for any company in any sector [1]. For context, that pace would make even the most aggressive SaaS companies of the last decade look pedestrian. Zoom, during its pandemic-fueled hypergrowth, never came close to this multiple in such a compressed timeframe.

The question analysts wrestle with—and Amodei clearly tries to preempt—is whether this trajectory is sustainable. The company heads into an IPO at a moment when the broader tech market increasingly bifurcates between AI winners and everyone else. NVIDIA, which filed its 10-Q with the SEC on May 20, 2026 [5], continues to ride the GPU demand wave, but the infrastructure layer is becoming commoditized. The real value capture, investors start to realize, happens at the application and model layer.

Anthropic bets that Claude’s enterprise adoption will continue to accelerate as companies move beyond experimentation into production deployment. The company’s internal metrics suggest the model is not just a cost center but a force multiplier for engineering productivity. If Anthropic can convince public markets that its own 80% code-automation figure is replicable across its customer base, the valuation narrative becomes much easier to sell.

But a tension exists that the company’s S-1 will need to address. The same models driving Anthropic’s revenue are also, in a very real sense, competing with the human labor that generates traditional enterprise software revenue. If Claude can write 80% of Anthropic’s own production code, what happens to the addressable market for developer tools? The company essentially eats its own tail. While philosophically elegant, this creates a tricky dynamic for long-term revenue modeling.

The Infrastructure Arms Race: NVIDIA and Microsoft Enter the Chat

Anthropic’s IPO preparations unfold against a rapidly shifting infrastructure landscape that will directly impact the company’s cost structure and competitive positioning. Just days before the revenue disclosure, NVIDIA and Microsoft announced a partnership to build a unified stack for agentic AI deployment spanning Windows devices, Azure cloud, and local environments [4].

The announcement, made at Microsoft Build, signals that the infrastructure layer is consolidating around a few key players. Jensen Huang and Satya Nadella effectively bet that the next phase of AI adoption—agentic AI, where models don’t just generate text but take actions across systems—requires a vertically integrated stack combining fast hardware, secure runtimes, responsive data layers, and models tuned for long-running reasoning [4].

This directly affects Anthropic’s IPO story because it highlights the dependency risk every AI company faces. Anthropic builds world-class models, but those models run on NVIDIA GPUs, and increasingly, they’ll deploy through Microsoft’s cloud infrastructure. The NVIDIA-Microsoft partnership [4] creates a powerful duopoly at the infrastructure layer that could squeeze margins for model providers over time.

For Anthropic, the calculus is straightforward: the company must demonstrate that its model quality and safety advantages are durable enough to command premium pricing, even as underlying compute costs are dictated by partners who also invest in competing models. Microsoft, after all, has a deep relationship with OpenAI, and NVIDIA has no incentive to favor one model provider over another.

The Real Estate Signal and the Talent War

The Wired report about Bay Area homeowners accepting Anthropic stock in lieu of cash for real estate transactions [3] is more than a quirky anecdote—it signals how the market prices the company’s future. In a region where housing inventory is constrained and prices remain astronomical, the willingness of sellers to accept equity as payment suggests deep conviction that Anthropic’s stock will appreciate significantly post-IPO.

This dynamic also speaks to the talent war Anthropic wages. The company was founded in 2021 by former OpenAI members, including siblings Daniela Amodei and Dario Amodei [1], and has positioned itself as the safety-first alternative in the AI arms race. That narrative has attracted top researchers and engineers who could command massive salaries elsewhere but are willing to take equity-heavy compensation packages in exchange for a shot at a life-changing liquidity event.

The real estate phenomenon is essentially a secondary market signal that employees and early investors believe the company’s valuation will continue to climb. It also reminds us that Anthropic, despite its revenue growth, is still a privately held public benefit corporation [1] that has yet to face the full scrutiny of public markets. The IPO will force the company to open its books and answer questions private investors have been willing to defer.

The Safety Paradox and the IPO Narrative

One of the more interesting subtexts of Anthropic’s pre-IPO positioning is how the company reconciles its safety-focused mission with the relentless growth demands of public markets. Anthropic was founded with an explicit focus on AI safety, and the company has consistently argued that responsible development is not a constraint on innovation but a prerequisite for long-term value creation.

That argument is about to face a very public stress test. Public market investors are not known for their patience with mission-driven trade-offs that might slow revenue growth. If Anthropic’s safety protocols require slower deployment cycles or more conservative model releases, quarterly earnings calls will become a battleground between the company’s founding principles and the market’s demand for acceleration.

