The Spiraling Cost of Making AI

Published
Score
10

Why it matters

Core event: OpenAI and Anthropic are projected to spend nearly $65 billion combined in 2026 on training and operating AI models, exemplifying the spiraling costs of AI development amid explosive infrastructure demands.[8] This ties into Anthropic's April 6-7 announcements of expanded deals with Broadcom and Google, securing 3.5 gigawatts of TPU-based compute capacity starting 2027 to fuel growth, plus Broadcom's long-term supply of custom AI chips to Google through 2031.[3][5]

Key players: Anthropic (AI startup, $30B revenue pace in 2026, seeking business users), OpenAI, Broadcom (chip supplier), Google (providing TPUs and infrastructure), with broader involvement from hyperscalers driving $3-4T cumulative AI capex by decade's end; IDC notes $82B AI compute spend in Q2 2025 alone.[2][3][5]

Context and timeline: AI costs have surged—enterprise compute spending up 166% YoY in Q2 2025, budgets doubling to 1.7% of revenue in 2026 despite 80-85% forecast misses—fueled by hyperscaler data centers ($6.7T needed total) and model training.[2] Model prices fell (e.g., Claude Opus 4.6 67% cheaper), but infrastructure opacity and CEO-led decisions amplify overruns; Anthropic's deals address its "unprecedented growth" from $9B run-rate end-2025.[2][4][5] NVIDIA reports 86% of firms raising AI budgets, prioritizing infrastructure.[1]

Newsworthy now: Fresh April 6-7 deals highlight cost escalation as AI hits maturity—$500B global spend projected 2026—while 80% of enterprises face budget risks without controls, contrasting ROI gains (87% cost reductions) and making infrastructure races a market flashpoint.[1][2][9] Broadcom shares rose 2-6% post-announcement, signaling investor focus.[3]

Sources

mail

Get notified about new Artificial Intelligence developments

Primary sources. No fluff. Straight to your inbox.

See more entries tagged Artificial Intelligence.