Our last issue called Anthropic's Q2 numbers a run-rate story. They are not a run-rate story anymore. The Q2 operating profit of roughly $559 million on $10.9 billion in revenue, with the compute-to-revenue ratio falling to roughly 56 cents and a figure that includes model-training costs but excludes stock-based compensation, has now been independently modeled by SemiAnalysis, which puts the Q2 EBIT margin at approximately 36 percent. That is the confirmation this story needed to move from projection to pattern. The firm's July 2026 deep-dive then extended the line forward: SemiAnalysis projects Anthropic will achieve GAAP EBIT of $1 billion in Q3 2026, at a 6 percent margin, with annualized recurring revenue surging from $9 billion at end-2025 to over $60 billion. A single profitable quarter can be a weather event. Two consecutive quarters with the second doubling the first is something closer to a climate change.
The mechanism behind the shift is worth tracing carefully, because it is not the obvious one. The engine behind these numbers is not a consumer chatbot; it is Claude Code, Anthropic's enterprise AI coding tool, and its agentic architecture. The API business now boasts gross margins exceeding 80 percent and a net revenue retention rate of 500 percent. Claude Code now accounts for over 7 percent of GitHub code commits. Claude Code hit $2.5 billion in annualized revenue in under nine months, with over 1,000 business customers spending more than $1 million per year. Usage-based pricing in a category where customers expand rather than churn produces the kind of retention curve that makes investors reach for superlatives; a 500 percent net revenue retention number means that the average enterprise dollar from one year ago has become five dollars today.
The compute picture is where the story gets genuinely complicated, and where the skeptics deserve a fair hearing. According to documents filed as part of SpaceX's IPO, Anthropic will make monthly payments of $1.25 billion until May 2029, with fees reduced during the ramp-up period in May and June 2026. That last clause is the one critics seized on immediately. Anthropic is paying SpaceX $1.25 billion a month, but at a reduced fee as it ramps up through precisely the months it is using to tell investors it has an operating profit, which implies a normal monthly bill of $15 billion a year in compute costs. The projected Q2 operating profit is real but may be temporary, as Anthropic told investors that heavy compute commitments are expected to compress margins in late 2026 and early 2027; read it as a window of profitability, not yet a permanently profitable company. The SemiAnalysis Q3 projection addresses this directly: the firm does not identify a source for the Q3 profit figure beyond its own financial modeling, and the analysis reflects a forecast rather than disclosed audited results, as Anthropic has not published standalone financial statements as a private company.
Where the numbers land with force is in the direct comparison to the rest of the frontier lab landscape. OpenAI expects to spend roughly $22 billion in 2026 against $13 billion in revenue, producing a $14 billion operating loss, with cumulative projected cash burn through 2028 exceeding $143 billion. Anthropic's Q3 2026 quarterly operating profit is projected to arrive two to three years ahead of OpenAI's breakeven. SemiAnalysis calculates that if both companies reach $100 billion in ARR, OpenAI's cost of services would be higher due to supporting its massive free user base, resulting in approximately $25 billion less in gross profit than Anthropic. That structural gap, not just the quarter, is what resets expectations for every lab still burning cash on consumer distribution.
All of this lands directly into an active IPO process. Anthropic confidentially submitted a draft S-1 registration statement with the SEC on June 1, 2026, following the announcement of a $65 billion Series H funding round at a $965 billion post-money valuation. The company is targeting a possible public listing as early as October 2026 and has reportedly engaged law firm Wilson Sonsini, which managed Google's 2004 IPO, to assist with public-market readiness. The profitability estimates now circulating, including the SemiAnalysis Q3 figure, are widely read as pre-IPO financial framing, and public-market investors will demand more precision than 'roughly $1 billion' once an S-1 is unsealed; the gap between analyst estimates and actual disclosed GAAP numbers is a real risk to the valuation Anthropic is targeting above its $965 billion Series H mark. Anthropic's gross-versus-net revenue accounting practice also inflates headline figures relative to net-reporting peers, a question the full S-1 will need to resolve. The audited numbers, when they arrive, will be the most consequential financial disclosure in the short history of the AI industry, either confirming that the profitability turn is structural or revealing that two good quarters were engineered by a ramp-period discount.
The real test is not whether Anthropic earned $559 million in Q2 but whether an S-1 filed under SEC review in the next sixty days shows the same arithmetic still holds when SpaceX bills full price.