Compute demand surges
AI compute demand is ratcheting up: CoreWeave said it expanded business with Meta under a reported $21 billion agreement while Anthropic added workloads, driving the provider's stock higher. Market forecasts also expect semiconductor spending to hit roughly $1.3 trillion in 2026, and industry releases warn memory supply strains are reshaping electronics manufacturing. (benzinga.com; finance.yahoo.com; globenewswire.com)
CoreWeave said on April 9 that Meta expanded its artificial intelligence cloud deal to about $21 billion through December 2032, then added Anthropic as another multiyear customer a day later. (coreweave.com) CoreWeave said Meta will use the added capacity for artificial intelligence inference, the step where trained models answer prompts, and that some deployments will use Nvidia’s Vera Rubin platform across multiple sites. (investors.coreweave.com) On April 10, CoreWeave said Anthropic signed a multiyear agreement to run and develop Claude models on its cloud, with compute scheduled to come online later in 2026. (investors.coreweave.com) Investors pushed CoreWeave shares higher after the announcements. The stock closed at $117.20 on April 14, up 6.28% for the day, and traded around $117.25 early on April 15. (finance.yahoo.com; google.com) The rush for cloud contracts sits on top of a larger chip buildout. Gartner said on April 8 that global semiconductor revenue is projected to exceed $1.3 trillion in 2026, up 64% from 2025 and the fastest growth in more than two decades. (gartner.com) Gartner said dynamic random-access memory, the short-term working memory inside servers and devices, is set for the sharpest jump: prices are forecast to rise 125% in 2026, with storage tightness extending into 2027. (gartner.com) A separate April 13 industry release said demand for high-bandwidth memory, the fast memory stacked next to artificial intelligence chips, is pulling factory capacity away from conventional memory used in phones, personal computers, cars, and industrial gear. (globenewswire.com) That helps explain why cloud providers are signing long contracts now. The bottleneck is no longer only the graphics processor unit itself, but also the memory and networking parts needed to keep large artificial intelligence systems running at scale. (gartner.com; globenewswire.com) CoreWeave’s filings show how concentrated that spending has become: one contract can stretch to 2032 and run into the tens of billions of dollars as model makers lock up future capacity before the next chip cycle tightens further. (sec.gov)
