OpenAI brings the heat with 'Jalapeño'

The biggest story of the day: OpenAI is officially in the chip business. The Sam Altman-led company, working with Broadcom, unveiled Jalapeño, its first in-house custom processor. It's built for AI inference — the work of actually running models to answer user queries, as opposed to training them — and it's designed to power ChatGPT and OpenAI's next generation of large language models.

Inference chips are the workhorses of deployed AI: every time someone types a prompt, an inference chip does the lifting. By designing its own, OpenAI is making a clear bid to lean less heavily on Nvidia, whose GPUs have been the default — and a costly, scarce one — across the industry. The reveal carries extra weight because it lands after a much larger arrangement with Nvidia, reported at around $100 billion, failed to come together. Building from scratch with Broadcom is OpenAI's answer to tightening supply and its own soaring compute needs.

OpenAI isn't alone. Reports describe a broader race in which tech giants — including Google, Amazon, and Microsoft alongside OpenAI — are increasingly designing the hardware that powers the AI boom themselves, rather than buying it all off the shelf. Jalapeño is the splashiest entry yet, but it's part of a structural shift: the companies driving AI want to own the silicon underneath it.

IBM breaks the 1-nanometer barrier

In the labs, IBM debuted what it calls the world's first sub-1-nanometer chip technology, announced June 25. The breakthrough centers on a 0.7-nanometer design — a milestone that, per EE Times, points toward continued gains in density and efficiency even as the industry bumps against the physical limits of how small transistors can go. It's a reminder that the frontier of chipmaking is still moving, and that the names racing to design AI processors will need ever-more-advanced manufacturing to build them.

Apple rethinks the Mac roadmap — and chases Chinese memory

Apple is reworking how it builds the brains of its highest-end Macs. According to Bloomberg's Mark Gurman, the company plans to skip the higher-end M6 Pro and Max chips entirely and leap straight to an AI-focused M7 generation, targeted for 2027. A base M6 would still arrive for entry-level Macs, but the premium tier would jump ahead — a sign of how thoroughly AI is reshaping even Apple's famously deliberate silicon cadence.

Apple is also making a quieter, more delicate move. Per the Financial Times, the company is pressing the Trump administration for permission to buy memory chips from CXMT, a Chinese manufacturer currently on a U.S. blacklist. It's a telling tension: Apple wants supply and cost relief, but the source sits squarely in the middle of U.S.–China trade restrictions, putting the iPhone maker in the position of lobbying Washington to ease its own rules.

Korea courts Nvidia

Finally, a glimpse of how badly regions want a piece of the AI hardware economy. The newly elected governor of South Korea's North Jeolla Province (Jeonbuk) is pitching Nvidia directly, hoping to lure investment to the Saemangeum development zone. Nvidia may be facing fresh competition from its own customers, but it remains the most coveted name in AI chips — and local officials are lining up to bring it home.

The throughline

Today's stories rhyme: everyone wants out from under a single supplier. OpenAI is building its way out with Jalapeño, Apple is hedging on both design and sourcing, IBM is pushing the manufacturing frontier that makes it all possible, and governments are courting the incumbent even as challengers emerge. The AI hardware map is being redrawn — and it's the companies that run the models, not just the ones that sell the chips, doing the drawing.