(This article is authored by Outside the Box; published with authorization from Titanium Media.)
Article | Beyond the Page, Author | Drawing, Editor-in-Chief
The story of Silicon Valley often begins with a parting of ways.
Five years ago, a group of core researchers at OpenAI chose to leave. They held differing views on the direction of AI development and decided to start their own venture as a result.
This departure later gave rise to an AI company whose valuation and revenue have both surged rapidly: Anthropic.
In 2021, in an old factory building in San Francisco’s Pioneer Square, Dario Amodei and his sister Daniela Amodei were packing their belongings into cardboard boxes.
Those who left OpenAI alongside them included core researchers who later participated in the development of GPT-4, as well as key members of OpenAI’s safety team.
The reason for leaving OpenAI is not complicated: CEO Sam Altman is pushing OpenAI to transition from a nonprofit research organization into a company that pursues high valuations and commercialization. Dario believes this rapid sprint is causing safety to spiral out of control.
They named the new company Anthropic (meaning human-centric) and drafted a charter aimed at establishing a laboratory that prioritizes restraint and safety amid Silicon Valley’s AI frenzy.
Dario thought he had left Sam Altman’s business battlefield. But he soon realized that, along the technological trajectory of large models, the deeper one pursues safety and performance, the more pressing the demand for computing power becomes.
It is precisely this determination to remain independent of Sam Altman that has propelled it into a capital game from which there is no turning back.
I. Startup Capital and the Chosen Chips
To train cutting-edge models capable of competing with OpenAI, computational power is an inescapable physical barrier, and such power requires massive funding to sustain.
To maintain the independence of its research, Dario deliberately avoided bringing in traditional Silicon Valley venture capitalists in its early stages.
But fate took a dramatic, interconnected turn here. In 2022, the early $580 million funding for Anthropic came from Sam Bankman-Fried (SBF), the founder of the cryptocurrency exchange FTX.
However, just a few months later, FTX collapsed spectacularly, and SBF was brought to court.
As part of assets that were nearly deemed subject to liquidation, Anthropic’s equity faced the risk of being sold at a judicial auction, leaving the company on the brink of survival.
The market waits for no one, and AI competition evolves by the day—or even by the hour. This sudden funding crunch has left Dario with just a very short window to find a buyer large enough to provide stable computing power.
Cloud giants, sensing an opportunity, have jumped on board.
In the fall of 2023, Amazon announced an investment of up to $4 billion in Anthropic, followed by Google with $2 billion. While this funding provided immediate relief, it also revealed the true cost of the rescue effort.
To implement this agreement, AWS even built a dedicated large-scale cluster of Trainium2 chips. Anthropic also agreed to grant Amazon early priority access to tune its architectural improvements, tightly integrating chip design with algorithmic optimization.
II. Refusing the Throne
Just as Anthropic was struggling to make progress within the infrastructure frameworks of the industry giants, a turning point that seemed poised to reshape the landscape suddenly emerged.
On the evening of November 17, 2023, OpenAI’s board abruptly fired CEO Sam Altman. Over the following 48 hours, amid chaos, OpenAI’s board secretly called Dario Amodei and proposed an unexpected plan: inviting Dario to take over as OpenAI’s CEO and merging the two companies.
This is the moment when Dario is closest to regaining control of his former team and consolidating resources.
But he chose to refuse.
He saw through the tangled web of interests behind that governance crisis and ultimately returned to Anthropic’s lab to continue advancing the development of Claude.
This refusal also prompted Sam Altman, who has returned to OpenAI, to accelerate the multi-hundred-billion-dollar plan to expand computing power.
Sam Altman, who excels at the capital game, has been relentlessly pressing Anthropic’s market share with aggressive moves. To avoid being left behind, Anthropic has had to exponentially scale up its training cluster for the Claude 3 and 3.5 series.
By the time Claude 3.5 Sonnet was iterated, the computational cost of a single full training run and alignment process alone had already reached several times that of the previous-generation flagship model.
To avoid falling behind OpenAI in model iterations, Dario has no choice but to push Claude’s training scale and computational costs to even higher levels.
And to support this unprecedented spending on computing power, Anthropic needs to find a business model that can generate revenue quickly.
III. Business Models Driven by Compulsion
Under the financial pressure of computing power costs doubling every month, Anthropic couldn’t, like OpenAI, gradually cultivate end-user payment habits through a general-purpose chat interface aimed at the public; instead, it went straight to customers willing to pay big for efficiency.
Accordingly, Anthropic made a pivotal strategic shift, abandoning the race for consumer‑level traffic with ChatGPT and pivoting to the enterprise market.
In 2025, Anthropic launched Claude Code, a programming tool that runs directly embedded in the terminal command line.
This breakthrough stems from the team’s refactoring of the underlying prompt and computational graph architecture.
In the past, developers had to manually copy and paste the entire project’s code files into a web interface, whereas Claude Code can directly mount the local file system and precisely invoke indexing in the command-line background, automatically performing cross-file refactoring at the scale of tens of thousands of lines—all without exposing the company’s core assets.
