The $559M-versus-$12.3B quarter matters more than the $65B run rate

Bloomberg reported on Monday that Anthropic's annualized revenue run rate passed $65 billion at the end of July, up from $9 billion at the end of 2025, based on figures the company gave investors in a regular update ^[1]. The number that should change a procurement decision sits underneath that milestone: in the same quarter, Anthropic recorded $559 million in operating profit, its first ever, while OpenAI's operating loss widened to $12.3 billion from $9.3 billion the quarter before ^[2][3].

Deal with the run rate comparison first, then set it aside. Anthropic's most recent completed quarter brought more than $11.5 billion in revenue against $787 million a year earlier, roughly 14x, with sequential growth above 50% ^[5]. OpenAI reported $6.7 billion in Q2, up 18% from $5.7 billion, on a run rate above $40 billion ^[4]. On each company's most recently disclosed basis, Anthropic's run rate is ahead for the first time ^[6]. But run rate annualizes a point in time, usually the final month of a quarter, which is why $11.5 billion times four gives $46 billion rather than $65 billion, and $6.7 billion times four gives $26.8 billion rather than $40 billion ^[7][8]. Bloomberg flagged that the two companies may not calculate the metric the same way ^[9], and the metric says nothing about margin, customer concentration or the durability of current usage ^[18].

The income statements say something. Anthropic's $559 million on more than $11.5 billion of revenue is an operating margin under 5% ^[19]: profitable, barely. OpenAI's quarterly loss is roughly 1.8 times its quarterly revenue ^[20], and the loss grew about 32% sequentially while revenue grew 18% ^[21]. Same demand, same quarter, and the two results are close to $13 billion apart ^[22].

The gap is structural, not managerial. OpenAI carries hundreds of millions of free users who pay no subscription while every conversation consumes compute ^[11]. Anthropic's revenue is mostly enterprise, with reporting pointing to enterprise adoption of Claude Code as a major driver ^[12], and reporting also credits Anthropic with more efficient use of compute ^[13]. In one model incremental usage is a cost line; in the other it is an invoice.

That is where future pricing and rate limits come from. A sub-5% operating margin ^[19] leaves no cushion to absorb a compute or energy price move, so the plausible lever on the profitable side is metering: quotas, throughput and tier design on the compute-heavy products, not headline list prices. On the other side, HSBC estimates cited in the same reporting have OpenAI losing roughly $14 billion across 2026, cumulative losses near $44 billion from 2023 to 2028, and no profitability before 2030 ^[14]. That is fundable while capital keeps arriving; if it stops, the cheapest thing to cut is the part of the book that produces no revenue ^[11].

The caveat is not decorative. None of these figures are audited, Anthropic is private and has no disclosure obligation, and SiliconANGLE noted that the company did not say what method it used to calculate operating profit ^[10]. Anthropic confidentially submitted a draft registration statement to the SEC on June 1 and has not set a share count or price ^[15]; the prospectus is the first document that will show recognized revenue, operating expense and cash flow together.

Watch three things: whether the IPO arrives as early as fall 2026 at the reported $1 trillion valuation ^[17], how the audited numbers compare with the May Series H, which raised $65 billion at a $965 billion post-money valuation, roughly 15 times the run rate ^[16], and whether the reported 2028 projection of $190 billion to $200 billion in revenue survives contact with a registration statement ^[17].