Cheaper Than Open Source: OpenAI Pitches AI for Chip Design and Claims It Undercuts China on Price

For most of the past two years, the pricing conversation in AI ran in one direction: Chinese open-source models were the cheap option, and American labs charged a premium. At Goldman Sachs’ Communacopia + Technology conference on September 9, OpenAI CFO Sarah Friar tried to flip that narrative on its head — arguing that OpenAI’s models, after a brutal round of price cuts, can now be cheaper to actually run than “free” open-source alternatives, and that the company is steering its enterprise push into specialized verticals where that argument lands hardest: semiconductor design and life sciences.

What Friar Said

The headline claim is economic, not technical. OpenAI recently cut the price of its lower-cost Luna model by 80%, and Friar said that cut helped drive a roughly tenfold increase in usage. Her argument to the audience of investors and telecom and tech executives: when you account for the full cost of running an open-source model — the cloud GPUs you rent to serve it, the engineering time to operate it — deploying OpenAI’s API can come out ahead of running Chinese open-source alternatives through cloud providers.

That is a direct attack on the value proposition of models like DeepSeek’s family and other open-weight releases that have gained ground with American enterprises this year. CNBC reported in July that open-source Chinese models could be 60% to 90% cheaper than leading OpenAI and Anthropic offerings, according to industry buyers — a gap that forced the labs’ hands. The July 30 Luna cut brought pricing down to roughly $0.20 per million input tokens and $1.20 per million output tokens, deliberately sliding under DeepSeek on input cost at the time.

Friar backed the pitch with fresh growth numbers. Enterprise revenue increased 32% from June to July, she said, outpacing the 20% growth in OpenAI’s overall annualized revenue during the same period. Codex, the company’s coding product, now counts 25 million users. The message: price cuts are not a margin surrender but a demand unlock — the usage surge more than compensates on volume.

Why Chip Design and Life Sciences

The vertical targeting is the strategically interesting part. OpenAI is expanding beyond general-purpose chatbots and coding assistants into specialized industrial work, and Friar named chip design and life sciences as the beachheads.

Chip design is a telling choice. It is a domain with enormous economic value per solved problem, a chronic engineering-talent shortage, and — conveniently for OpenAI — a customer base that already spends billions on simulation and EDA tooling. AI-assisted design work could shorten development cycles and cut the cost of producing new silicon, a pitch that lands with the same semiconductor firms OpenAI otherwise negotiates with as suppliers. The company has confirmed it is working with Samsung on next-generation memory chips for AI systems; selling design intelligence into the chip industry while buying chips from it makes OpenAI simultaneously a customer, a partner, and now a vendor to the same ecosystem.

Life sciences is the other natural vertical: regulated enough that buyers pay for reliability and accountability, data-rich enough that frontier models have real advantages over smaller open-weight alternatives, and already validated by AI-driven drug discovery programs that have produced clinical candidates this year.

The Open-Source Pincer

To understand why OpenAI is making this argument now, look at the competitive landscape. A March 2026 report from the U.S.-China Economic and Security Review Commission described China’s open AI strategy as reinforcing its industrial dominance — open-weight models as a deliberate instrument to seed Chinese AI stacks globally. American enterprises, facing ballooning AI budgets, have been willing to mix them in: multi-provider routing that sends cheap traffic to Chinese open-source models and hard tasks to frontier labs became standard practice at cost-conscious startups over the summer.

OpenAI’s counter has three prongs:

1. Absolute price cuts — the 80% Luna reduction, plus similar cuts across the GPT-5.6 family, collapsing the gap that made mixing attractive.
2. Total-cost-of-ownership framing — Friar’s argument that self-hosted “free” models carry hidden serving costs on rented cloud GPUs.
3. Vertical lock-in — specialized deployments in chip design and life sciences where switching costs are high and the work is too demanding for budget models.

The risk in this strategy is the classic one: a price war against competitors whose marginal cost is near zero. Open-weight models don’t need to recover training costs from every inference call — the entities releasing them have industrial and strategic reasons to keep them cheap. If Chinese labs respond with further cuts, OpenAI is fighting the land war it explicitly tried to avoid for two years.

Enterprise Is Carrying the Business

The June-to-July numbers Friar cited continue a trend that has been building all year: business customers are now the growth engine. Enterprise revenue growing 32% month-over-month — 1.6x the pace of overall annualized revenue — suggests the ChatGPT consumer story, while enormous, is maturing relative to API and seat-based business sales.

That reframes what the Luna price cut was for. A 10x usage surge on a model priced 80% lower means OpenAI deliberately traded unit economics for volume and data flow — the same playbook cloud providers ran for a decade. Each enterprise workload that moves onto Luna is a workload that is not validating a Chinese open-source alternative, not generating telemetry for a competitor, and not building tooling around someone else’s ecosystem.

What to Watch

Three things will determine whether Friar’s conference pitch ages well. First, whether OpenAI discloses what the post-cut Luna margins actually look like when the next financial details leak — a 10x usage surge on 20% of the price means revenue per workload fell dramatically, and the math only works if infrastructure costs keep falling in step. Second, whether named chip-design or life-sciences customers emerge; vertical pitches are easy to announce and hard to land, and the semiconductor industry moves on multi-year cycles. Third, how DeepSeek, Alibaba, and the other open-weight players respond on pricing — because the moment the cost advantage flips back, the enterprises that came for the price will leave for it too.

For now, though, the framing shift itself is the news. The lab that spent 2024 and 2025 defending premium pricing has started arguing it is the budget option — and it has the usage graph to make the case.