Meta is reducing its AI spend, particularly on inference compute, as token costs strain budgets.

Meta is reducing its AI spend, particularly on inference compute, as token costs strain budgets. According to a Digitimes report (https://www.digitimes.com/news/a20260617PD230/meta-ai-cost-business-2026.html), Meta has put the brakes on AI usage due to rising token costs, suggesting a pivot to cost optimization over scale. Separately, a deep-dive from The Pragmatic Engineer (https://newsletter.pragmaticengineer.com/p/why-is-meta-destroying-its-engineering) details how Meta’s engineering organization is being disrupted under an AI-fueled leadership push, which could slow deployment and operational efficiency. An FT article (https://www.ft.com/content/c9523494-db86-4e9d-858a-d7f39d5bde45) notes Meta is opening the door to Wall Street through Dina Powell McCormick, possibly to secure capital or restructure amid mounting AI costs. Taken together, these signals indicate that Meta is tempering its AI infrastructure ambitions, potentially reducing demand for GPUs, networking gear, and foundry services from key suppliers such as NVIDIA, AMD, TSMC, Broadcom, and Marvell. The pullback is likely to manifest in lower procurement volumes or delayed data center buildouts over the next 45 days.