AI Infrastructure Capex Supercycle Extends; GPU Demand Holds

The AI capex bull thesis has proven remarkably resilient to Iran war shocks, geopolitical tail risks, and rotation narratives out of mega-cap software. Traders continue to argue that major cloud providers, private equity infrastructure funds, and AI start-ups (including OpenAI, which is at the center of a high-profile litigation revealing internal rivalries and a $852 billion valuation) all remain committed to multi-year infrastructure spending that will sustain elevated demand for GPUs, memory chips, and data-center hardware through 2027.

CoreWeave, the infrastructure provider backed by venture capital and reportedly tied to Nvidia's buildout ecosystem, is seen as a strategic play on the thesis. Its CEO recently signaled that Nvidia must expand capacity or risk losing customers to AMD, implying both supply constraints and demand intensity. Cerebras Systems, a chip design company targeting AI inference and training, has raised its IPOInitial Public Offering - a company's first public sale of stock. price range to $150-$160 per share (from $115-$125) and upped share count, reflecting explosive demand from institutional investors betting on the AI infrastructure cycle. Broadcom and other semiconductor equipment makers have also benefited from the conviction that the capex cycle is durable.

The narrative is strengthened by commentary that AI mania is masking the true damage from the Iran war. A Financial Times analysis noted that since the conflict began, the biggest US companies have gained $5.4 trillion in market cap, but the semiconductor sector accounts for most of the gains, suggesting AI is the only narrative insulating markets from geopolitical shocks. Alphabet's 160% rally over a year is cited as proof that mega-cap tech owns 'most of the stack' in AI: search, cloud, chips, and services.

Sceptics argue that the capex cycle mirrors the 2021-2022 cloud buildout, which led to oversupply and capex cuts in 2023-2024. They point out that if training data becomes scarce or model improvements plateau, capex could decelerate. Additionally, if energy prices spike due to the Iran war and persist, data-center power costs could compress margins for cloud providers, forcing them to ration buildout. The bull case hinges on the assumption that AI model scaling follows an exponential path (Nvidia's Jensen Huang has claimed this), but that path has not been validated empirically.