Hyperscalers are bypassing grid queues by ordering 400 MW gas turbines directly; GE Hitachi and peers are adding assembly lines but cannot meet 7-year backlog, creating a seller's…

Caterpillar's Solar Turbines division produces 38 MW gas turbines on semi-trailers; 40 units make a GW, and even these smaller units have backlogs, showing depth of behind-the-met…

Boom Supersonic is pivoting its supersonic jet engine design to manufacture natural gas turbines for data center power, capitalizing on extreme demand and 7-year backlogs for larg…

Emerald AI's workload migration software lets data centers shift compute across sites to match grid capacity, a capital-light way to unlock stranded grid capacity without new wire…

Emerald AI enables data centers to act as interruptible loads via software job migration, unlocking 100GW of grid capacity without new construction; backed by Nvidia and Rome Nam.

Agent Gentic deploys 4-hour battery storage at data centers to charge at night and avoid peak grid draw, qualifying for fast-track interconnection under new Texas/FERC interruptib…

Agentic deploys 4-hour battery storage at data centers to charge at night and avoid peak draw, qualifying as 'interruptible load' under new Texas/FERC rules that cut interconnecti…

Weave Grid's software time-slices EV charging at the neighborhood transformer level, reducing last-mile grid stress and creating headroom for AI data center loads; Rome is an inve…

Elon's stated goal to build 100 GW of terrestrial solar is a stepping stone to develop the manufacturing capacity needed for space-based solar arrays that would power orbital data…

AMD's hardware is now on par with Nvidia's, and the ability to use AI to recompile CUDA code for AMD chips eliminates the software switching cost that protected Nvidia's dominance.

Nvidia's CUDA software moat is being broken by AI tools that can automatically recompile code to run efficiently on competitor chips like AMD and Cerebras, eroding Nvidia's pricin…

Lamorian is developing software that automatically recompiles CUDA code to run efficiently on AMD, Cerebras, and other non-Nvidia hardware, directly attacking Nvidia's software mo…

The Nuclear Company aims to finance and build fleets of AP1000-style reactors simultaneously to amortize first-of-a-kind costs and build supply chain momentum, solving the single-…

Valar Atomics secured a $1B raise at a $6B valuation for a small modular reactor design despite not yet having a working reactor, signaling intense investor appetite for nuclear f…

Nam called X-Energy an amazing company and said he liked its reactor design immediately after another speaker noted that the company had recently gone public.

Aalo's factory-built, shipping-container-scale reactor design maximizes manufacturing learning rates and minimizes on-site construction, making it Rome's top pick in the SMR space.

CFS uses high-temperature superconducting tapes to shrink tokamak magnets dramatically, enabling a 600MW commercial reactor vs. ITER's 5GW; it has the most experimental validation…

CFS leverages proven tokamak physics from ITER but uses novel high-temperature superconducting magnets to dramatically shrink reactor size and cost, targeting 600 MW net energy vs…

CFS leverages high-temperature superconducting magnets to shrink tokamak reactors from 5 GW (ITER scale) to ~600 MW, dramatically reducing capex and cooling needs; it has the stro…

Helion's pulsed magneto-inertial fusion captures electricity directly without steam turbines, offering a path to lowest cost per kWh; Microsoft has signed a 50MW power purchase ag…

Microsoft's PPA with Helion validates fusion's near-term commercial viability and provides a revenue anchor for Helion's 2028 demonstration plant.

Avalanche Fusion aims for a reactor small enough to power a vehicle, representing the most ambitious compactness goal in fusion but with lower scientific confidence than tokamak a…

Launching 10 GW/year of orbital data centers requires 5-6 Starship launches/day; FAA safety regulations (grounding after any failure) and 200 launches/year practical limit make th…

Factory-built, wave-powered data center buoys in the Southern Ocean achieve ultra-cheap baseload power (2¢/kWh) and free 40°F seawater cooling that extends GPU life; Rome regrets…

This company deploys factory-built, bobby-pin-shaped data centers on the ocean that generate power from wave motion (targeting 2¢/kWh) and use seawater for free cooling, reducing…

Next-gen geothermal (Fervo, Quaise) can unlock cheap baseload power anywhere, not just near volcanic hotspots, potentially joining solar, nuclear, space, and ocean as a scalable A…

Fervo applies oil-and-gas horizontal drilling to geothermal, enabling scalable baseload power anywhere; CEO Tim Latimer is a known contact of Rome's.

Millimeter-wave plasma drilling enables 10-20 km depths, accessing superhot rock everywhere, which would make geothermal a globally deployable baseload source for AI data centers.

Quaise's gyrotron-based plasma drilling can reach 10-20km depths, making geothermal viable anywhere on Earth, not just near volcanic hotspots.