Anthropic has informed existing investors that its annualized revenue run rate surpassed $65 billion as of the end of July, according to sources familiar with the matter. This marks a significant jump from the $47 billion reported in May and represents a more than sevenfold increase from approximately $9 billion at the end of 2025. For the parent company of Claude, which is preparing for its public market debut, the message is clear: it aims to prove that AI revenue is not merely a fundraising narrative but a tangible, recurring business.
The growth trajectory is steep when broken down quarterly. First-quarter revenue was approximately $4.73 billion, with preliminary second-quarter figures exceeding $11.5 billion, representing a sequential increase of over 140%. Compared to the same quarter last year at $787 million, this is roughly a 14-fold expansion. Analysts note that the company has likely turned positive on an adjusted operating profit basis, though net income may not yet follow suit. While a run rate annualizes recent sales rather than reflecting audited full-year revenue, the direction is compelling: enterprise APIs and developer tools are converting usage into repeatable billing.
Product-wise, Claude Code is accelerating. This agentic tool for programming workflows was fully launched in mid-2025, reached approximately $1 billion in annualized revenue by November of that year, and grew to about $2.5 billion by February 2026, continuing to capture enterprise R&D budgets. Eight of the Fortune 10 are now customers, with enterprise seats significantly up year-over-year. Token-based API billing remains the core revenue stream, while subscriptions (Pro, Max, Team) provide more stable income from individuals and small teams.
Growth has not been without disruptions. Quartz reported that the company briefly removed some models due to export control directives and faced defense supply chain risk assessments, yet the latest run rate continues to climb. The timeline compresses into a clear arc: roughly $87 million in annualized revenue in early 2024, about $1 billion by end-2024, $9 billion by end-2025, $30 billion in April 2026, $47 billion in May, and $65 billion by late July. For context, Salesforce took roughly two decades to reach $30 billion in annual revenue; Anthropic has crossed the same threshold in a far shorter period on a run-rate basis. This is the raw material investment banks are now using to test valuations.
Where the IPO stands
On May 28, Anthropic closed a $65 billion Series H round at a post-money valuation of $965 billion, surpassing OpenAI's private valuation of approximately $852 billion at the time. Lead investors included Altimeter, Dragoneer, Greenoaks, and Sequoia. On June 1, the company confidentially filed its S-1 with the SEC, with Morgan Stanley, Goldman Sachs, and JPMorgan as underwriters. Market speculation points to a listing this autumn, possibly as early as next month, potentially ahead of OpenAI. OpenAI's run rate is around $40 billion, though the metrics may not be directly comparable; its IPO has reportedly been delayed to next year, partly linked to post-IPO volatility in SpaceX's stock.
The valuation multiple is not cheap, but it leans more heavily on revenue than competitors. Based on the May run rate of $47 billion, the $965 billion valuation implies roughly 20 times; using the latest $65 billion, the multiple compresses to about 15 times. Some investors are modeling a year-end 2026 run rate of $100–120 billion and 2028 revenue of approximately $190–200 billion to support a higher listing price. These are investor projections, not company guidance. Private secondary markets had briefly assigned higher implied valuations, but secondary premiums often evaporate on IPO day. SpaceX's volatile trading since its June listing has already stress-tested what a "historic offering" might look like.
Computing costs drive M&A talks
The other side of growth is the computing bill. Anthropic rarely makes large acquisitions, yet ahead of its IPO it is in talks to acquire Decart for roughly $6–7 billion, mostly in stock. Decart focuses on chip efficiency and inference optimization, handling real-time video and world models; if the team integrates into the inference and performance division, the goal would be to extract more tokens from existing GPUs, TPUs, and Trainium hardware rather than building another data center. The deal is not yet closed, and Nvidia had offered a higher price, but the founding team and some shareholders favor Anthropic. For a company about to face public investors, this is a "buy throughput, control costs" transaction: no matter how steep the revenue curve, if unit inference costs do not decline, the margin story weakens.
Amazon and Google remain key computing and capital partners, and the Series H also brought in memory and supply chain names like Micron, Samsung, and SK Hynix. AI infrastructure capital expenditure is shifting from one-sided cloud expansion to a shared bill among model companies, chipmakers, and memory suppliers. This is also why, during this week's memory stock selloff, the market read Anthropic's numbers as "demand remains, but financing and debt pressures persist."
What it signals for markets
For semiconductors and cloud, the $65 billion run rate is hard evidence of demand-side strength: as enterprises delegate coding, customer service, and knowledge work to models, usage will continue consuming high-end GPUs, HBM, and power. For bond markets, it reinforces the narrative that hyperscaler capex will not slow soon, with tech debt supply still competing with Treasuries for buyers. For equities, the implication is more split: revenue is materializing, but if the listing price anchors on a $100 billion-plus run rate and a trillion-dollar market cap, any growth deceleration could be framed as a broken narrative.
Three numbers need to be separated. First, a run rate is not GAAP full-year revenue; seasonality, discounts, and one-off enterprise contracts can distort the extrapolation. Second, turning positive on adjusted operating profit does not equal positive free cash flow; training next-generation models will still burn cash upfront. Third, OpenAI, Google, and other coding tools will compete for the same enterprise budgets, and Claude Code's share should not be assumed to extrapolate linearly forever.
The correct reading of this update is not "the next SpaceX is already priced," but rather: Anthropic has moved AI from a concept stock into a phase where it can be discussed in terms of quarterly revenue. Whether the IPO can digest a valuation above $965 billion depends on whether growth in the coming months can approach the first-half pace, and whether inference costs can be driven down through acquisitions and custom silicon. If it lists this autumn, investors are not buying a chatbot story but a company turning enterprise workflows into recurring revenue—and the question of whether it can sustain triple-digit growth at a trillion-dollar market cap.