Morgan Stanley: Power Gap Widens to 38GW, NVIDIA Is Top Semiconductor Pick
nashnova research
Morgan Stanley raised its US data-center power-gap forecast to 38GW and flagged that roughly a third of the shortfall is not yet priced in; with power now the binding constraint, Nvidia's per-watt efficiency edge makes it the bank's top semiconductor pick.
Why did the power gap just jump by 5GW?
Morgan Stanley lifted its 2026–2028 US data-center net power gap from 33GW to 38GW. This means → data centers will need 38GW more electricity than the grid can deliver — roughly the output of 38 large nuclear plants.
The trigger: investors are trying to match $1.5 trillion (2027) and $1.6 trillion (2028) in planned capex against actual power availability. In plain terms = the money is approved, but the outlets are not.
Nvidia rallied on the report, rising nearly 3% on Monday to a market cap of $5.23 trillion.
Why hasn't the market priced this gap in yet?
Morgan Stanley says roughly one-third of the power gap sits outside market consensus. Many investors believe companies will "creatively find ways around the bottleneck," citing still-strong supply-chain orders as evidence.
This means → there is a real split on how severe the shortage is. Optimists bet on ingenuity; Morgan Stanley bets that physics does not negotiate.
This reflects a deeper signal: when Wall Street lacks consensus on a bottleneck, the probability of a pricing error is highest — and the snap re-pricing, up or down, will be sharper.
Under a power crunch, why does Nvidia beat custom ASIC chips?
Morgan Stanley argues that in a power-constrained world, ASIC chips — silicon designed for a single task, such as Google's TPU — face share-loss risk relative to GPUs.
In plain terms = when electricity is scarce, buyers pick the chip that computes the most per watt. Nvidia's GPU leads on tokens per watt (how many AI task units each watt of power can process); the scarcer power gets, the more that edge is worth.
The bank's top picks are Nvidia and Broadcom, citing four reasons: conservative guidance, chipment visibility, geographic expansion outside the US, and Nvidia's superior per-watt efficiency.
If data-center construction stalls, who gets hurt first?
Morgan Stanley flags downside risk: if construction pauses, the disruption will likely hit the tail end of the "bullwhip effect" — a supply-chain phenomenon where small demand swings at the front get amplified into large inventory swings at the back.
Specifically, memory and optical modules are the most exposed downstream components; in a bear case, power-management and analog chips may also face pullbacks from over-ordering.
This means → some of the strong orders in these segments are driven by fear of shortage, not pure end demand — whether the power gap materializes as forecast is the single most important validation point for the entire thesis.
市场有风险，内容仅供研究参考，不构成投资建议。