AI Has a Power Problem: Why the U.S. Power Grid Can't Keep Up | The Real Eisman Playbook Ep 69

Summary

Steve Eisman and Ben Callo focus on the physical constraints behind the AI boom, arguing that power availability, grid interconnects, and construction labor will determine how fast data centers can actually come online. Callo cites very large incremental electricity needs, emphasizes that many data-center projects are still early and subject to double counting, and says the shift to 800-volt Nvidia-based architectures raises power density and equipment demand. He highlights GE Vernova as a long-cycle beneficiary through gas turbines, grid hardware, and an SMR nuclear partnership, while also pointing to Tesla’s energy storage business as a meaningful profit engine alongside autonomy. The conversation broadens into EV competition, Chinese brands gaining share in Europe, First Solar’s tariff sensitivity, MP Materials’ role in magnets and rare earths, and the idea that capital, governance, and energy scarcity all sit at the center of the next industrial buildout.

Notes

• AI buildout is constrained by labor and energy before chip supply even matters.
• Project churn creates false cancellation signals when developers or contractors change midstream.
• Callo says data-center development is still lumpy and early, with many announced sites likely collapsing into fewer real builds.
• Interconnection access is a gating item for hyperscale load.
• The only real gigawatt-scale data center built so far is described as SpaceX.
• 800-volt Nvidia-based architectures are increasing power demand inside the data center.
• The shift also increases demand for outside-the-building grid equipment and internal power electronics.
• Data centers need AC-to-DC conversion and other power electronics to run off direct current.
• Bloom Energy is cited as producing 800-volt DC power.
• U.S. load growth by 2035 is framed as anywhere from 100 GW to 350 GW of new capacity.
• The upper-end case would more than double current U.S. electricity generation.
• Today the U.S. is described as having just over 150 GW of generation capacity.
• Callo expects roughly 30 GW of annual U.S. capacity growth over the next five years.
• One megawatt is framed as enough power for roughly 1,000 homes.
• Baseload is defined as 24/7 firm power rather than intermittent output.
• Nuclear, hydro, and geothermal are treated as the main baseload options.
• Wind is criticized for a low capacity factor compared with firm generation.
• Nuclear is cited at about a 99% capacity factor.
• Gas turbines are said to run in the high-80% capacity-factor range.
• New York curtailment days are becoming more frequent, signaling grid stress.
• GE Vernova’s power segment is centered on very large natural-gas turbines.
• Those turbines are multiple-hundred-megawatt machines.
• GE Vernova’s power equipment sales are said to be about 90% gas-turbine related.
• Backlog visibility for some GE Vernova work extends out to 2030-2031.
• GE Vernova also sells high-voltage transformers and other grid-hookup equipment.
• Cross-selling into data centers is becoming part of GE Vernova’s sales mix.
• Gas-plant air permitting can take about six months.
• Hyperscalers still care about CO2 targets, but local NOx and SOx limits are the immediate permitting issue.
• GE Vernova’s SMR partnership with Hitachi is aimed at data centers and small-city applications.
• SMRs are framed as a late-decade opportunity rather than an immediate earnings driver.
• Typical SMR sizes cited are 250-400 MW, so they are not tiny units.
• SMRs need a new supply chain and a financing model that can withstand first-of-a-kind risk.
• Tesla’s U.S. EV business is described as flat at best while the broader U.S. market still grows in single digits.
• China is the strongest EV market, followed by Europe and then the U.S.
• Chinese EV brands are gaining share in Europe, with BYD showrooms visible across major cities.
• Tesla is characterized as one of the most innovative companies in the world.
• Tesla is a top destination for elite engineering talent alongside SpaceX.
• Tesla would look overvalued if it were only an EV company.
• Tesla’s robotaxi rollout is expected to start in Texas first.
• Austin, Houston, and Dallas are described as fully permitted for Tesla’s rollout.
• Waymo and Zoox are framed as active competitors, suggesting a multi-player autonomy market.
• Tesla’s autonomy thesis depends on a camera-only stack rather than lidar or radar.
• Fleet data from all Tesla vehicles feeds machine learning whether or not owners bought FSD.
• Tesla’s energy business contributes about 20% of operating income.
• Megapacks are being deployed as standalone grid batteries as well as paired with renewables.
• Tesla has Megapack plants in California, Texas, and Shanghai.
• Tesla energy is described as growing over 30% annually with healthy margins and meaningful cash flow.
• Tesla energy cash flow is said to help fund robotaxi and Optimus.
• Elon Musk’s desired 25% Tesla ownership is framed as control over AI.
• A combined SpaceX-Tesla structure is portrayed as a backdoor to governance and capital control.
• Keeping the companies together would make capital raising easier in the AI and data-center arms race.
• SpaceX and Tesla are said to be collaborating on a 100 GW solar manufacturing facility.
• The proposed solar manufacturing scale is described as China-scale.
• Terafab is described as a chip project for self-driving vehicles or for competing with Nvidia.
• Elon is portrayed as frustrated by separate board approvals on joint projects.
• Callo says a SpaceX-Tesla combination could happen within 18 months, maybe sooner.
• Space-based data centers are used as an illustration that energy, not compute, is the limiting resource.
• MP Materials owns the largest U.S. rare-earth mine and is building magnet capacity.
• First MP Materials magnet output is already sold out to GM and Apple.
• China is said to control about 90% of rare-earth supply and even more of the magnet layer.
• Residential solar is described as a tough door-to-door sales business.
• Residential solar has very high customer-acquisition costs and complaint volume.
• Utility-scale solar is favored over residential solar on scale and cost grounds.
• First Solar is still waiting on clarity around Section 232 tariffs for polysilicon and related panels.
• First Solar is shifting capacity from Malaysia and Vietnam back to the U.S.
• First Solar wants specific dollar tariffs rather than percentage tariffs.
• Callo sees about $100 upside in First Solar if tariffs break favorably.
• He gives a rough First Solar earnings bridge: every penny of ASP is worth about $2 in earnings.
• First Solar could do about $12 EPS this year, or as much as $17-18 if everything goes right.
• Rivian is preferred over Lucid as an EV investment.
• Rivian’s R2 is expected in the $40,000-$45,000 range.
• The R2 is positioned as a smaller SUV and a possible Model Y alternative.
• Callo says the U.S. energy buildout has about a decade of visibility.
• GE Vernova is said to be sold out to at least 2035.
• Labor, especially electricians, is presented as a major bottleneck for energy and data-center construction.
• Continuous over-the-air changes may accelerate vehicle obsolescence and affect used-car values.
• Battery degradation concerns are described as mostly a cold-weather issue, with warranties covering many cases.
• The U.S. is described as building around 30 MW of power per year, according to Eisman’s closing recap.
• GE Vernova turbine orders can take until 2030-2031 to reach utility operation, underscoring the long cycle.
• GE Vernova’s electrification and nuclear businesses could become meaningful by 2035.
• Labor scarcity is called out as one of the biggest constraints on the buildout.