Nvidia Claims $1T in AI Orders; Hyperscalers Are Building Around It

Nvidia sees $1 trillion in AI orders through 2027; hyperscalers are building custom chips to dodge that price tag.

🖥️ Jensen Huang: "The Inference Inflection Has Arrived" — $1 Trillion by 2027

Decoded: Nvidia CEO Jensen Huang delivered his GTC 2026 keynote Monday in San Jose, declaring "at least $1 trillion" in orders for Blackwell and Vera Rubin systems through 2027 — doubling the $500 billion demand forecast cited on Nvidia's February earnings call. Huang declared that inference has now surpassed training as the primary AI workload, and unveiled new software for orchestrating AI inference clusters at scale. (CNBC, Reuters, March 17)

Why it matters: The $1 trillion forecast directly counters analyst concern that AI capex would plateau in 2026. Nvidia is signaling demand is accelerating — and that inference, not training, is where the next leg of GPU revenue lives. The Vera Rubin platform is positioned as Nvidia's inference-first flagship for the next hardware cycle.

🖥️ TrendForce: CSPs Are Investing in Custom ASICs — And Nvidia Is Responding

Decoded: A TrendForce report published March 18 found that major cloud service providers — including Google, Microsoft, Amazon, and Meta — are increasing investment in self-developed AI chips to reduce reliance on Nvidia's GPU stack. In direct response, Nvidia shifted GTC 2026 messaging toward AI inference rather than training — the workload where its hardware holds the largest performance lead over custom silicon. (TrendForce, March 18)

Why it matters: Every hyperscaler building its own chip — Google TPU, AWS Trainium, Microsoft Maia — buys fewer Nvidia GPUs at full price. TrendForce's finding shows Nvidia is consciously repositioning its GTC narrative around inference to defend against that structural pressure. Jensen Huang's $1 trillion forecast is real — but so is the long-term erosion of Nvidia's training-market dominance.

That's your Wednesday signal. See you tomorrow.

— The AI Decoded Team