Thoughts on the Market (Morgan Stanley) · · 12 min

Can AI Chip Demand Outrun the Power Crunch?

With Joe Moore, Stephen Byrd

Summary

Morgan Stanley strategists Stephen Byrd and Joe Moore weigh whether AI chip demand can keep growing despite a looming US data center power crunch. Byrd raises his estimate of the US data center power shortfall through 2028 to the mid-30 GW range because new governors in Texas and Ohio will impose additional approval processes, which he frames as a delay rather than a cancellation. He argues that off-grid generation, international expansion (Malaysia, India, Japan, China) and grid flexibility do not close the gap in time, citing Virginia's ban on on-site gas generation, a UK project terminated over slow grid access, and utilities' worst-case planning. Moore counters that component shortages, especially memory, are the binding constraint, that data center fulfillment sits near 50%, and that demand stays strong into 2028 despite repeated warnings of a slowdown. He notes memory stocks trade near three times earnings after falling 60-70%, and advises that companies avoid guiding to 2030 because the market already prices in a 75% four-year CAGR, leaving the market to judge AI economics as more cost-of-sales data emerges.

Key takeaways

  • Stephen Byrd on AI infrastructure & data centers: US data center power shortfall widens to mid-30s GW in 2028 from Texas and Ohio delays, not cancellations.
  • Stephen Byrd on AI infrastructure & data centers: Off-grid and international data center power offsets together are not yet enough to close the US shortfall.
  • Stephen Byrd on Power demand & the grid: Grid optimization to free up spare capacity for data centers is overhyped; utilities plan for worst-case peak days.
  • Joe Moore on AI chips & semiconductors: AI chip supply is getting tighter into 2027, with severe component shortages, memory the top constraint.
  • Joe Moore on AI chips & semiconductors: Market is pricing in an AI chip slowdown in 2028 that Moore does not see in demand.

What was said, by company and theme

8 views from 2 speakers. Every quote is checked against the episode's transcript (how).

AI chips & semiconductors

Where everyone stands on AI chips & semiconductors

  • Joe Moore · Morgan Stanley

    AI chip supply is getting tighter into 2027, with severe component shortages, memory the top constraint.

    Moore says the semiconductor market is not just staying tight for 2027 but getting tighter, with all component shortages severe. Multiple hyperscaler CEOs say memory is their biggest problem, not land and power, in the short term. Land and power are becoming a bigger focus, but for now the constraint is components.

    “So I think for 2027, it's not just staying tight, it's getting tighter. Like all of the component shortages are severe.”at 7:11
  • Joe Moore · Morgan Stanley

    Memory stocks look cheap, trading around 3x earnings after falling 60-70%, despite range-bound trading.

    Consumer PC and smartphone memory demand is weak and PC inventories remain high. Data center demand, however, is the true driver, and that keeps memory stocks range-bound. Moore points out that the stocks are down 60-70% and trade at about three times earnings, so any sustainability in the business would let investors buy back much of the company within two years.

    “What that does to me for memory is it keeps the stocks very range bound. But look at the range. I mean, we're down 60%, 70%, and we're trading at like three times earnings.”at 8:50
  • Joe Moore · Morgan Stanley

    Market is pricing in an AI chip slowdown in 2028 that Moore does not see in demand.

    Moore sees demand as still very strong into 2028. Data center fulfillment is very low, around 50%, and compute is ubiquitous, so there is still demand for chips even when some projects get delayed. Customers keep saying 2028 will be a challenge, but he has heard the same concern at the end of 2024, 2025 and 2026 without a slowdown appearing. He thinks the market is already valuing the stocks as if the slowdown in 2028 is happening, which he doesn't believe the demand data shows. He is bullish across the group overall.

    “So I'm pretty bullish on all of it... What about '28? ... It just doesn't feel that way to me and given that the market is basically saying now that it does, we're being valued as it slows in '28.”at 9:06
  • Joe Moore · Morgan Stanley

    Guiding AI chip demand out to 2030 would add risk for stocks in a duration trade; prove it quarterly.

