The AI revenue story just got a $20B haircut. The haircut isn't the story.

On the afternoon of October 8, the Nasdaq fell 1.25%. Nvidia dropped 2.9%. CoreWeave fell close to 8%. Oracle lost 5.5%. The proximate cause was a Financial Times report putting OpenAI's September annualized revenue at roughly $50 billion — about $20 billion below the $70 billion figure that had been circulating among investors since late September.

Okay so here's the thing: both numbers are probably right. They're measuring different companies in different ways.

Source spread

• Financial Times (via The Next Web) — skeptic. Reports $50B September figure and the discrepancy with investor-circulated $70B; notes OpenAI declined to comment.
• Axios — "OpenAI's annual recurring revenue nears $70B" — hype. September 29 scoop that put the run rate above $70B; the framing is positive and used by investor decks.
• Wealth Professional / Reuters — builder. Neutral on the methodology gap; focused on market impact.
• The Standard HK (Reuters wire) — skeptic. Confirms the methodology explanation: OpenAI excludes revenue via AWS and Google Cloud, while Anthropic includes it.

The methodology gap, explained

The two numbers describe the same underlying business. They're different because OpenAI and Anthropic count revenue differently.

Anthropic includes API revenue that flows through cloud partners — if you access Claude through Amazon Bedrock or Google Cloud, that revenue shows up in Anthropic's figures. OpenAI does not count the equivalent revenue from ChatGPT's presence on Azure. So when investors tried to do a direct apples-to-apples comparison of the two companies, they took Anthropic's figures — which include cloud-partner revenue — and applied that methodology to OpenAI's pipeline. That produced the $70B. OpenAI's own $50B figure excludes it.

Neither number is the final revenue. Both are run rates — one month multiplied by twelve — which is a useful proxy and also a way of presenting a number that hasn't happened yet.

What's real

• The $50B number is probably accurate on its own terms. Reuters confirmed it through a separate source.
• OpenAI is guiding to $70 billion by year-end — which implies it expects to close most of that gap in Q4 through direct-channel growth.
• Both numbers represent genuinely extraordinary revenue trajectories. Two years ago neither company had meaningful revenue. The question isn't whether the growth is real; it's whether the valuations priced in more growth than actually showed up.
• Anthropic's figures are heading into IPO due diligence too, and whatever methodology they've been using to report to investors will eventually face an auditor.

What deserves a side-eye

• Run rates are constructed numbers. Multiplying one month's revenue by twelve is useful and also misleading — it ignores seasonality, deal timing, and whether the pace holds. Pre-IPO companies use them because they make the trajectory legible. Investors should read them as directional.
• The $20B gap caused a 1.25% Nasdaq drop because investors had been using the $70B figure in AI infrastructure spending models. That's not irrational — it means real dollars of projected chip demand and data center buildout were priced on a $20B error. When the error corrected, so did the stocks.
• The cloud-partner exclusion is a real structural difference between the two companies, not just a rounding choice. OpenAI's IPO prospectus will need to clarify which number it claims as revenue. That answer will matter.

Samwise's take

What builders need to know

• This doesn't change pricing or access for you today. The revenue reporting discussion is pre-IPO finance. Your API bills are not going up because of this story.
• The IPO timelines are real signals. Anthropic's November target and OpenAI's year-end target mean both companies are optimizing for metrics and optics right now. Watch how they package developer-facing pricing and enterprise deals over the next 6 weeks — some of those decisions will be IPO math, not product strategy.
• Anthropic's $100B guidance is more aggressive. Going from $65B in July to $100B by December requires roughly a 50% increase in four months. That's possible — enterprise contracts close in big lumps — but it's a stretch target. If they hit it, the revenue story dominates the IPO narrative. If they miss, the story is the gap.
• The methodology question will get resolved at IPO. When the S-1 goes final and an auditor signs off, we'll have the actual number. Neither $50B nor $70B is that number yet.

Further reading

• The Next Web — OpenAI revenue $70B year-end target — covers the FT report and the $70B guidance
• Axios — "OpenAI's annual recurring revenue nears $70B" — the September 29 scoop that set investor expectations
• The Standard HK — Reuters wire — methodology explanation
• Wealth Professional — Nasdaq selloff context — market impact and sector moves