OpenAI Moves Into Chip Design and Outcome-Based Pricing as Enterprise Competition Intensifies

What is OpenAI actually selling now?

The answer is no longer just chatbots. OpenAI is pushing its AI capabilities into highly specialised industrial verticals — chip design, life sciences, financial services — while simultaneously restructuring how it charges for them. Chief Financial Officer Sarah Friar made this clear at Goldman Sachs’ Communacopia + Technology Conference in San Francisco on Monday, framing the strategy as a direct response to enterprise customers who want measurable returns, not just access to a model.

This is a meaningful pivot. Selling API access to a general-purpose language model is one business. Selling AI embedded in a semiconductor design workflow, or a drug discovery pipeline, is quite another. It demands domain credibility, integration depth, and a pricing logic that aligns with how those industries actually measure value. OpenAI is now attempting all three simultaneously.

Why chip design specifically?

OpenAI has skin in this game. Friar disclosed that the company used its own models to develop its internal chip, codenamed Jalapeno, which reached tape-out — the stage at which a design is finalised and dispatched to a fabrication facility — within nine months. That timeline matters. Chip development cycles are notoriously long and expensive, and compressing them is a commercially compelling proof point.

This is not purely a marketing story. Using your own product to build your own infrastructure, and then citing the outcome as a sales case, reflects a particular kind of institutional confidence. It also signals that OpenAI sees the semiconductor sector as a viable enterprise vertical, not merely a technical curiosity. The audience for this pitch would include fabless chip designers, EDA software firms, and the hyperscalers building custom silicon for AI workloads — a market with deep pockets and acute cost sensitivity.

How is OpenAI competing on price against open-source models?

This is where the positioning gets sharper. Open-source and open-weight models — particularly those emerging from Chinese laboratories — have long been positioned as the cost-efficient alternative to frontier proprietary models. The assumption is that self-hosting an open-weight model eliminates licensing costs. Friar challenged that assumption directly.

She stated that deploying OpenAI’s lower-cost Luna model can be cheaper than running Chinese open-source alternatives through cloud infrastructure. Her specific comparison was against Z.ai’s GLM 5.3, a capable open-weight model with growing enterprise adoption. “If you’re deploying Luna and compare that to GLM 5.3, for example, on a cloud layer, we are cheaper,” she said. The implication is that once you factor in cloud compute costs, engineering overhead, and model maintenance, the total cost of ownership for self-hosted open-source is not as low as it appears.

OpenAI has also cut Luna’s price by 80 per cent, a reduction that drove roughly a tenfold increase in usage. The elasticity of that response is significant. It suggests that price, not capability ceiling, has been the primary barrier for a substantial segment of potential enterprise users. Friar added that Codex, OpenAI’s coding-focused tool, now counts 25 million users — a figure that points to developer-layer adoption as a durable growth channel.

What does the enterprise revenue data actually show?

The numbers Friar cited are striking in their trajectory. Enterprise revenue grew 32 per cent from June to July alone, compared with 20 per cent growth in overall annualised revenue during the same period. Enterprise is accelerating faster than the business as a whole. That gap matters because enterprise contracts tend to be stickier, more predictable, and structurally more defensible than consumer subscriptions.

By mid-year, enterprise and consumer revenues had reached roughly equal footing — a balance OpenAI had originally targeted for year-end. Achieving it six months early is either a sign of genuine enterprise momentum or a reflection of consumer growth plateauing. Possibly both. The distinction matters for how one reads the underlying health of the business, and Friar did not elaborate on the consumer trajectory in detail.

What is outcome-based pricing, and why does it matter?

Friar mentioned that OpenAI is experimenting with pricing tied to business outcomes rather than token consumption or API calls. This is a structurally different commercial model. Usage-based pricing is straightforward but misaligns incentives — a customer pays whether or not the AI delivers value. Outcome-based pricing, by contrast, ties revenue to results: a successful drug candidate, a completed chip design, a closed sales deal.

Executing this model requires OpenAI to agree on what a measurable outcome looks like in each vertical, which is technically and contractually complex. It also requires the kind of domain expertise that is difficult to scale quickly. But if it works, it repositions OpenAI from infrastructure vendor to value partner — a far more defensible commercial relationship. The move reflects pressure from enterprise buyers who have grown sceptical of AI spending that cannot demonstrate a clear return. That scepticism is well-founded, and OpenAI’s willingness to price against it is a notable strategic concession to market reality.

What does this mean for the broader competitive landscape?

OpenAI faces pressure from multiple directions. Anthropic is targeting the same enterprise verticals with its Claude model family. Chinese open-weight models from DeepSeek, Z.ai, and others have disrupted the assumption that frontier capability requires frontier pricing. Meta’s Llama releases have made capable open-source infrastructure broadly accessible. Against this backdrop, OpenAI’s strategy appears to rest on three pillars: aggressive price cuts on lower-tier models to win volume, domain-specific credibility through verticals like chip design and life sciences, and a pricing innovation that could, if it scales, redefine how enterprise AI contracts are structured.

Whether the Jalapeno chip story translates into sustained semiconductor sector sales, whether Luna’s price cuts hold margin at scale, and whether outcome-based pricing survives contact with enterprise procurement processes — these remain open questions. What is clear is that OpenAI is no longer content to compete purely on model capability. It is competing on cost, on domain depth, and on commercial model design. That is a more sophisticated contest, and one where the outcome is genuinely uncertain.