Why AI's Most Idealistic Labs Were Forced to Choose Profit
Why AI's Most Idealistic Labs Were Forced to Choose Profit
Two AI labs. Two founding ideals. One gravitational force.
OpenAI and Anthropic each began with a conviction that the most consequential technology in human history was too important to be left entirely to the market. Both learned, at scale, that the market doesn't particularly care about that conviction.
I. The Letter
On the evening of June 12, 2026, the Commerce Department delivered a letter to Anthropic CEO Dario Amodei. The directive invoked national security authority and ordered the immediate suspension of Claude Fable 5 and Mythos 5 for all foreign nationals, everywhere on earth, including Anthropic's own non-citizen employees. The models had been publicly available for three days. Anthropic, unable to filter its user base by citizenship in real time, shut both models down entirely. The company described the situation as a misunderstanding and said it was working to restore access. As of this writing, the models remain offline.
The letter was the visible event. The structure beneath it had been forming for years.
II. The Company Anthropic Set Out to Be
Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and several colleagues who had departed OpenAI. The departure was not quiet. The founders carried with them a specific concern: that the pressure to commercialize AI was outrunning the ability to understand and control it. They built Anthropic around a framework called Constitutional AI, a method of training models to evaluate their own outputs against a set of explicit principles. The founding premise was that safety and capability were not in opposition — that a model could be both highly capable and reliably aligned with human values, provided the development process was structured around that goal from the beginning.
That premise required capital to test. A lot of it.
$13 billion...Amazon's total investment in Anthropic, with reported plans to exceed $25 billion
Anthropic raised its first significant outside capital in 2022. Google invested. Then Amazon invested — first $4 billion, then $8 billion more, then a commitment in April 2026 to spend over $100 billion on AWS infrastructure. Anthropic's primary cloud provider became its largest financial backer. That same backer held a board seat. The company that had set out to demonstrate that AI development could be done differently was, by the spring of 2026, operating inside one of the most entangled financial relationships in the history of the technology industry.
In May 2026, The New York Times reported that Anthropic was preparing to file for an initial public offering, targeting a debut in the fall. The company that was founded to be something other than a conventional technology company was preparing to become one.
III. The Company OpenAI Set Out to Be
OpenAI was founded in 2015 as a nonprofit. The founding documents described the organization's purpose as ensuring that artificial general intelligence would benefit all of humanity. The structure was deliberate: by organizing as a nonprofit, the founders signaled that the work was too important to be governed by the interests of shareholders. The mission was the point. The organization existed to serve it.
The capital requirements of frontier AI research made that structure difficult to sustain. In 2019, OpenAI created a "capped profit" subsidiary — a hybrid structure designed to attract outside investment while preserving the nonprofit's oversight role. Microsoft invested $1 billion. Then more. Eventually, Microsoft's total commitment to OpenAI exceeded $13 billion. The nonprofit's board retained formal authority. In November 2023, that board exercised it, briefly firing CEO Sam Altman over concerns about candor and the pace of development. Within days, Altman was reinstated. The board members who had voted to remove him were gone.
The nonprofit's board retained formal authority. In November 2023, it exercised that authority. Within days, the board members who had voted to remove Altman were gone.
By 2025, OpenAI had filed with regulators to restructure as a conventional for-profit company, formally severing the governance link between the nonprofit mission and the operating entity. The nonprofit retained a financial stake. It no longer held the controls. OpenAI filed for an IPO in 2026. The company that had been founded to ensure AI benefited all of humanity was preparing to issue shares on the public market.
IV. The Rest of the Field
It is worth pausing here to note what the rest of the frontier AI landscape looks like, because the contrast is instructive.
Microsoft's Copilot is a product of one of the most profitable companies in the world. Microsoft was never a nonprofit, never organized around a safety mission, and never pretended otherwise. Its investment in OpenAI was a strategic acquisition of capability, not an ideological commitment. When OpenAI's governance structure created friction, Microsoft's interests were clear and consistent.
