According to Bloomberg, Anthropic could publicly file its IPO documents as early as late August, for a deal the company anticipates at or above the record set by SpaceX. Annualised revenue is said to have reached around $65 billion by the end of July, up sharply, driven by enterprise spending on Claude and agents.
We wrote a week ago that OpenAI's IPO would be the most revealing event of the year, before learning that its timeline was slipping into 2027. It is ultimately its rival that could go first.
What the numbers say
These figures need to be handled with care, because they circulate in different forms.
We reported on Monday that the company announced around $10.9 billion in revenue and its first operating profit for the second quarter. The $65 billion figure refers to something else: an annualised run rate, that is, the last month's revenue multiplied by twelve. It is a projection, not a reported result.
This distinction is not pedantry. An annualised run rate well above four times the previous quarter signals very rapid acceleration. It also signals that the figure reflects a snapshot, and that it will vary.
Why this is a shift in posture
The most interesting point is not financial, it is strategic. We described in our comparison of the two labs a clash of doctrines: one betting on maximum integration, the other on independence and caution.
But going public means accepting a quarterly constraint. A listed company must publish, explain, and account for itself every three months. That sits poorly with a long-term posture, and even less with decisions like a deliberate slowdown in development or an acknowledged conflict with a major public client.
The question becomes: can a company whose central argument is that it refuses certain things on principle hold that line before shareholders who did not sign up for it?
This is the part that interests us most, and it applies to the whole sector. An IPO document discloses, under legal liability, the real cost of inference, the split between consumer, enterprise and API revenue, customer concentration, and above all off-balance-sheet commitments. We have documented data centre leases and a recent twenty-year commitment on 191 megawatts. Seeing those amounts set down in an audited document, against revenue, will be the first honest measure of the economics of a frontier lab.
The caveats
Three precautions are in order.
Nothing has been filed. These are reported preparations. IPO timelines regularly slip, as OpenAI's just did.
Growth must hold. A valuation tied to an annualised run rate assumes that rate continues. But we noted, regarding the quarterly profit, that an exceptional ramp-up at a major client may have flattered the period.
Spending follows. Spectacular revenue growth accompanied by even more spectacular infrastructure commitments says nothing about sustainable profitability. That is precisely what the prospectus will settle.
What to take away
The timing is remarkable. This preparation comes the same week as the watermark bypass announced a few days earlier and in a climate where the company's own chief executive describes a crisis of trust in the sector.
An IPO in this context is a bet on two things at once: that markets believe in the AI economy, and that they accept a company refusing certain markets on principle. The second point is the most uncertain, and it is what will make this case interesting to follow well beyond the share price.