It doesn’t waste resources on generic AI conversations; instead, it integrates directly into the software development workflow, handling code refactoring, bug triage, and architecture debugging for enterprises. Engineers no longer view Claude as a mere Q&A tool but as an engineering productivity asset whose input–output ratio can be quantified.
Revenue from pay-as-you-go pricing based on actual usage (API) has surged, with Anthropic’s revenue reaching $4.585 billion in 2025; by the first half of 2026, this figure had jumped to $16.286 billion, and the number of enterprise customers spending over $100,000 annually has grown to more than 11,900.
However, the commercial boom did not resolve Dario’s predicament; instead, it pushed him into an even more pressing situation.
The details in the prospectus reveal the cold-blooded nature of this business model. In the third quarter of 2026, more than 80% of revenue came from new model versions that had been released for less than six months.
This means that the rapid iteration of model capabilities can cause the commercial value of older versions to decline within a short period.
The more enterprise customers rely on Claude to handle complex tasks, the more Anthropic needs to continuously invest in acquiring higher-performance computing power to train new models.
A product developed to generate revenue for paying computing power fees ultimately requires the company to invest even more computational resources.
IV. Hidden Clauses in the Prospectus
On October 26, Anthropic officially filed its prospectus with the U.S. Securities and Exchange Commission (SEC), planning to list on Nasdaq under the stock symbol ANT.
The document reveals the interconnected business results of recent years.
In the first half of 2026, the company reported a book loss of as much as $80.748 billion. Of this amount, approximately $72.4 billion was attributable to changes in the fair value of convertible notes resulting from rising valuations. Under Non-GAAP measures, however, its operating profit for the first half reached $1.618 billion, and operating cash flow turned positive at $3.929 billion.
But what truly brings this five-year saga to a close is the hundreds-of-pages-long infrastructure procurement contract disclosed in the prospectus’ footnotes.
To fulfill its previous commitments to the tech giant regarding computing power and to sustain model iterations, Anthropic has secured long-term arrangements totaling approximately $637 billion, which include:
If combined with other computing power contracts from Microsoft, Lambda, and others, the overall long-term commitment scale is even higher.
Behind Broadcom’s $161.2 billion hardware partnership lies a customized integration of custom ASIC chips and next-generation optical interconnect buses. This means that, at the physical hardware level, Anthropic’s future computing infrastructure is already tightly aligned with Broadcom’s supply chain.
The data also shows that in 2025, 47% of Anthropic’s sales—approximately $2.16 billion—were generated through Amazon AWS and Google Cloud Marketplace, for which it paid $351 million in distribution channel fees.
Amazon and Google are both shareholders of Anthropic, as well as its largest computing power providers and distribution channels, while also developing their own Gemini models and hardware chips.
V. The Fate of Silicon Valley
Anthropic’s IPO is more than just a company going public; it also represents a rare collective bet in Silicon Valley in over two decades.
Since Google went public in 2004 and Meta (Facebook) rang the bell in 2012, Silicon Valley has not witnessed such a large-scale IPO that so closely integrates equity with infrastructure.
But this is no longer the Silicon Valley story that people once knew.
Over the past half-century, Silicon Valley has prided itself on the myth of garage startups combined with venture capital.
Two young people wrote code in a garage, secured millions of dollars in venture capital, and swiftly disrupted massive industry giants through their product experience.
That was the era of Fairchild Semiconductor, the era of Apple and Microsoft, and also the era of early Google and Facebook.
Meanwhile, Anthropic’s IPO marked the end of this solo‑player model.
No startup can complete this marathon on its own without relying on the cloud infrastructure of industry giants.
At its core, this IPO represents how Silicon Valley giants—leveraging capital, chips, and data centers—have co-opted a former rebel that once sought to challenge the old system, bringing it back into the established Silicon Valley industrial order.
Amazon, Google, Microsoft, and Broadcom are no longer just passive financial investors in the audience; they are investors, landlords, and channel partners—and the biggest winners of this technological revolution.
Even top university labs and research institutions, once independent and neutral players, have been forced to outsource their research projects to the cloud computing resources of tech giants because they can no longer afford to build compute clusters comprising tens of thousands of GPUs. This concentration of resources in the hands of a few major players is reshaping the entire Silicon Valley innovation model.
This is Silicon Valley’s new position in the AI wave: it remains a global hub for innovation and talent, but the reins of technological advancement have shifted from decentralized individuals and labs to a handful of tech giants that control computing power and capital.
Dario Amodei, the idealist who once stepped out of an old factory in San Francisco, spent five years completing a long circular journey.
He fled Sam Altman’s battlefield, only to ultimately deliver his answer to the new order of Silicon Valley.
[Off-Page] says:
Each generation of technological revolution lowers the barrier to innovation.
As the era when a company could be launched simply with code becomes increasingly difficult, the amount of computing power one possesses and the capital one can mobilize are also beginning to determine who is eligible to compete.
Technology continues to advance, yet the threshold for innovation is rising.
The next round of reshuffling in Silicon Valley may well begin right here.
For more exciting content, follow TMTPost on WeChat (ID: taimeiti), or download the TMTPost App.