Google's Gemini emerged from a company whose core business has always been advertising revenue and shareholder returns. Google DeepMind does serious scientific work. It does that work inside a publicly traded corporation with obligations to its investors that predate any individual AI model by decades. There was no mission drift at Google because there was no alternative mission to drift from.
Elon Musk's xAI, which produces the Grok models, was capitalized through private funding with explicit commercial intent from its founding. Musk departed OpenAI's board partly over disagreements about commercialization. He then built a commercial AI company. The structure was always transparent.
Mistral, the French AI lab, and DeepSeek, the Chinese lab, each operate with significant state-adjacent resources and neither was organized around a nonprofit or mission-first premise. The capital source, in both cases, was never arm's-length from broader national interests. The tension between mission and money that defines the OpenAI and Anthropic stories simply does not arise in the same form when the capital and the mission are aligned from the start — or when there is no stated mission to protect.
This leaves two companies in a category of their own: the ones that tried.
V. The Physics of the Situation
The argument here is not that OpenAI or Anthropic failed, or that their founders compromised their principles, or that the investors who supplied capital acted in bad faith. The argument is simpler and more structural than any of that.
Developing a frontier AI model requires compute at a scale that is difficult to fully convey. Training runs consume tens of thousands of specialized chips running continuously for months. The infrastructure required to serve a model to millions of users is measured in data centers, not server rooms. The engineering talent capable of doing this work commands compensation that reflects its scarcity. None of this is cheap. None of it was going to become cheap on a timeline relevant to the competitive race that began in earnest around 2022.
$100 billion...Anthropic's commitment to AWS infrastructure, announced April 2026 — the price of staying competitive at the frontier
A nonprofit, or a mission-first organization structured to resist commercial pressure, can absorb one funding round at this scale. It can absorb two. At some point, the capital requirements and the governance structure become incompatible. The investors supplying billions of dollars are not doing so out of philanthropic interest. They have obligations of their own — to their shareholders, to their limited partners, to the federal government in the case of contractors. Those obligations do not disappear because the recipient of the capital has a safety charter.
This is not a story about villains or heroes. It is a story about arithmetic. When the cost of the product exceeds what a mission-driven structure can raise, the structure changes to accommodate the cost. This has happened before in technology, in pharmaceuticals, in defense research. The specific form it takes in AI — the nonprofit wrapper, the capped profit hybrid, the safety charter coexisting with the billion-dollar investment round — is new. The underlying dynamic is not.
This is not a story about villains or heroes. It is a story about arithmetic.
VI. Where This Leaves Things
As of June 16, 2026, Anthropic's most capable models are offline. The company is in an unresolved dispute with the federal government over the conditions of their reinstatement. Its largest investor triggered the government action that pulled the models. That investor holds a board seat. Anthropic's IPO timeline is uncertain. The safety mission that distinguished the company from its competitors is now, in some readings, the source of its conflict with the Department of Defense, which sought access to Claude for uses that Anthropic's Constitutional AI framework was specifically designed to prevent.
OpenAI, meanwhile, is a for-profit company with a public offering on the horizon and a nonprofit that retains a financial stake but no operational control. The board crisis of November 2023 — which briefly looked like a governance structure asserting itself against commercial pressure — resolved in favor of the commercial operation. The capped profit structure is gone. The mission lives on as language in a filing.
Both companies began with a version of the same conviction: that the development of transformative AI was too important to be governed solely by market incentives, and that a different kind of organization was needed to do it responsibly. Both organizations are now, in their operational reality, governed substantially by the same forces that govern every other large technology company: capital requirements, investor expectations, competitive pressure, and the demands of the public market.
Whether a structure exists that can hold a genuine safety mission together with the capital requirements of frontier AI development is a question that neither company has answered. What the last decade of their parallel histories suggests is that nobody has found it yet.
The question remains open. So does the market.
Tech Reader Magazine
TechReaderMagazine.com