AI capex cycle debate ignites as Nvidia faces dual pressure from memory competition and infrastructure costs

The AI narrative has fractured into two competing stories: memory chip makers celebrating a supercycle, versus skeptics questioning whether the capex cycle can sustain valuations that have already surged substantially. Nvidia found itself under scrutiny this week after CoreWeave announced major infrastructure partnerships; the deal sparked concern among traders that Nvidia's competitive advantage in AI chips might be eroding if infrastructure firms could deploy alternative architectures. CoreWeave's CEO stated publicly that Nvidia must expand AI capacity or risk losing customers to AMD, a blunt signal that buyers are demanding optionality and cheaper alternatives.

The infrastructure buildout narrative hinges on sustained hyperscaler spending from Amazon, Google, Meta and Microsoft to support large language model training and inference. Goldman Sachs and others have raised bullish notes on AI winners, but an undertone of skepticism has emerged: if current capex projections materialize, do the returns justify the spending? Some analysts and traders have begun flagging the circular investment theme, wherein infrastructure providers like CoreWeave invest in Nvidia chips to build out data centers that then drive Nvidia demand. This creates the appearance of organic demand while masking the reality that the cycle is vendor-driven and potentially unsustainable if hyperscalers moderate spending.

Nvidia's own forward guidanceCompany-issued forecasts of future financial performance. will be critical; the company has guided to continued strength but faces pressure from both memory chip competition and rising infrastructure costs. AMD has signaled that AI custom silicon and chiplet architectures could offer hyperscalers cheaper alternatives, though adoption remains limited. Broadcom and other fabless semiconductor firms are also racing to develop AI-specific designs that could disrupt Nvidia's dominance. The market has largely dismissed these competitive threats, but if hyperscalers diversify their supply chains, Nvidia's growth trajectory could decelerate materially.

Bulls counter that AI capex will prove self-reinforcing and that the emerging infrastructure ecosystem creates new winners rather than destroying existing ones. However, the juxtaposition of extreme valuations on Nvidia versus compressed multiples on memory makers suggests investors are not yet pricing in the full magnitude of infrastructure vendor competition and capex sustainability risks.