OpenAI unveiled custom AI inference chip ('Jalapeño') targeting 50% cost reduction versus Nvidia GPUs; GPU rental prices fall 31%
GPU rental prices have plunged 31% in just three weeks as the AI industry shifts from training massive models to deploying them—a transition known as inference. This pivot threatens Nvidia's dominance: inference is expected to account for two-thirds of global AI compute by 2026, a market where custom chips hold significant advantages over standard graphics processors. The stock closed the week at €168.80 to €168.94, representing a 16% to 17% decline from its May record high and reflecting a 7% weekly loss that has shifted investor sentiment from optimism to caution.
The most direct challenge emerged on June 24, when OpenAI unveiled Jalapeño, its first proprietary chip, developed in partnership with Broadcom and engineered specifically for inference. According to Broadcom CEO Hock Tan, such custom designs can reduce costs by approximately 50% compared to off-the-shelf GPUs. OpenAI designed the chip in just nine months, leveraging its own AI models to accelerate development—a self-reinforcing cycle where AI accelerates AI hardware creation. Notably, Nvidia's major customers are not waiting. Google, Amazon, Microsoft, and Meta are all developing proprietary chips, transforming today's largest Nvidia customers into potential rivals.
Geopolitical headwinds compound these competitive pressures. China once contributed a fifth of Nvidia's data-center revenue, but strict export controls have largely eliminated that business, allowing local rivals like Huawei to gain ground. Meanwhile, Chinese AI models such as GLM-5.2 from Z.ai are matching American benchmarks at substantially lower costs, intensifying software competition and forcing Nvidia to reconsider how its Blackwell architecture is deployed.
Yet significant industrial opportunities provide a counterweight. Nvidia is pivoting toward specialized AI factories. Roche, the pharmaceutical giant, is deploying over 3,500 Blackwell chips to accelerate drug discovery, targeting a 25% improvement in how quickly molecules advance through research. In Boden, Sweden, a massive data center is under construction to lease up to 10,000 Nvidia chips to state and industrial clients. The company's order backlog stands at €119 billion, providing substantial buffer against market volatility. Nvidia's technology powers more than 400 of the world's 500 fastest supercomputers, and nine in ten new systems in the latest ISC High Performance ranking rely on it. At the Hamburg conference, Nvidia positioned its new Vera-Rubin system as a scientific platform, while 35 new AI supercomputers are under construction across 23 European countries—a market segment that custom chips from cloud giants cannot easily serve.
Financially, Nvidia remains robust. Quarterly revenue reached €81.6 billion, the company launched an €80 billion share buyback program, and instituted a quarterly dividend of $0.25 per share. Yet year-to-date gains range from a modest 5% to 27% depending on the measurement method, reflecting investor uncertainty. The average analyst price target hovers around €261, implying roughly 55% upside from current levels—assumptions that rest on Nvidia's software ecosystem retaining customers and demand remaining substantial enough for both custom chips and GPUs.
Technical indicators paint a mixed picture. The stock is trading barely 3% above its 200-day moving average of €163.66, a critical support level. The 100-day moving average at €168.66 is also being tested, and the relative strength index at 38.3 signals an almost oversold condition.
Nvidia faces two competing forces. The first is a contracting GPU rental market and the proliferation of cheaper custom alternatives, exemplified by OpenAI's Jalapeño. The second is a deepening industrial pivot toward AI factories that could cement Nvidia's role as the foundation of scientific and enterprise computing. The company's competitive moat shows visible cracks—not collapse, but clear fractures as customers begin pursuing alternatives. How deep those excavations run will determine Nvidia's trajectory.