The AI bubble isn't just inflating valuations and destroying jobs. It's physically consuming the components that the rest of the technology industry needs to exist. Khein-Seng Pua, CEO of Phison — one of the world's largest flash storage controller manufacturers — issued a warning in February 2026 that reads less like market analysis and more like a casualty report: many consumer electronics manufacturers will go bankrupt or exit product lines by the end of this year because AI data centers are eating all the memory chips.
The math is brutal. A single Nvidia B300 GPU consumes 1,400 watts and requires massive amounts of high-bandwidth memory. Multiply that across the tens of millions of units that data centers are ordering, and you get an industry that is physically redirecting the world's semiconductor output away from phones, laptops, TVs, and consumer SSDs toward AI training infrastructure. Kioxia, a major NAND manufacturer, has confirmed that its entire 2026 production volume is already sold out. Not allocated. Sold.
Memory manufacturers aren't just rationing supply — they're changing the economics entirely. According to Pua, manufacturers are now demanding three years of prepayment, a requirement unprecedented in the electronics industry. Only companies with massive cash reserves — Apple, Samsung, the hyperscalers — can afford that. Everyone else is locked out.
Pua's forecast is specific: mobile phone production will be reduced by 200-250 million units globally. PC and TV production will see significant cuts. The companies that survive will be the ones with enough cash to prepay for components years in advance — Apple, Samsung, the handful of manufacturers that own their own fabrication. The companies that don't survive will be the mid-tier brands, the budget device makers, the independent SSD brands that don't own fabs and can't win bidding wars against data centers willing to pay any price for memory.
This is where the AI bubble thesis meets physical reality. The companies building AI infrastructure are so desperate for compute and memory that they're creating scarcity for everyone else. That scarcity drives up costs for every technology company, which drives up prices for consumers, which reduces demand, which kills the companies that can't absorb the margin compression. The AI boom doesn't just risk a financial correction — it's actively destabilizing the supply chains that the broader technology economy depends on.
New fabrication capacity won't save anyone fast enough. Samsung, Micron, SK Hynix, and Kioxia are all investing, but new fabs take at least two years from announcement to production. China's new capacity will account for only 3-5% of global supply in the initial stage. Equipment manufacturers can't keep up with demand. The bottleneck isn't just chips — it's the machines that make the chips.
If you're building a company that depends on hardware availability — IoT devices, consumer electronics, any physical product with memory in it — your supply chain risk just became existential. If you're building software, the lesson is different but equally urgent: the infrastructure you're building on top of is getting more expensive, not less. Cloud costs are rising because the providers are competing for the same scarce memory and compute. That cost gets passed to you.
The companies raising hundreds of billions to build AI aren't just competing with each other. They're competing with every other technology company for the physical resources required to operate. When AI eats all the memory chips, everyone else starves. The only question is who runs out of runway first — the VC-backed unicorns burning cash, or the consumer electronics companies that can't source components at any price.