Wells Fargo sees AI slowdown as unlikely despite safety calls

Wells Fargo expects competitive dynamics to prevent a coordinated slowdown in frontier AI development, despite safety warnings from Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman. The bank expects any AI deployment delays to stem from power availability, permitting and local opposition, extending rather than reducing the AI investment cycle. It favors relatively insulated midstream names DTM, KMI, TRP and WMB, along with GEV, while flagging greater near-term deployment risk for BE and GNRC; WMB also benefits from its behind-the-meter power work with Meta.

Analysis

The relevant differentiation is not whether AI capex persists, but where the constraint sits. Grid interconnection and generation scarcity shift project economics toward firms monetizing delivered power and gas transport, while exposing equipment names whose valuations require rapid datacenter energization. WMB's contracted Gulf Coast and power-adjacent exposure offers better duration than merchant-like distributed-power narratives; KMI, TRP and DTM are lower-beta beneficiaries but have less direct incremental AI sensitivity.

Near term, the Fed decision and hyperscaler capex commentary will dominate this thematic signal. Over the next 1-3 months, interconnection-queue data, utility load forecasts, gas-basis spreads and META's disclosed power procurement are more investable catalysts than voluntary-safety rhetoric. A delayed energization schedule can compress BE and GNRC revenue multiples before it impairs long-dated demand, since their order conversion, working-capital needs and fixed-cost absorption are materially more dependent on deployment timing.

The consensus risk is treating all AI infrastructure as a single duration trade. Supply bottlenecks can increase the value of scarce, dispatchable behind-the-meter capacity, but they can also push customers toward utility-scale grid solutions, gas turbines, batteries, or delayed campuses; WMB is not a pure-play off-grid power provider and its AI linkage needs verification through contract disclosures. GEV remains structurally advantaged if power shortages trigger turbine orders, but its premium multiple leaves limited tolerance for backlog conversion delays or margin dilution from capacity expansion.

A reversal would come from a material hyperscaler capex cut, evidence that model efficiency reduces peak power intensity, or federal/state permitting reforms that rapidly relieve interconnection constraints. For the long infrastructure/short near-term deployment basket, invalidate on BE/GNRC bookings and backlog conversion accelerating despite delayed datacenter starts, or on WMB/GEV guidance failing to identify incremental power-load demand by the next two reporting cycles.

Key Decisions for Investors

• Initiate a 3-6 month pair: long WMB / short BE, sized beta-neutral. The trade captures duration value from contracted gas infrastructure versus a valuation more exposed to project timing; target 10-15% relative outperformance, with a 7% relative stop if BE reports materially accelerating bookings and conversion.
• Add GEV only on a 8-10% pullback or after confirmed turbine/backlog disclosures tied to North American datacenter load. Use a 6-12 month horizon; upside requires backlog conversion and pricing to offset capacity-investment drag, while a downward revision to margin or delivery timing is the exit signal.
• Maintain KMI or DTM as lower-volatility AI-power optionality rather than chase WMB after a news-driven move. Reassess following utility integrated-resource plans and 2027-30 load forecasts; lack of incremental contracted volumes would mean the AI narrative is not translating into cash flow.
• Avoid adding BE and GNRC ahead of their next earnings unless order backlog, cancellation rates, financing terms and project energization dates support near-term revenue recognition. A tactical short is appropriate only if consensus revenue estimates remain unchanged despite disclosed permitting/interconnection delays.