Daniel Romero
Technology Equity Analyst, Semiconductors/Data Centers/Physical AI
Data center and power plays are very undervalued because demand is set to outpace supply by a wide margin. TrendForce expects a 170GW gap by 2030, and a widening gap means massive pricing power gains for Nebius, Cipher, TeraWulf, Hut 8 and Keel Infrastructure that are not priced in. x.comÂ
The sell-off in data-center stocks such as Nebius, Cipher, TeraWulf, Hut 8 and Keel is irrational because cheaper intelligence raises demand for power. Models like Kimi are very cheap yet need roughly the same energy, so demand for both intelligence and power grows, and energy becomes the bottleneck. Moonshot, OpenAI and Anthropic all lack enough compute; Moonshot said Kimi K3 demand pushed it near its capacity limits, forcing it to pause new subscriptions. Valuations now look attractive. x.com x.comÂ
Nebius, IREN and CoreWeave stand to benefit from a deepening shortage of US data-center power. Morgan Stanley raised its US data-center power shortfall forecast to 33 GW. x.comÂ
Electricity is a clear bottleneck for AI, favoring AI-focused power and data-center infrastructure names over residential utilities. The US could be short 107 GW to power data centers through 2029, while China adds capacity year after year. Medium-term preferences are Cipher, Hut 8, Riot, TeraWulf, Galaxy, Equinix, Bloom Energy, Solaris, Vistra, GE Vernova, INIO, Williams, Liberty Energy, EROC and Cummins. x.com x.comÂ
Nebius' business model faces pressure once GPU supply and demand normalize, perhaps within five years. It faces enormous supplier power, more competitors and likely rising buyer power, and lenders and equity investors may view it differently once utilization and pricing stop being exceptional. China could flood global markets with low-cost compute, and the market already seems to price a very long AI boom. The case weakens if AI demand proves durable, like an 18-year real estate cycle. x.comÂ
AI's electricity shortfall creates a massive opportunity for companies solving the power bottleneck. Morgan Stanley estimates a 32 GW US data-center power shortfall by 2028. Constellation Energy and Vistra are nuclear names, Bloom Energy in fuel cells, GE Vernova in power generation and Eaton in grid infrastructure. x.comÂ
Meta's unmet AI demand pays its bottleneck suppliers. Super Micro's $19.5B quarterly outlook and under 8x forward earnings point to a double. CoreWeave's Meta deals total $35.2B; it should easily top $100. Builders say nearly all new sites use Dell or Super Micro; Dell could easily double. Micron's demand is off the charts at 5x forward earnings. Sources say Vertiv's gear is sold out for 2-3 years. VivoPower has 2GW+ potential in 5-7 years. Digi Power X has a $1.1B Cerebras deal. x.com x.com … Show more
AI power generation is now all about deployment speed, with a shortfall of about 38 GW through 2028, of which only about 15 GW will be met by the grid, per Morgan Stanley. Bloom Energy looks pricey at about 82x earnings, but a bull case of about $25B of 2030 revenue implies $16.70+ EPS and $668 at 40x. Gas winners include Howmet, GE Vernova, Caterpillar and Baker Hughes. Cheap solar plus batteries deploys in 18-24 months, favoring Nextracker, Shoals, Fluence, Tesla and Albemarle. x.comÂ
An imminent FERC ruling on how data centers connect to the grid could catalyze PJM-exposed names such as Talen, Constellation, TeraWulf, Keel and MARA. FERC is expected to rule by end of June on PJM co-location rules and a DOE-directed rulemaking that may make hyperscalers pay for grid upgrades. Existing interconnects gain value if hyperscalers self-fund, and certainty alone could start a new PJM investment cycle. PJM fell 6.6 GW short in its December 2025 auction at a record $333/MW-day. x.comÂ
AI infrastructure operators such as Core Scientific, CoreWeave, Galaxy Digital and TeraWulf benefit if out-year gigawatt projections are not met. Constraints on power generation, grid transmission, labor and wafer capacity limit how many debt-financed data centers and chips get built, which is healthier longer term if the build-out is funded from operating cash flow. @LEJCapital counters that this assumes demand keeps growing exponentially and stays far above supply. x.com x.comÂ
Hubbell sells both the utility-side and data-center-side gear for an AI power buildout bottlenecked by transformer and switchgear shortages, a setup retail isn't pricing in. Half of US 2026 data centers are delayed on those shortages, equipment prices are up 77-95% since 2019, and US peak load is forecast to grow 166 GW in five years. A 2.4 PEG leaves no room for a guidance cut, the stock sits below its 50-day average, and lumpy utility capex could push orders out. x.com x.com x.com x.com x.com … Show more
Powered-shell suppliers such as Applied Digital, Core Scientific, CleanSpark, TeraWulf and Fermi look well positioned as a large gap opens between 2027 compute demand and available gigawatts. Their financing needs, however, make interest rates and the credit environment bigger risks for them. x.comÂ
