Wells Fargo says AI capex is surging to $174 billion in Q1 2026, up 72.8% year over year, and argues investors should keep owning AI until growth slows or inflation reaccelerates. The bank said AI spending drove 42% of 1Q GDP growth and 2.4% of U.S. GDP, while Goldman estimated hyperscaler AI capex at $400 billion in 2025 and $700 billion in 2026. The article is broadly bullish on AI spending and the trade, though it also flags bubble risk and valuation comparisons to railroads and the dot-com era.
The market is beginning to treat AI infrastructure as a self-funding utility cycle rather than a speculative spend cycle. That matters because once hyperscaler capex is large enough to meaningfully move GDP, the winners broaden from the obvious megacap platforms into the full “picks-and-shovels” stack: power generation, grid equipment, data-center cooling, networking, and memory/storage. The second-order effect is that AI spend becomes less sensitive to near-term software monetization and more sensitive to financing capacity, power availability, and execution on buildout cadence. The key risk is not valuation in isolation; it is duration of the capex cycle versus margin compression from higher funding costs and power inputs. If inflation reaccelerates or oil keeps pressure on input costs, the market can sustain the narrative for a while, but the lag effect will show up first in balance-sheet-heavy builders with long-dated commitments and then in suppliers whose order books are concentrated in a few hyperscalers. That creates a classic late-cycle setup where the first derivative remains strong even as forward returns narrow. Oracle is the most interesting expression because the debate is no longer about demand but about whether the market is over-penalizing the capital intensity required to convert backlog into revenue. If the capex-to-commitment gap keeps shrinking, ORCL can rerate even before free cash flow inflects, but this is also where disappointment risk is highest if delivery timelines slip. Meanwhile, the relative underappreciation is in non-obvious beneficiaries like utilities, electrical equipment, and semiconductor memory vendors, which can still benefit if AI capex keeps compounding into 2026 even as megacap multiples compress. The contrarian view is that consensus is focused on bubble duration, not bubble breadth. When a capex boom is large enough to support multiple ecosystems, the best short is rarely the marquee AI name; it is often the companies whose earnings are most exposed to the next-stage funding bottleneck or to falling utilization if buildout pauses. The timing window is months, not days: the trade should work as long as capex revisions keep moving up, but reverses quickly if growth prints roll over or inflation forces rates higher.
AI-powered research, real-time alerts, and portfolio analytics for institutional investors.
Request DemoOverall Sentiment
mildly positive
Sentiment Score
0.18
Ticker Sentiment
