Money flows into chip bets
Investors and cloud providers are directing cash toward silicon as the AI build-out matures, not just toward model makers. One market piece says the semiconductor index hit a record high as AI industrialisation lifts valuations, and reporting notes Amazon’s in‑house silicon business has passed about a $20 billion annual run rate. ( )
Wall Street spent 2023 and 2024 chasing the companies that made chatbots. In April 2026, money is piling into the companies that make the chips, package the chips, and rent the chips by the hour, with the Philadelphia Semiconductor Index closing at 8,889.69 on April 10 after touching 8,962.95 intraday near a record high. (indexes.nasdaq.com) That index is a scoreboard for the hardware layer of artificial intelligence: the silicon that trains models, the memory that feeds them, and the factories that assemble them. When that basket rises more than 123% over one year, investors are saying the bottleneck has shifted from ideas to equipment. (indexes.nasdaq.com) Amazon put a hard number on that shift on April 9. Chief executive Andy Jassy wrote that Amazon’s chips business, including Graviton, Trainium, and Nitro, is now running at more than $20 billion a year and growing at triple-digit rates year over year. (aboutamazon.com) That is not Amazon selling chips in stores. It is Amazon Web Services building its own processors, putting them inside its cloud data centers, and charging customers for the computing time those processors deliver. (aboutamazon.com) Graviton is Amazon’s general cloud processor, the chip that handles everyday computing jobs like web servers and databases. Amazon says it has launched five generations since 2018, and the newest Graviton5 has 192 cores and delivers up to 25% better performance than Graviton4. (aboutamazon.com) Trainium is the heavier machine for artificial intelligence, built to train and run large models at lower cost than standard graphics-processing-unit rentals. Amazon says the Trainium family now includes Trainium1, Trainium2, and Trainium3, with Trainium2 already in service. (aws.amazon.com) The point of owning both kinds of chips is control. If Amazon designs the processor, the server, and the rental price, it can squeeze more work out of each data center and keep more of the economics inside Amazon Web Services. (aboutamazon.com) Jassy’s letter showed how tight demand has become. He wrote that two large customers asked whether they could buy all available Graviton capacity in 2026, which means the constraint is no longer just “who has the best model” but “who can get enough machines.” (aboutamazon.com; datacenterdynamics.com) That is why semiconductor stocks are rising together instead of only one famous name rising at a time. The artificial-intelligence build-out now looks more like railroads or electric grids, where the winners include the companies selling picks, shovels, transformers, and track. (indexes.nasdaq.com; aboutamazon.com) The choke point is not only chip design. Advanced packaging, which is the step that stacks memory next to processors and wires them together at very short distances, remains scarce enough that DigiTimes reported on April 10 that global capacity is in severe shortage and Nvidia has reserved most of Taiwan Semiconductor Manufacturing Company’s leading CoWoS packaging supply. (digitimes.com) So the market is repricing the whole supply chain at once. Amazon’s $20 billion chip run rate says cloud companies are becoming chip companies, and the semiconductor index near record highs says investors think the next phase of artificial intelligence will be built less like an app launch and more like an industrial expansion. (aboutamazon.com; indexes.nasdaq.com)
