GAIB (@gaib_ai): State of AI Infra – Week of May 22

State of AI Infra – Week of May 22 In Short This week’s AI infrastructure theme: from incremental cloud expansion to full industrial capex supercycle. Hyperscaler spending forecasts are being revised upward in real time, GPU demand is broadening well beyond traditional cloud buyers, and capital markets are issuing long-duration debt against contracted AI infrastructure revenues. The pattern across every story this week is consistent: compute infrastructure is now being underwritten as a permanent, large-scale economic asset, not a cyclical technology bet. Highlight Nvidia Posts Record $81.6B Quarterly Revenue as AI Infrastructure Demand Broadens Beyond Hyperscalers Nvidia reported record quarterly revenue of $81.6 billion — up 85% YoY and 20% QoQ — with the Data Center segment generating $75.2 billion alone (up 92% YoY). The quarter also marked a structural shift in how Nvidia frames its business: the company reorganized into Data Center and Edge Computing segments, with Data Center split further into Hyperscale and ACIE (AI Clouds, Industrial, and Enterprise). The reorganization formally acknowledges that demand is now coming from AI-native cloud providers, enterprise deployments, sovereign AI programs, and industrial AI, not just the big four hyperscalers. The revenue mix tells a secondary story. Networking revenue came in at $14.8 billion, up 199% YoY, nearly double the rate of compute growth. Scaling AI infrastructure is increasingly a bandwidth problem as much as a GPU problem. Meanwhile, the Edge Computing segment generated $6.4 billion (up 29% YoY) spanning robotics, AI-RAN, and autonomous systems. Why it matters: A broadening customer base, networking outpacing compute, and a new edge revenue line confirm that the AI infrastructure build-out is entering a more distributed, infrastructure-grade phase. This is not a single-customer cycle. (Source) 4 More Top Stories Moody’s Raises Hyperscaler Capex Forecast by $85B, Now Approaching $1T by 2027 Moody’s revised its forecast upward by $85 billion, now projecting ~$785B in 2026 and approaching $1 trillion by 2027 across Amazon, Microsoft, Meta, Alphabet, Oracle, and CoreWeave. The revision pace is itself the signal — only months ago the figure was ~$700B. Hyperscalers collectively added ~$700B in RPOs over the last two quarters, a leading indicator of future infrastructure demand. Moody’s also flagged rising financial pressure: capex intensity and debt levels are now large enough to materially influence balance sheet quality at the mega-cap level. (Source) Akamai Launches $2.6B Zero-Coupon Bond Offering Following Major AI Cloud Contracts Akamai issued $2.6B in convertible senior notes (2030/2032 maturities) at 0% coupon, an unusually favorable structure that follows a $1.8B / seven-year contract with Anthropic and a separate $200M CIS agreement. Zero-coupon issuance anchored by contracted AI demand is a clean example of capital markets pricing AI infrastructure as durable, investment-grade revenue. (Source) Nscale Secures $790M Financing for Norway AI Campus Anchored by Microsoft Demand Nscale secured $790M (plus a $790M accordion facility) for its 230MW Narvik campus, expected to be anchored by Microsoft with ~30,000 Nvidia Rubin GPUs. The deal continues a pattern of lenders financing large-scale AI infrastructure when hyperscaler demand visibility exists upfront. Norway’s renewable energy and cooling economics are making it a structural destination for hyperscale AI deployment. (Source) Alibaba Cloud Says AI Demand Requires 10x More Compute Than 2022 Levels Alibaba Cloud CEO Eddie Wu said the company expects to need roughly 10x its 2022 compute capacity to meet AI demand. Cloud Intelligence Group reported ~$6.0B in quarterly revenue (up 38% YoY), with AI services at ~$1.3B,  the company’s 11th consecutive quarter of triple-digit AI revenue growth. AI products are now ~30% of external revenue, with management guiding for >50% within one year. Chinese hyperscalers are entering the global infrastructure expansion cycle at scale. (Source)