OpenAI investors question $852b valuation as strategy shifts

Some OpenAI investors question its US$852 billion valuation as it shifts to enterprise and prepares a possible IPO this year, the Financial Times reported.

OpenAI raised US$122 billion last month, but it changed its product roadmap twice in the past six months.

Some backers said the moves could weaken its position against Anthropic and a resurgent Google.

One early backer criticized the company as unfocused for emphasizing enterprise and coding products despite ChatGPT’s scale, while OpenAI CFO Sarah Friar said claims that investors are not supportive of the company’s strategy do not match the facts.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Anthropic’s enterprise approach brings in more revenue per user than OpenAI

A comparison estimates Anthropic earns about $211 per monthly user, while OpenAI brings in about $25 per weekly user 1.
The gap ties back to business mix. The cited analysis describes Anthropic as aiming at enterprise buyers and application programming interface (API) sales, with 80% of revenue coming from business customers 1.
That setup lets Anthropic reach about 40% of OpenAI’s revenue scale with only about 5% of OpenAI’s consumer user base 1.

AI costs are pushing software companies to rethink pricing

High computing bills are squeezing profit margins for AI applications across the industry 2.
Microsoft reportedly lost an average of more than $20 per month per GitHub Copilot user at a $10-per-month price point, while many fast-growing AI startups run negative gross margins 2.
To stay afloat, software firms are moving away from flat-rate software-as-a-service (software as a service (SaaS)) plans toward usage-based pricing or blended models 2.
Some teams are cutting dependence on outside model providers. Cursor, a coding assistant startup, is described as building specialized AI models to lower third-party API costs, with one analysis projecting gross margins rising from about 74% to 85% by 2027 as it shifts to a mix of open-source and proprietary models 2.