Nvidia, CoreWeave competition fuels AI infrastructure spending race

The race to build out generative AI inference and fine-tuning infrastructure has entered a new phase of intensity. Cerebras Systems raised its IPOInitial Public Offering - a company's first public sale of stock. price guidanceCompany-issued forecasts of future financial performance. to $150-$160 per share (from prior $115-$125 range) on strong institutional demand, signaling conviction that specialized AI compute will command premium valuations. CoreWeave, a data-center GPU specialist, announced a transformative deal with Iren (backed by venture and strategic capital) to expand GPU rental capacity, effectively positioning itself as a competitor to hyperscalers' internal GPU clouds. CoreWeave's CEO publicly flagged that Nvidia must expand AI capacity faster or risks losing customers to AMD, hinting at supply constraints and customer frustration with long lead times.

The narrative of AI infrastructure capex reaching escape velocity gained traction from multiple quarters of record orders, backlog extensions into 2027, and bullish guidanceCompany-issued forecasts of future financial performance. from semiconductor equipment makers. Goldman Sachs noted that South Korean memory demand is surging, with JPMorgan hiking its Kospi bull case target to 10,000 on semiconductor cycle improvement. However, skeptics point out that such aggressive capex by hyperscalers and startups mirrors prior tech bubbles where massive spending on overcapacity ultimately crushed returns. The fact that CoreWeave is positioning itself as a GPU-less alternative via AMD GPUs (cheaper but lower-margin for customers) suggests Nvidia pricing power is being tested and that competitive AI clouds are viable alternatives.

Market implications span semiconductors, data-center real estate, power utilities, and networking vendors. Nvidia benefits from the training bottleneck and remains the default standard, but inference margins are lower and more contestable. AMD's EPYC processors and MI-series GPUs are gaining traction in price-sensitive inference workloads. Broadcom, Intel, and Marvell all see TAM expansion from higher networking and interconnect demands. Real-estate plays linked to data-center colocation (CoreWeave partnerships with colo providers) could see outsized growth. Power and cooling becomes a gating constraint, with rising energy costs from the Iran war adding headwind to margin expansion stories.

The debate hinges on whether capex growth can justify current valuations or whether a deceleration in new model releases or training efficiency improvements could crimp spending growth. Earnings risk for equipment vendors and memory suppliers will likely materialize if hyperscaler guidanceCompany-issued forecasts of future financial performance. turns cautious in 2H 2026.