The internal data about Claude writing 80% of production code [2] cuts both ways on this front. On one hand, it demonstrates that safety and speed are not necessarily in conflict—Anthropic moves faster than ever while presumably maintaining its safety standards. On the other hand, it raises questions about oversight. If the model writes the code that runs the model, who audits the auditor? The company will need to convince regulators and investors that its recursive self-improvement loop [2] includes robust guardrails.

The Developer Ecosystem: Winners, Losers, and the Open-Source Question

Anthropic’s IPO also arrives at a moment when the developer tooling ecosystem is being reshaped by the very models the company sells. The VentureBeat report notes that the 80% code-automation figure triggered an 8x increase in the volume of code shipped per engineer [2]. For Anthropic’s own team, this is a productivity miracle. For the broader developer tools market, it’s an existential threat.

If Claude can write production-grade code at this scale, what happens to the market for traditional IDE plugins, code review tools, and documentation platforms? These tools will need to either integrate deeply with AI models or become obsolete. We’re already seeing this play out in the open-source community, where models tracked by Daily Neural Digest’s HuggingFace data see massive adoption. The whisper-large-v3-turbo model, for instance, has been downloaded 8,625,103 times, while the gpt-oss-20b and gpt-oss-120b models have accumulated 7,780,249 and 4,549,787 downloads respectively.

The open-source ecosystem is clearly thriving, but it’s also fragmenting. NVIDIA’s NeMo framework, which has 16,885 stars and 3,357 forks on GitHub, represents one approach to building scalable generative AI systems. Anthropic’s Claude represents another. The tension between open-source flexibility and proprietary performance will define the post-IPO era.

For developers, the calculus is becoming more complex. Tools like the OpenAI Downtime Monitor, which tracks API uptime and latencies for various LLM providers, are becoming essential infrastructure as companies increasingly rely on model APIs for production workloads. The era of treating AI models as experimental add-ons is over; they are now core infrastructure, and reliability requirements are correspondingly severe.

The Verdict: What the Mainstream Media Is Missing

Mainstream coverage of Anthropic’s IPO has focused on the obvious story: a hot AI company with massive revenue growth is going public, and the founders are confident. That’s not wrong, but it misses the deeper structural dynamics at play.

What’s actually happening is that Anthropic attempts to prove that AI can generate returns that are not just large but sustainable in a way previous technology waves could not. The 80% code-automation figure [2] is the most important data point in the entire IPO narrative because it suggests the company’s product is not just a tool for others but a force multiplier for its own operations. If that pattern holds, Anthropic’s margins could improve dramatically over time as the cost of engineering scales sub-linearly with revenue.

The risk, of course, is that the $47 billion revenue figure [1] includes a significant amount of pull-forward demand—customers buying AI credits in anticipation of future use cases that may or may not materialize. The enterprise AI market is still in its early innings, and a digestion period is possible as companies figure out how to integrate these models into their workflows.

Daniela Amodei’s confidence is not misplaced, but it’s also not risk-free. The IPO will be a referendum on whether the market believes AI’s returns are real enough to justify the valuations assigned to the entire ecosystem. If Anthropic succeeds, it will validate the thesis that AI is not just a cost center but a genuine productivity revolution. If it stumbles, the repercussions will be felt far beyond San Francisco real estate listings.

For now, the company bets that its own codebase is the best proof of concept. In a market starved for evidence that AI actually works at scale, that might be enough.

---

References

[1] Editorial_board — Original article — https://techcrunch.com/2026/06/04/ahead-of-its-ipo-anthropics-daniela-amodei-shrugs-off-doubts-about-ais-returns/

[2] VentureBeat — Anthropic says 80% of its new production code is now authored by Claude — how your enterprise can keep up — https://venturebeat.com/technology/anthropic-says-80-of-its-new-production-code-is-now-authored-by-claude-how-your-enterprise-can-keep-up

[3] Wired — What’s Worth More Than Cash in San Francisco Real Estate? Anthropic Stock — https://www.wired.com/story/whats-worth-more-than-san-francisco-real-estate-anthropic-stock/

[4] NVIDIA Blog — NVIDIA Partners With Microsoft on Unified Stack for Agentic AI Deployment, From Windows Devices to Cloud to Local — https://blogs.nvidia.com/blog/microsoft-build-windows-local-cloud-devices/

[5] SEC EDGAR — NVIDIA — last_filing — https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001045810