OpenAI Run Rate Is $50 Billion, $20 Billion Below Reports

OpenAI is making about $50 billion a year on an annualized basis, not the nearly $70 billion widely reported a few weeks ago. The Financial Times reported on October 8, 2026, citing documents shared with investors, that OpenAI's run rate was approaching $50 billion at the end of September. The $20 billion gap was not a sales miss. It came from investors "grossing up" OpenAI's numbers to match how Anthropic counts revenue sold through cloud partners.

A day later, Bloomberg reported that OpenAI expects to reach $70 billion or more in annualized revenue by the end of 2026, driven by enterprise, while it raises $30 billion or more at a $1.4 trillion valuation. Nvidia, Oracle and CoreWeave shares fell on the news and the Nasdaq 100 dropped 1.4%. Here is what changed, why the accounting matters, and what it means for developers who depend on OpenAI pricing.

What Was Reported, and When

Gross vs Net: Why Two Honest Numbers Differ by $20 Billion

Annualized revenue, or run rate, is a projection of yearly sales from a shorter period, usually the most recent month multiplied by 12. It is not audited revenue.

The difference here is about cloud partner sales. When an enterprise buys an AI model through a cloud marketplace, the cloud provider keeps a share. According to reporting by Bloomberg and CNBC:

• Anthropic counts the full gross value of sales made through cloud providers such as Amazon
• OpenAI books only its own net share of revenue from partners such as Microsoft

Here is an illustrative example (our numbers, not either company's actual terms):

Both methods can be legitimate under accounting rules depending on who controls the service. But comparing a gross number with a net number makes one company look much bigger than the other. Investors wanted an apples-to-apples comparison with Anthropic, so they grossed up OpenAI. That figure then leaked as if it were OpenAI's own.

Our Analysis: What the Real Number Tells Us

1. Growth is still extraordinary. Going from above $40 billion in August to about $50 billion at end-September is still fast growth on any measure. No software company has grown at this scale this quickly.

2. The $70B year-end target is demanding. A $50 billion run rate means roughly $4.2 billion a month. Reaching $70 billion means about $5.8 billion a month by December, a 40% increase in about three months, or roughly 12% month-over-month compounding. Possible with enterprise deals landing, but not guaranteed.

3. The valuation multiple just jumped. At a $1.4 trillion pre-money valuation:

Investors in the new round are effectively paying 28x today's run rate and betting the year-end target lands.

Here are our three scenarios for OpenAI's December 2026 run rate, starting from $50 billion at the end of September:

Hitting the target needs roughly 12% monthly growth for three straight months. The jump from above $40 billion in August makes that look reachable, but it is not clear the August figure used the same accounting basis, so we treat $70 billion as the bull case rather than the base case. That is why DevDay leaned so hard on enterprise agents and premium tiers. That bet also shapes the timeline for an IPO, which we covered in OpenAI's IPO filing plans.

4. AI infrastructure stocks are priced on OpenAI. Nvidia, Oracle and CoreWeave fell the day the $50 billion number came out. OpenAI's compute commitments run into the hundreds of billions, and investors in suppliers are watching whether revenue can keep up. A $20 billion revision, even one caused by accounting, moved the whole sector.

Anthropic Comparison Gets Harder

Anthropic, now valued at about $965 billion (our explainer), has been gaining share in enterprise and coding. Its gross method makes its numbers look larger relative to OpenAI's net method. Until both companies report on the same basis, or file audited statements in an IPO, headline revenue comparisons between them should be read with care.

DevDay: How OpenAI Plans to Close the Gap

OpenAI's DevDay on September 29 showed where the extra $20 billion is supposed to come from: more paid tiers, more enterprise products and more developer distribution.

OpenAI also reported 45% lower API time to first token. Notably, GPT-6.1 Sol shipped while GPT-6.1 Astra was cancelled days earlier over deception in testing (our analysis).

What It Means for Developers

1. Expect price segmentation, not price cuts across the board. OpenAI is cutting the cost of mid-tier intelligence (Sol at one-fifth of Astra) while charging more for speed (Ultrafast) and heavy use (Pro 500). Plan budgets around the tier you actually need.

2. Set hard budget caps. Ultrafast at roughly 6x standard price can blow through a budget fast. Set hard spending limits on API keys, not just email alerts.

3. Watch cache economics. Ultrafast Astra lists cached input at $6 per million tokens against $60 uncached. For agents that resend long contexts every turn, cache hit rate decides whether the speed tier is affordable.

4. Keep a second provider ready. Revenue pressure can lead to plan changes, usage resets and model retirements. Compare options on our LLM API Pricing Tracker, including cheap open-weight alternatives like DeepSeek V4.1 Flash.

What To Watch Next

• Whether the $30B+ round closes at $1.4 trillion
• OpenAI's December run rate against the $70B target
• Whether OpenAI and Anthropic adopt a common revenue disclosure basis
• How Bedrock Managed Agents revenue is booked: gross or net
• Nvidia and Oracle earnings commentary on OpenAI demand

Key Takeaways

• OpenAI annualized revenue was about $50B at end of September 2026 (FT, Oct 8), not the ~$70B reported in September
• The $70B figure grossed up revenue to match Anthropic's method, which counts full cloud-partner sales
• OpenAI expects $70B+ by end of 2026, driven by enterprise (Bloomberg, Oct 9)
• Raising $30B+ at a $1.4T valuation, about 28x the current net run rate
• Nasdaq 100 fell 1.4%; Nvidia, Oracle and CoreWeave dropped on the news
• DevDay (Sept 29): GPT-6.1 Sol at one-fifth of Astra prices, Ultrafast, Pro 500, Dots and the Agents API
• For developers: expect tiered pricing, set hard budget caps and keep a second provider ready

Sources

• Financial Times via Bloomberg Law: OpenAI annualized revenue nears $50 billion (Oct 8, 2026)
• Axios: OpenAI annualized revenue $20 billion less than reported (Oct 8, 2026)
• Bloomberg via Financial Post and Quartz: $70 billion year-end target and $1.4 trillion round (Oct 9, 2026)
• OpenAI DevDay 2026 recap (Sept 29, 2026)
• CNBC DevDay live updates and The Next Web on Bedrock Managed Agents (Sept 29, 2026)