OpenAI Trims Raise to $1.25T Valuation, Adds UAE Compute

OpenAI is close to closing a round of up to USD 30 billion at a pre money valuation of USD 1.25 trillion, according to reports on 23 October, led jointly by Abu Dhabi investors and the venture firm Coatue. The company told investors it aims to have signatures by the end of next week, though the final size is still under discussion. The figure is down from the USD 1.4 trillion that Morgan Stanley and Dragoneer had pitched to private wealth clients earlier in the month, after demand came in softer than in previous rounds.

The context is a funding race between two labs heading for public markets. OpenAI filed confidentially with the SEC in June, its chief financial officer has told staff a listing could come in 2027, and reporting now points to the first half of next year. Anthropic is pushing for a November debut. Investors told the Financial Times they would rather see how Anthropic prices before committing to its competitor, which is why OpenAI wants commitments signed first. The new round follows USD 122 billion raised in March at an USD 852 billion valuation, and it comes with a side agreement under negotiation to secure additional compute from the United Arab Emirates.

That compute clause is the part buyers should track. OpenAI told investors in February that it targets roughly USD 600 billion of total compute spending by 2030. A valuation reset from USD 1.4 trillion to USD 1.25 trillion does not cancel that plan, but it does change the internal rate of return every purchase order has to clear, and procurement teams at labs facing a down round tend to negotiate harder on unit economics.

The product side moved in the opposite direction on 23 October. OpenAI made its Decisions API generally available, built on GPT-6 Luna, positioning it for near real time decisions on text, image and structured input at up to ten times the speed of calling the same model through the Responses API. Pricing is input tokens only, with output tokens free, and a single request can carry up to 400 questions, double the previous limit. Cheap output tokens change the shape of inference demand: more calls, shorter answers, and a heavier load on the interconnect and power delivery around the accelerators rather than on memory bandwidth alone.

For component buyers, three implications. Inference optimised silicon keeps pulling advanced packaging, high speed memory and power stages. Regional compute deals, such as the UAE arrangement, shift hardware demand into new geographies with their own qualification and logistics requirements. And a down round is a warning that the demand curve at the top of the chain is financed by capital markets, not by retained earnings, so contract terms matter as much as forecasts. Our component catalog and quote desk and PCBA assembly service teams can support qualification and alternate sourcing as those programmes move.