Overview
- Nvidia, which reports fiscal first‑quarter results on Wednesday, is being treated as a gauge for AI chip demand, cloud spending, and the health of the data‑center buildout.
- U.S. officials cleared about 10 Chinese firms to buy H200 chips with a 750,000‑unit ceiling, yet Beijing has not approved purchases and no H200s have shipped, leaving the license as upside rather than baseline demand.
- Big Tech’s 2026 AI spending has swelled to about $725 billion, and Nvidia’s supply commitments jumped to $95.2 billion late last fiscal year, signaling confidence in multi‑quarter orders from hyperscalers and sovereign buyers.
- Deployment is running into real‑world limits, with U.S. grid connections often taking 36 to 84 months and high‑voltage transformer lead times stretching to four years; some operators are adding on‑site generation that costs more per megawatt‑hour to keep projects on schedule.
- Nvidia’s dominance in training faces a tougher inference market as Google’s TPUs, Amazon’s Trainium, and new AMD parts target lower‑cost, always‑on workloads, pushing Nvidia to add inference‑focused systems after acquiring technology from Groq.