Build2 publishers3 min readPublished
OpenAI's near-$70 billion run rate tops Anthropic's July figure
OpenAI is nearing a $70 billion annualized run rate, with business revenue more than doubling since July, according to sources who spoke to Axios. Until the IPO filings show costs, buyers are comparing an unaudited monthly pace with an Anthropic figure from July.
The Engineer · Build desk
What happened
- OpenAI's annualized run rate rose more than 70% from the start of the third quarter, according to the people who briefed Axios.
- Anthropic's run rate was above $65 billion by late July, according to Bloomberg figures that Axios cited in August.
- At DevDay, OpenAI said 2.5 million businesses use its products and more than 35 million people use ChatGPT Work and Codex each week.
- Anthropic's prospectus, reviewed by Reuters, lists $518 billion in future cloud, computing and infrastructure obligations.
- OpenAI did not supply expense details alongside the new run-rate figure, Axios reported.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- contradiction Runtime Wire reads the numbers as a reversal, but The Decoder says Anthropic may already match or exceed OpenAI, so which lab leads depends on how current the Anthropic figure is.
- constraint Contracted revenue, API use and other products are mixed into one pace, so a buyer cannot tell how much of OpenAI's new business spend is committed and how much could leave after the next price cut.
- exposure The one filing on record shows a frontier lab spending well over a dollar to earn one, so today's per-token prices could be repriced once public shareholders ask for margins.
- decision A team that signs a multi-year commitment with either lab now signs before OpenAI's public prospectus shows whether its growth holds up after compute costs, while shorter terms leave that choice open.
A run rate takes the current month of revenue and projects it over a full year [7]. At close to $70 billion, OpenAI is billing about $5.8 billion a month [2]. When revenue climbs this fast, that pace runs far ahead of booked revenue. Anthropic's late-July run rate is about 14 times the nearly $4.6 billion of revenue it reported for all of 2025 [15][7].
The Axios report uses both "annual recurring revenue" and "annualized revenue run rate" for OpenAI's figure, and it does not say how either was computed [5]. The numbers came from people familiar with the company's finances. They are not audited results [4].
Even so, the OpenAI numbers are consistent with each other. A rise of more than 70% that ends near $70 billion puts the start of the third quarter at about $41 billion or less [1]. In August, Bloomberg cited people familiar with the matter and reported more than $40 billion [8].
The Anthropic comparison is weaker. Runtime Wire calls OpenAI's numbers the September figures [22]. The Anthropic figure they are measured against dates from late July [9]. On paper the gap is at most about $5 billion, or 8% of Anthropic's July pace [3]. Runtime Wire says OpenAI is ahead only if the two estimates use comparable definitions and dates [10]. It also calls the figures a rapid reversal in the enterprise race [12]. But the Anthropic number in that comparison is a total run rate. The evidence supports a narrow lead on total pace. It does not show an enterprise lead.
OpenAI's own mix is changing on both the business and consumer sides. In March it said enterprise made up more than 40% of revenue and was on track to match consumer revenue by the end of 2026 [13]. Assume that share held into July. Business revenue doubling inside a total that grew 70% would then put enterprise near 47% [4]. OpenAI also added more consumer revenue in the third quarter than in all of 2025 [14].
The only cost figures on record for either lab come from Anthropic's prospectus. Reuters reviewed it and reported more than $8 billion of 2025 operating losses, excluding certain financing-related accounting charges [15]. That is about $1.74 or more of operating loss per dollar of revenue [5]. Its future obligations equal about eight years of revenue at its July run rate [6]. The Decoder says OpenAI's growth comes from enterprise sales and an aggressive price war against Claude and Chinese models, which OpenAI stepped up with GPT-6.1-Sol [20].
I think the OpenAI figures are good evidence of volume and weak evidence of a lasting lead. The first public document with OpenAI's margins will be its prospectus. OpenAI said in June it had filed confidentially for a U.S. IPO, according to the Associated Press [18]. Anthropic is preparing to list as early as November [19].
What to watch
- OpenAI's public prospectus, the first document that would show its margins and how much of its revenue is contracted.
- Anthropic's IPO, possibly as early as November, and any newer run rate showing whether it has passed OpenAI's September pace.
- Per-token price changes after GPT-6.1-Sol, which would show whether the price war The Decoder describes is still running.