Analysts warn packaging and power‑supply bottlenecks could constrain AI chip rollout after Nvidia’s record quarter
- Nvidia reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, and analysts said packaging and power limits could slow deployments. - The clearest constraint is 60 megawatts: Iris Energy said Nvidia’s five-year, $3.4 billion Texas agreement covers internal AI workloads at Childress. - AMD said on May 21 its 6th Gen EPYC “Venice” entered production ramp as AI infrastructure demand broadens.
Nvidia reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, with data center sales rising 92% from a year earlier to $75.2 billion, according to the company. Within days of that result, analysts and industry reports said the next limit on AI expansion was no longer demand alone, but the physical system needed to deliver chips into servers and power those servers once installed. Reports from Benzinga and DigiTimes pointed to chip-on-wafer-on-substrate, or CoWoS, packaging, electrical capacity and data-center buildouts as the main constraints on near-term rollout. ### Where is the bottleneck showing up first? CoWoS packaging has emerged as the most cited near-term choke point after Nvidia’s quarter, according to Benzinga’s roundup of analyst views and DigiTimes’ reporting on supply-chain strain. Those reports said advanced AI processors are running into limits not just at wafer fabrication, but in the specialized packaging step that stacks and connects compute dies with high-bandwidth memory. (investor.nvidia.com) DigiTimes reported that Nvidia’s rapid product iteration is forcing supply-chain partners to accelerate development cycles, raise spending and manage higher quality risks. That matters because packaging capacity cannot be expanded as quickly as demand for new accelerators, and suppliers face added pressure when product transitions come faster. (quiverquant.com) ### Why does power now matter almost as much as chips? Iris Energy said this week that a five-year, $3.4 billion agreement with Nvidia covers 60 megawatts of gross power for internal AI workloads using air-cooled Blackwell systems at its Childress, Texas campus. The clarification drew attention because it tied a large Nvidia deployment directly to power capacity, not just to chip purchases. (quiverquant.com) The same Iris clarification said the company still has 390 megawatts of uncontracted capacity at the site, a figure investors and market watchers have treated as a measure of future AI hosting potential. That framing has pushed power, cooling and site readiness closer to the center of the AI buildout story. (quiverquant.com) ### Who stands to benefit if Nvidia’s supply chain stays tight? AMD has been making the case that CPUs remain a critical part of AI systems, especially as “agentic AI” increases demand for orchestration, memory handling and general-purpose compute inside data centers. In a May 21 release, AMD said its 6th Gen EPYC processor, code-named “Venice,” had entered production ramp on TSMC’s 2-nanometer process, and the company linked that milestone to demand for accelerated AI infrastructure deployments. (quiverquant.com) The Motley Fool, citing Wall Street interest in CPU-heavy AI workloads, said AMD’s EPYC line was benefiting from the rise of agentic AI. That is not a direct substitute for Nvidia’s accelerator business, but it shows how tighter GPU supply and more complex AI clusters can lift spending across adjacent parts of the server stack. (amd.com) ### Is this a demand problem or an infrastructure problem? Nvidia’s May 20 report showed demand remains strong by almost any conventional measure, with quarterly revenue up 20% sequentially and the company authorizing an additional $80 billion in share repurchases. The constraint described in recent reporting is different: factories, packaging lines, power contracts and completed data-center capacity must all line up before revenue can convert into installed systems. (fool.com) Yahoo Finance said investors are weighing those two forces together — continued global AI demand against rising infrastructure constraints. That helps explain why enthusiasm for the broader AI trade has held even as some reporting after Nvidia’s earnings focused on muted stock-market reaction and execution limits. ### What should readers watch next? (investor.nvidia.com) Nvidia’s next supply-chain signals are likely to come from supplier capacity updates, customer deployment timelines and any changes to Blackwell system rollout. AMD has already set one near-term marker by starting the production ramp for “Venice,” while Iris Energy’s Childress project gives investors a named site, a 60-megawatt commitment and a five-year contract to watch as Texas AI workloads scale. (investor.nvidia.com) (amd.com)