The heavyweight comes into focus
All eyes turn to Taiwan this week. TSMC, the world's largest contract chipmaker, reports earnings on July 16, and the read-through stretches far beyond Taipei. Ahead of the call, Citi raised its price target on the stock, a vote of confidence tied squarely to the AI chip crunch. TSMC isn't waiting to celebrate — the company is planning three new advanced chip-packaging fabs in Chiayi, a homegrown expansion aimed directly at surging AI demand. Advanced packaging, the art of stitching chips together into ever-denser modules, has become as strategically vital as the transistors themselves.
The capacity scramble runs deeper still. Memory chipmakers are racing to feed what one account calls AI's 'almost unlimited' appetite, with Korea's memory giants pushing hard to expand output. When the industry's memory and foundry leaders are both sprinting to build, it's a signal that demand isn't cooling.
The Nvidia question
Nvidia's dominance is the subplot in nearly every story today — and the bull and bear cases are both getting louder. On one side, AI search company Perplexity picked Nvidia over AMD to power its AI coding stack, a choice financial commentators read as a message larger than a single purchasing decision. Data-center firm IREN is teaming with BE Networks to test-drive Nvidia's next-generation Blackwell infrastructure before it goes live, another datapoint in Nvidia's favor.
The counterweight comes from Mountain View. A widely syndicated analysis argues that Google's homegrown AI chips pose a bigger threat to Nvidia than Wall Street currently assumes — the kind of in-house silicon push that could quietly erode the incumbent's grip.
Everybody wants their own chip
That custom-silicon impulse is spreading. Bloomberg's Mark Gurman reports that Apple is rewriting the roadmap for the chips inside its Macs, with AI as the driving force — and, intriguingly, lessons salvaged from its abandoned self-driving car project. Samsung, meanwhile, is reportedly building a dedicated AI accelerator for PCs, codenamed Gaia and led by its LSI division, that could show up in machines around 2027. The message: the AI chip is becoming a first-class citizen inside consumer devices, not just data centers.
The hard physics problem
As chips shrink and stack, they cook themselves — and a cluster of new academic and government research, highlighted by Semiconductor Engineering, is racing to master the heat. Thermal management is quietly becoming a limiting factor for how far the industry can push.
Quantum steps closer
Quantum computing is drawing fresh attention as signals mount that practical machines may arrive sooner than expected. A standout: researchers led by the National University of Singapore demonstrated a photonic quantum processor that runs at room temperature — no deep freeze required — and is compatible with standard CMOS manufacturing. A quantum chip that fits existing fabs is exactly the kind of pragmatism the field has been missing.
And, for the tinkerers
A smaller but heartening note: developers are porting Nvidia's GPU driver to the Haiku operating system, finally bringing hardware-accelerated 3D graphics to the long-underserved platform. Proof that even amid the AI gold rush, the community keeps closing the everyday gaps.