Anthropic is heading for an IPO that could value it at up to $2 trillion — but the growth numbers behind that bet just triggered a global tech selloff. Bloomberg reported on August 18 that the company's July annualized revenue run rate (ARR) hit $65 billion, below the $70–80 billion that buy-side trackers had been modeling, and markets read the miss as the first real revenue check on the AI capex boom.
The $2 trillion IPO that could break SpaceX's record
According to the Financial Times, Anthropic could go public as soon as October at a target valuation of up to $2 trillion. The company confidentially filed for a US IPO back in June; timing and pricing are still not final. If the valuation holds, it would beat the $1.77 trillion record SpaceX set in June — in under four months.
But the past week has turned the spotlight from the IPO itself to the numbers underneath it. A Bloomberg disclosure, an analyst survey, and a leaked OpenAI quarter landed in quick succession, and each one sharpened the same question: can the revenue justify the capex?
Why $65 billion in ARR caused a selloff
The headline numbers are still exceptional. July ARR of $65 billion is up 38% from May's $47 billion and roughly seven times the $9 billion run rate at the end of 2025. Preliminary Q2 revenue passed $11.5 billion, up nearly 14x year over year, with adjusted operating profit positive.
The problem was expectations, not growth. Third-party trackers like Yipitdata and TickerTrends had modeled $70–80 billion, and a sell-side research survey came in at $78 billion. The actual $65 billion landed below even the conservative end of that range.
Markets did not shrug. The Philadelphia Semiconductor Index fell nearly 5% the same day, with SanDisk and Micron down over 7%; the Nasdaq dropped another 1.33% the following session, and China's STAR 50 and ChiNext indices fell more than 5% on August 20. A single ARR print moved global markets in three days.
The mechanism: Anthropic's ARR is the most important anchor in the AI capital-expenditure narrative. The four hyperscalers — Amazon, Microsoft, Alphabet, and Meta — have pushed combined annual capex from roughly $456 billion to near $750 billion in two years. That money only pays off if model vendors turn compute into revenue. Anthropic, as the first pure-play large-model company heading to public markets, is the clearest gauge of whether that loop is working.
Claude Code: the engine behind the reversal
Twelve months ago the gap looked unbridgeable: Anthropic was at roughly $9 billion ARR against OpenAI's $20 billion. The line crossed around March this year, and by late July Anthropic was above $65 billion while OpenAI sat near $40 billion.
The driver was a single product. Claude Code hit $500 million ARR within three months of launch, crossed $1 billion in about six, passed $2.5 billion by early 2026 — and then went from $2.5 billion to $15.1 billion in about seven months. TickerTrends tracking as of August 10 puts Claude Code's ARR at $15.12 billion, implying an Anthropic total of roughly $69 billion.
Coding worked because it is the highest-frequency paid AI use case with the deepest toolchain lock-in: individual subscriptions naturally upgrade into enterprise contracts and API usage. More importantly, Claude Code proved with real revenue that the biggest paid AI scenario is doing work directly. Cursor, Windsurf, Replit, GitHub Copilot, Google Jules, and a refocused Codex all followed — and the ecosystem now extends to autonomous workflows like Cursor's /goal mode, where agents chase long-running goals until the job is done. For a broader map of the category, see our AI programming tools guide.
OpenAI's widening losses add fuel
OpenAI's numbers, reported by the Wall Street Journal, intensified the mood. Q2 revenue came in at $6.7 billion, up just 18% quarter over quarter from $5.7 billion, while operating losses widened from $9.3 billion to $12.3 billion — roughly $1.80 burned for every $1 earned. Group ARR passed $40 billion in August with Q3 ARR up 35% quarter over quarter, but the deceleration in the quarterly print was enough to make the market reassess the pace of AI monetization.
The competitive detail matters too. Claude Code's tracked ARR growth slowed to roughly 5% month over month, while OpenAI's Codex — at $8.83 billion tracked ARR — grew 20.8% in the same period. The smaller product is chasing at four times the speed, and the leader's deceleration is exactly what a valuation reset feeds on.
The first revenue check on a $750 billion bet
Two years of a "hardware-first" narrative — build compute, revenue will follow — are now being tested by demand-side data. The loop is simple: hyperscalers build data centers, model vendors sell tokens, enterprises pay for outcomes, and revenue flows back to justify the next round of capex. Anthropic's ARR was the fastest-moving gauge of that loop, and its slowdown is the first signal that supply and demand now have to validate each other in public.
The same doubt is spreading in China. Zhipu's chief scientist recently conceded that trillion-parameter scaling was a detour the whole industry took together — a rare admission from a major lab that efficiency, not raw scale, is where the next phase gets decided.
Three numbers to watch: the final IPO pricing against the $2 trillion target; Anthropic's Q3 ARR growth; and the Claude Code–Codex gap. Any one of them moving sharply will reprice the whole sector.
What builders should do now
For product teams, the signal is unambiguous: coding agents are the only AI category with proven, repeatable paid revenue at scale. That is where budgets will keep flowing, so it pays to build workflows around tools that already show durable usage — Claude Code, Codex, and autonomous-agent setups like Cursor /goal.
For anyone pricing AI products, the next battleground is cost per task, not raw model quality. As unit economics tighten across the industry, products that make tokens cheaper to spend — not just smarter — will win renewals.
For investors, the story era is over. Revenue growth rate and burn are now the price drivers, and the IPO itself is the biggest event risk on the calendar: if pricing holds at $2 trillion, it resets the ceiling for every AI company; if it slips, expect the repricing wave to continue.
FAQ
Is Anthropic really going public at a $2 trillion valuation?
Financial Times reporting says the company could list as soon as October at a target valuation of up to $2 trillion, after a confidential IPO filing in June. Timing and final pricing are not confirmed.
Why did Anthropic's ARR number crash AI stocks?
July ARR of $65 billion missed buy-side forecasts of $70–80 billion. Because Anthropic's revenue is the anchor for roughly $750 billion of hyperscaler capex, the slowdown was read as the first demand-side warning — the Philadelphia Semiconductor Index fell nearly 5% the same day.
Is Claude Code still growing faster than OpenAI's Codex?
Claude Code's tracked ARR was $15.12 billion as of August 10, about 1.7x Codex's $8.83 billion. But Codex is growing at 20.8% month over month versus roughly 5% for Claude Code, so the gap is closing fast.
