Morgan Stanley’s Stephen Byrd calculated that US data center developers face a power shortfall of 70 gigawatts through 2028, ZeroHedge reported on 1 November. After counting turbines, fuel cells and converted bitcoin mines, ZeroHedge wrote, “the gap is still 43 GW, or 38% of all the power AI developers need”.
A year earlier, in its November 2025 note, the bank put the net gap at about 13 GW once those “Time to Power” solutions were counted, according to a Longbridge summary of that report. By our arithmetic, 43 GW is about 3.3 times 13 GW, though the forecast windows differ: 2025 to 2028 for the earlier note and 2026 to 2028 for the latest.
Business Pill · Grid Cost Allocation Explained
An evergreen Business Engineer explainer on who pays for grid upgrades. It teaches the general idea only and does not discuss Morgan Stanley’s estimate.
Animated: Morgan Stanley’s shortfall before time-to-power fixes by report vintage, as reported (windows differ). Chart: FourWeekMBA.
The key insight: Morgan Stanley’s latest estimate, as reported by ZeroHedge, leaves a 43 GW net shortfall in US data center power through 2028 after every time-to-power fix the bank counts, about 3.3 times the 13 GW net gap in its November 2025 note by our arithmetic, on a slightly different window.
A ZeroHedge chart compares three Morgan Stanley vintages. In December 2024 the bank put US data center power needed for 2025 to 2028 at 57 GW, with a 36 GW shortfall before any fixes. The November 2025 note raised need to 69 GW and the pre-fix shortfall to 44 GW.
The October 2026 vintage on the chart shows 116 GW needed for 2026 to 2028 and a 70 GW shortfall before fixes, with 43 GW left after them. The chart labels the net gap “3.3x in a year”.
Longbridge’s summary of the November 2025 report listed the fixes then: natural gas turbine projects of about 15 to 20 GW, Bloom Energy fuel cells of 5 to 8 GW, and direct deals with existing nuclear plants of about 5 to 15 GW.
Morgan Stanley’s US data center power estimates by report vintage, as shown on a ZeroHedge chart and in summaries of the earlier notes: power needed, the shortfall before time-to-power fixes and the net shortfall after them. Forecast windows differ (2025 to 2028 for the first two vintages, 2026 to 2028 for the latest); the 3.3 times multiple is our arithmetic.
The Numbers Beside It
ZeroHedge set the estimate against a strong month for supply: US generating capacity rose by a net 5.6 GW in September, the strongest September in at least a decade, according to EIA data compiled by Goldman’s utilities team. By our arithmetic, the 43 GW net gap equals about 7.7 months at that pace.
Our earlier pieces give two more yardsticks. Musk’s 1.2 million Blackwells would draw about 2.2 GW of rack power before cooling, by our arithmetic, so 43 GW is roughly 20 campuses of that size. SemiAnalysis models 62.2 GW of US net additions in 2028 alone; 43 GW equals about 69% of that figure, though the two measure different things.
The Business Lever: Who Supplies the Missing Gigawatts
The estimate collides with another recent view. A Google data center energy executive said US power supply is sufficient but permits are slow, according to Techmeme’s summary of a report. As we read it, the two views differ on whether the constraint is generation itself or the time it takes to connect it.
As we read the three vintages, need outran the fixes: by our arithmetic, time-to-power fixes closed about 31 GW of the gap in the November 2025 vintage (44 GW to 13 GW) and about 27 GW in the latest (70 GW to 43 GW), while the pre-fix shortfall rose from 36 GW to 44 GW to 70 GW on ZeroHedge’s chart.
As we read it, the November 2025 note treated time-to-power solutions as closing most of the gap, leaving 13 GW, while the latest vintage leaves 43 GW after them.
As we read the September data ZeroHedge cites, even the strongest September in at least a decade would need about 7.7 months at that pace to cover 43 GW, by our arithmetic.
ZeroHedge, 1 November 2026
“the gap is still 43 GW, or 38% of all the power AI developers need”
Three Implications
FOR DATA CENTER DEVELOPERS ZeroHedge wrote that the 43 GW remains after counting turbines, fuel cells and converted bitcoin mines.
FOR ON-SITE POWER SUPPLIERS Morgan Stanley’s November 2025 note sized natural gas turbine projects at about 15 to 20 GW and Bloom Energy fuel cells at 5 to 8 GW, per Longbridge’s summary.
FOR FORECASTERS The vintages use different windows: 2025 to 2028 for the first two and 2026 to 2028 for the latest, according to ZeroHedge’s chart.
The framework puts it this way: “But infrastructure reality is catching up with AI ambitions: a projected 19GW gap between new data center demand and available grid capacity threatens to deflate the AI bubble.”
As we read it, the framework’s 19GW figure is an earlier estimate; Morgan Stanley’s latest vintage, as ZeroHedge reports it, puts the net gap at 43 GW after fixes.
We did not read the October 2026 Morgan Stanley note itself; its figures come from ZeroHedge’s article and chart, and the 116 GW need figure appears only on that chart. Earlier vintages come from Longbridge’s summary of the November 2025 report and a Crisis Investing newsletter citing the December 2024 chart. The windows differ across vintages, and the 3.3 times, 7.7 months, 20 campuses and 69% figures are our arithmetic.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Morgan Stanley’s latest estimate leaves a 43 GW net US data center power shortfall through 2028 after every time-to-power fix it counts, ZeroHedge reported, up from about 13 GW a year earlier on a slightly different window.
A note on sourcing. The October 2026 Morgan Stanley figures come from ZeroHedge’s 1 November article and chart; earlier vintages come from a Longbridge summary of the November 2025 report and a Crisis Investing newsletter. We did not read the Morgan Stanley notes themselves. The 3.3 times, 7.7 months, roughly 20 campuses and 69% figures are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
