InvestNot yet confirmed elsewhere1 publisher2 min readPublished
UBS starts Core Scientific and Applied Digital at Buy on a 19 GW power shortfall for 2027
UBS began coverage of Core Scientific and Applied Digital at Buy, citing 57 GW of data center capacity needed in 2027 against 38 GW with confirmed power. Its forecast of about 80% annual EBITDA growth through 2029 needs sites that already have grid connections, because new requests take years to clear.
The Investor · Invest desk
Five UBS figures: 19 GW of 2027 capacity lacks confirmed power; about 80% annual EBITDA growth projected for 2027-2029; about 6 years median from grid request to operation; US queues near 2.3 TW; about 415 GW of adequate power by 2030 if initial projects stay on schedule.
- 19 GW 2027 shortfall in confirmed power, claim 15
- 80% Annual EBITDA growth, 2027-2029, claim 13 UBS projection for colocation names
- 6 years Median interconnection wait, claim 4 request to commercial operation
- 2.3 TW US interconnection queue, claim 3 around
- 415 GW Adequate power by 2030, claim 14 if initial projects come online on schedule
| Figure | Value | Note | Claim |
|---|---|---|---|
| 2027 shortfall in confirmed power | 19 GW | 15 | |
| Annual EBITDA growth, 2027-2029 | 80% | UBS projection for colocation names | 13 |
| Median interconnection wait | 6 years | request to commercial operation | 4 |
| US interconnection queue | 2.3 TW | around | 3 |
| Adequate power by 2030 | 415 GW | if initial projects come online on schedule | 14 |
What happened
- The initiation came on September 23, 2026 and covered several colocation data center stocks, with UBS citing constrained compute supply against rising demand.
- UBS notes a data center can be built in 12 to 24 months, while adding power capacity can take 5 to 10 years.
- US interconnection queues sit around 2.3 TW, roughly double the country's installed capacity as of 2024.
- By UBS's count, 276 GW of data center capacity worldwide is being built or is in planning.
- UBS estimates $511 billion is needed for new US power generation by 2030 to support 3% compound annual growth in electricity demand.
Why it matters
- exposure One third of the capacity needed in 2027, 19 GW of 57 GW, has no confirmed power, so that slice of AI demand depends on power nobody has yet confirmed.
- cost In states that make developers pay for grid upgrades, the cost moves from utilities and ratepayers to the builder, so the spending lands on the developer's books ahead of any tenant revenue.
- constraint New York's one-year moratorium on new data centers above 20 MW leaves large projects there waiting 12 months, and developers who want that scale have to look elsewhere.
- contradiction UBS projects about 415 GW of adequate power by 2030 if initial projects arrive on schedule, so the research that prices scarcity today also forecasts it ending, and the Buy ratings rest on the 2027 to 2029 stretch in between.
UBS's Buy ratings rest partly on what it describes as strong power development pipelines at the covered firms [11], and a pipeline is a queue. At the US median, a request takes about 6 years to reach commercial operation [4], so a request filed in 2026 energizes in 2032, three years after the 2027 to 2029 window in which UBS forecasts about 80% annual EBITDA growth for these names [17][13]. The growth therefore has to come from sites that already hold connections (or rather, from sites already through the queue, since a filed request is not a grant).
The growth rate is large. Read as two annual steps after 2027, 80% a year compounds to 3.24 times the 2027 level by 2029 [18]. Equipment sets its own delay: UBS says large transformers may face lead times of up to 210 weeks [8], or about 4.04 years, so an order placed in late 2026 at the outer end of that range arrives around 2030 [19].
Our view is that the Buy calls are a bet on who already holds scarce connections, and the 80% forecast is the test of that bet. The counter-thesis is easy to state and could be right. If Core Scientific and Applied Digital already hold the powered sites that drive 2027 to 2029, new interconnection requests are irrelevant to them and the 6-year median does not bind [4]. Company disclosure of energized and contracted megawatts that covers the growth path would settle it, and would show our objection to be wrong.
The cryptobriefing.com summary of UBS's "Signal over Noise #14" report does not include price targets or company-level megawatts [1].
What to watch
- Whether the 38 GW of sites with confirmed power moves toward the 57 GW UBS says is needed in 2027.
- Whether other states copy New York's moratorium on data centers above 20 MW or the developer-pays rule on grid upgrades.
- Whether the first power projects UBS counts on for its 415 GW forecast reach their in-service dates.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+20
- Incentives
- Insufficient
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
UBS research titled "Signal over Noise #14: The AI 'super-exponential' meets its speed limits" argues that power supply, grid capacity and data center space are driving the next phase of the AI trade; the cryptobriefing.com summary does not include price targets or company-level megawatts.
- [2]
UBS notes a data center can go up in 12 to 24 months, while adding power capacity can take 5 to 10 years.
- [3]
US interconnection queues sit at around 2.3 TW, roughly double the country's installed capacity as of 2024.
- [4]
The median time from interconnection request to commercial operation in the US is now about 6 years.
- [5]
UBS tracks 276 GW of global data center capacity that is either under construction or planned.
- [6]
UBS sees 57 GW of capacity needed in 2027, against just 38 GW of sites with confirmed power.
- [7]
UBS estimates $511 billion will be needed for new US power-generation capacity by 2030 to support a 3% compound annual growth rate in electricity demand.
- [8]
Large transformers may face lead times of up to 210 weeks.
- [9]
New York has enacted a one-year moratorium on new data centers exceeding 20 MW.
- [10]
Other states are requiring developers to pay for grid upgrades themselves, shifting costs from utilities and ratepayers onto the builders.
- [11]
On September 23, 2026, UBS initiated coverage on several colocation data center stocks with a bullish stance, citing the constrained supply of compute against rising demand and strong power development pipelines at these firms.
- [12]
UBS started Core Scientific (CORZ) and Applied Digital (APLD) at Buy.
- [13]
UBS projects around 80% annual EBITDA growth for these colocation names from 2027 to 2029.
- [14]
UBS projects there will be adequate power, approximately 415 GW, to meet AI demand by 2030, provided the initial projects come online as scheduled.
- [15]
The gap between capacity needed in 2027 and sites with confirmed power is 19 GW.
- [16]
About one third (33.3%) of the capacity UBS says is needed in 2027 has no confirmed power.
- [17]
At the 6-year US median, an interconnection request filed in 2026 reaches commercial operation in 2032, three years after the 2029 end of UBS's 80% EBITDA growth window.
- [18]
If 80% annual EBITDA growth applies to the two annual steps after 2027, EBITDA in 2029 is 3.24 times its 2027 level.
- [19]
A 210-week transformer lead time is about 4.04 years, so an order placed in late 2026 at the outer end of the range arrives around 2030.
Sources
1 independent publisher whose own reporting we read for this story.
- cryptobriefing.comUBS says power and grid constraints are the next phase of the AI trade
1 article · October 12, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.