Anthropic goes public and reveals the real cost of cutting-edge AI

The company behind Claude debuted on Nasdaq with a valuation of $93.5 billion and an opening price of $111.50. Behind the ticker, pressure is mounting on its margin, governance, and safety narrative.

Anthropic, the publisher of Claude, made its debut on Nasdaq Thursday with a valuation of $93.5 billion, before closing Friday at $88.91 after opening at $111.50. This IPO, described by CNBC as a "historic Nasdaq debut," places the company among the most highly valued firms currently, but it also highlights a much less glamorous topic: the cost of cutting-edge AI and how to finance it without compromising the safety that the company places at the core of its identity.

An IPO that tests the "safety first" narrative

The market did not just validate a financial operation. It also gave a first real-world test to Anthropic's narrative, which presents itself as a more cautious alternative to OpenAI. In the exchange published by The Verge, Hayden Field recalls that the company must now reconcile two demands pulling in opposite directions: continuing to invest in the safety of its models and satisfying shareholders who expect rapid growth, or even visible profitability in the short to medium term.

The timing is not neutral. The company took the lead while its competitor OpenAI carries a more troubled public reputation. According to Hayden Field, some executives would have even preferred to let Anthropic pave the way, to observe how the market values a large generative AI company before telling their own stock market story. The message sent to the sector is clear: the window for major publicly traded AI labs is open, and no one wants to stay in the hallway too long.

Le vrai sujet : combien coûte une IA de pointe à faire tourner ?

At the heart of the matter is a simple economic mechanism to state, less simple to execute. Anthropic rents computing capacity from cloud providers, pays for electricity, GPUs, and infrastructure, then must sell its models at a price higher than all these costs combined. The company must also finance the training of new models, which further increases the bill. According to Nilay Patel, the total compute-related cost reaches "roughly $900 billion over the coming years." [to be verified]

This constraint explains why Anthropic pushes its revenues further, even if it means passing some costs onto customers. The Verge mentions a price increase movement that sparked criticism among some users, followed by a slight public backtrack. At the same time, the company relies on another lever: making Claude sufficiently central in professional uses so that switching providers becomes costly, thus rare. In this logic, loyalty is not sentimental. It is billed. A bit harsh, but very Silicon Valley.

La sécurité comme argument commercial, et comme ligne de tension

Anthropic has built its positioning on the idea of a safer, more controllable, and more trustworthy AI. This narrative also serves sales, especially among large accounts and administrations, where data sovereignty, auditability, and compliance requirements weigh heavily. Companies, the exchange reminds, do not always want to change models with every price update; they want to know where their data goes and under what conditions they can leave if needed.

The problem is that this promise of caution is costly. The more the company wants to keep teams dedicated to safety, the more it potentially slows some launches and cuts into its margins. The faster it accelerates to stay in the race, the more it risks weakening what distinguishes it. This is where the IPO changes everything: before, this dilemma remained largely internal; now, it is measurable quarter after quarter, with investors on the front line.

Une gouvernance conçue pour résister aux actionnaires

Anthropic did not choose a classic public company structure. Power remains concentrated in the hands of the founders and a Long-Term Benefit Trust, or LTBT, a vehicle intended to protect the long-term mission. This trust notably includes Ben Bernanke, Richard Fontaine, Christopher Stubbs, and Neil Buddy Shah. But its role is more limited than it seems: it can appoint or remove certain directors, but not manage the company's daily operations.

The most unique structure remains that of the founders, through an entity called Founder LLC, which holds a single share representing 50.1% of voting rights. In practice, the co-founders therefore retain decisive influence over major directions. Ann Lipton, a law professor cited by The Verge, sums up the issue: one must either trust the founders and the trust to properly arbitrate between private interests and the general interest, or not trust them. There is not much middle ground.

Les chiffres qui comptent vraiment désormais

For the markets, Anthropic will not be judged solely on its narrative but on its operating figures. The first signal to watch is revenue growth, which some analysts consider insufficient given the current valuation. Steve Rattner mentions an increase of about 17 times over ten years to offer a return comparable to that of major tech names. [to be verified]

Another point of vigilance: the ability to buy enough compute without becoming financially strangled. Dario Amodei summarized the problem as the company's "deepest business dilemma," according to The Verge. Too little capacity and R&D slows; too much capacity and the balance sheet tightens. Between the two, it is necessary to sell more, better, and fast enough. Nasdaq, for its part, does not reward good intentions.

Ce que cette IPO dit du marché de l’IA

The Anthropic case also tells another story: the value of large models now depends as much on technical quality as on the chip supply chain, the cost of computing, and the bargaining power with suppliers. CNBC notes that the IPO comes as Nvidia closed, the following Friday, its longest series of weekly declines since 2008. This coincidence fuels a broader question about the robustness of AI infrastructure financing.

Anthropic already has financial ties with Nvidia, which agreed to buy $10 billion of the company's shares via a private placement, in addition to a $10 billion investment pledge announced last year. It remains to be seen whether this proximity will protect the startup or expose it to greater dependency. For observers, the real measure of the moment will not be the closing price on a given day, but what Anthropic chooses to sacrifice, or not, to keep its safety promise while growing at stock market speed.

To read