    Moore says that in a duration trade, companies should not guide to 2030, because that creates risk. Investors are already pricing in a 75% four-year CAGR, so the stocks have a hard time moving forward unless each quarter validates the trajectory. He expects visibility into AI economics to come as companies that now go public show actual cost of sales and as cloud providers' customers reveal how AI is used. He thinks the market will decide if the numbers are sustainable.

    “I don't think guiding to 2030 is the answer when you're in a duration trade. That just creates risk. It makes it hard for the stocks to move forward because you're already guiding for a seventy five percent four year CAGR.”at 10:57

AI infrastructure & data centers

Where everyone stands on AI infrastructure & data centers

  • Stephen Byrd · Morgan Stanley

    US data center power shortfall widens to mid-30s GW in 2028 from Texas and Ohio delays, not cancellations.

    Morgan Stanley's estimate of a 30 GW US data center power shortfall through 2028 is revised higher to the mid-30s GW because newly elected governors in Texas and Ohio will require developers to go through new processes. Byrd frames this as a delay rather than a cancellation: a path to building the data centers still exists but takes longer. Deployment is much bigger in 2029, so the shortfall is a relief valve that lengthens the growth cycle.

    “Mathematically, what we essentially are looking at is that bottleneck getting larger in 2028 because of a couple of gubernatorial outcomes, Texas and Ohio being specific, where we think it's a delay. It's not a cancellation.”at 1:24
  • Stephen Byrd · Morgan Stanley

    Off-grid and international data center power offsets together are not yet enough to close the US shortfall.

    Byrd says off-grid power solutions are attractive, but the main constraint is time, since engineering, labor and permits take long. Virginia's ban on on-site gas generation shows how state policy can block that path. Expanding internationally helps in places like Malaysia, India, Thailand, Japan and China, but Norway has a data center moratorium and projects in Australia and the UK have run into problems, including a UK project terminated over slow grid access. Adding all this up, the offsets are not enough to address the US shortfall today, though he expects more emphasis on solutions over time.

    “So it's a lot of puts and takes. When we add all this up, as of now, it's not enough to address the [US] shortfall, but I would expect over time just more and more emphasis on getting solutions.”at 3:56
  • Stephen Byrd · Morgan Stanley

    Developers funding local hospitals, teacher pay and public services can win community support for data centers.

    Byrd agrees that voter concerns about data centers are warranted in some cases, because some projects are not designed to work for local communities. He expects smarter development models that work for all constituents, which could reverse the post-election pause. Louisiana is his example: developers there are bankrolling new hospitals and pay raises for teachers, police and firefighters. Industrial activity in the state, including data centers, is large enough to rival entire nations, so he sees a path where the slowdown is temporary and takes time to work through.

    “what I think we'll see is very smart approaches to developing data centers that work for all the constituents... And when you do it in the right way, I think you can get support and create economic activity.”at 6:01

Power demand & the grid

Where everyone stands on Power demand & the grid

  • Stephen Byrd · Morgan Stanley

    Grid optimization to free up spare capacity for data centers is overhyped; utilities plan for worst-case peak days.

    The idea is that the US grid is utilized only about 50% of the time on average, so there is slack that flexible data center load could fill. Byrd argues this misreads how utilities operate: they plan for the worst possible day, such as a six-day winter storm when demand peaks and some power plants trip offline. Utilities therefore assume every data center developer will want full power at the same time, so the flexibility idea doesn't work under a utility's conservative planning approach. He is explicitly not bullish on grid optimization as a solution.

    “The idea is that [the] US grid is actually only utilized on average, like, 50% of the time... The problem is that's not how a utility executive would look at the situation.”at 4:21

Topics

  • 2028 AI chip demand outlook
  • AI chip supply tightness
  • data center community acceptance
  • data center power offsets
  • data center power shortfall
  • grid optimization
  • long-term demand visibility
  • memory stock valuation

More from Thoughts on the Market (Morgan Stanley)

All Thoughts on the Market (Morgan Stanley) episodes

Ask your AI about this episode

Chorus gives Claude, ChatGPT and other assistants every view from this episode and every other show Chorus follows. Ask who agrees, who changed their mind, and what else was said about the same companies.

Setup guide

A weekly email with the calls that mattered from the most important voices in investing. Unsubscribe any time. Privacy Policy

Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus