OpenAI Slashes Chip Dependency While NVDA Stock Climbs — What's Going On?
Nvidia finds itself in an unusual position: two of the most significant threats to its dominance emerged within weeks of each other, yet its stock continues to march higher. First, China demonstrated it could train a massive AI model without a single Nvidia GPU. Now, OpenAI — one of Nvidia's most important customers — has dramatically reduced how many of those chips it actually needs. And still, NVDA shares rose.
This apparent contradiction is worth unpacking.
OpenAI Attacks Costs from Two Directions
The efficiency push is happening on parallel tracks. On the software side, engineers at OpenAI have developed new optimization techniques that cut inference costs by more than half, according to a report from The Information. The company has not disclosed the technical specifics, but the impact is concrete: fewer Nvidia chips are required to handle the same volume of ChatGPT traffic. The savings could also enable OpenAI to reduce prices for users or expand usage thresholds.
On the hardware side, OpenAI made a more dramatic move. On June 24, the company joined forces with Broadcom to unveil a custom AI chip called Jalapeño — OpenAI's first proprietary silicon. Built specifically for large language model workloads powering ChatGPT, Codex, the API, and future agentic systems, the chip reportedly delivers far better performance per watt than current leading solutions. Notably, it was designed in just nine months.
Deployment is set to begin at gigawatt scale before the end of 2026, with Microsoft serving as the primary partner. Nvidia, however, still handles the majority of OpenAI's inference workloads for now.
A Broader Industry Shift Toward Custom Silicon
OpenAI is not pioneering this path alone. Google has been developing tensor processing units since 2016. Amazon followed with its own custom accelerators. According to research firm TrendForce, ASIC-based systems are projected to account for 27.8% of AI server shipments in 2026 — the highest share since 2023 — and custom chip growth is expected to outpace Nvidia's GPU shipments for the first time.
Broadcom and Marvell have emerged as key manufacturers in this custom silicon wave, securing partnerships with major hyperscalers looking to reduce their dependence on third-party GPU suppliers.
Geopolitical pressure is accelerating the same trend in China. Chinese tech company Meituan recently trained its LongCat-2.0 model — a 1.6 trillion parameter system — entirely on domestic chips, with no Nvidia hardware involved.
So Why Is Nvidia Stock Up?
Despite these headwinds, Nvidia shares gained nearly 2% on June 30, with the company sitting near a $4.8 trillion market valuation. The reason: the numbers remain extraordinary. In its most recent quarter, Nvidia reported data-center revenue of $62.3 billion — a 75% increase and a new all-time record.
Critically, most of the competitive pressure targets inference, not training. Nvidia still dominates the training segment, where its CUDA software ecosystem has been the default development environment since 2006. Custom chips, while increasingly capable, rarely offer the same breadth of flexibility or developer support.
Nvidia is also working to protect its position in inference. At its GTC conference, the company revealed that its upcoming Rubin platform will reduce inference cost per token by up to ten times compared to its current Blackwell architecture. Lower inference costs historically drive higher usage volumes, which in turn boosts total compute demand.
Not all investors are fully persuaded. Some have shifted capital into rival chip stocks, betting that the inference transition will compound over time and erode Nvidia's lead more severely. But Nvidia issued its latest guidance without factoring in any revenue from China — and still projected record demand.
The bottom line is straightforward: Nvidia continues to sell every chip it can produce. The real question is whether its largest customers can reduce their dependency fast enough to outrun the explosive growth of the AI market itself.
