Joel Litman is running a promotion for his Hidden Alpha newsletter built around something he calls “Dark Energy.” The name is borrowed from physics — the mysterious force that accelerates the expansion of the universe. Dark energy makes up roughly 68% of the cosmos. Nobody knows what it is. It’s the perfect name for a mystery.
What Litman is actually talking about is not a mystery. It’s natural gas turbines. Specifically, aeroderivative gas turbines — jet engines bolted to generators, mounted on flatbed trailers, burning natural gas to produce electricity behind the meter of AI data centers.
The branding is aggressive. The technology is real and proven. The gap between them is where this guide lives.
What Dark Energy is not
Let’s clear the field first because the name invites speculation.
It is not nuclear. SpaceX is not building mini reactors. It is not fusion — we are likely a decade out from grid-scale fusion, and SpaceX hasn’t shown up in that R&D space. It is not solar, wind, geothermal, or hydropower. It is not a new physics breakthrough or a fifth fundamental force.
It is natural gas combustion driving a turbine that spins a generator. The same basic thermodynamic cycle that has powered aviation for 80 years, adapted for stationary power generation. The novelty is not the technology. It is the application — deploying it at data center scale, behind the meter, without waiting for the grid.
What aeroderivative turbines actually are
Aeroderivative gas turbines are exactly what they sound like: jet engines that have been adapted to sit on the ground and spin a generator instead of a fan. Companies like GE, Siemens, and Rolls-Royce have been building them for decades. They are smaller, lighter, and faster to start than industrial frame turbines — the massive units that utility-scale power plants use.
An aeroderivative unit the size of a shipping container can produce 30 to 60 megawatts. Multiple units get ganged together in parallel for larger loads. They can go from cold start to full load in under ten minutes. Compare that to a combined-cycle gas plant that takes hours to ramp or a coal plant that takes days.
The most relevant specs: 99.7% uptime. That is not theoretical. That is the operating record of the installed fleet. Natural gas turbines in the field hit that number because the technology is mature, the supply chain exists, and the maintenance cycles are well understood. Every major airline depends on the same core technology running thousands of hours between overhauls.
What SpaceX is actually doing
SpaceX disclosed the deployment in an SEC filing, buried on page 224. The company is installing natural gas turbines at multiple sites across Tennessee and Mississippi. These are not experimental. They are operational.
The scale is significant — more than double Microsoft and Amazon’s biggest power projects, according to Litman’s coverage of the filing. State governments are waiving taxes on the sites. That’s usually a sign of industrial policy urgency, not speculative hype.
xAI’s Colossus cluster in Memphis runs 46 natural gas turbines with 1.2 gigawatts of permitted capacity. That cluster went from dirt to training in months, not years, because they did not wait for the local utility to build transmission. They brought generation to the compute.
This is the pattern that matters. Every major hyperscaler is facing the same bottleneck: interconnection queues that run five to ten years, transformer shortages with 18-month lead times, and local utilities that have never seen load growth at this rate. Northern Virginia is rationing power. Dallas-Fort Worth has 40 GW of interconnection requests sitting in queue. Dublin has a moratorium. The grid was not built for AI.
The workaround is moving generation to the load. Gas turbines on a flatbed, connected directly to the data center switchgear, burning pipeline natural gas. No utility middleman. No transmission loss. No queue.
Why the name “Dark Energy”
Litman is a marketer. The physics term is a natural metaphor — an invisible force that nobody can see, that is driving massive expansion. It fits the promotional narrative: hidden alpha in plain sight.
The real technology is not invisible. It is not mysterious. It is well-understood, well-documented, and operating right now. Calling it “Dark Energy” is a rhetorical choice, not a technical description.
That does not make the investment thesis wrong. Litman’s argument is that the AI data center buildout will drive massive demand for behind-the-meter power generation, and that most investors are watching the hyperscaler stocks while ignoring the companies that build and operate the infrastructure underneath. That thesis is grounded in real market dynamics. Goldman Sachs, Jefferies, and other institutional shops have published similar analyses. The packaging is aggressive. The underlying logic is solid.
The unsung workhorse of the AI buildout
The AI industry has a compute problem and an energy problem, and the energy problem is the harder one. Compute gets cheaper per FLOP. Power delivery gets harder per megawatt. Moore’s Law bent the cost curve on chips. Nothing has bent the cost curve on transformers, transmission lines, and interconnection approval times.
Natural gas turbines are the only technology that can add gigawatt-scale power in months today. Solar and wind are intermittent. Batteries can buffer but not sustain — a 1 GW data center with six hours of battery backup costs roughly $2 billion at current pack prices. Small modular reactors are five to ten years away from meaningful deployment. Fusion is a decade-plus out.
Gas turbines exist. They are modular, dispatchable, and fast to deploy. A 300 MW gas plant can be permitted and built in 18 to 24 months in favorable jurisdictions. The existing fleet has 99.7% uptime. The fuel is abundant. The United States has the largest natural gas reserves in the world and the infrastructure to deliver it.
The carbon trade-off is real and acknowledged by every hyperscaler running this play. Natural gas is not zero emission. It emits roughly half the CO₂ of coal per megawatt-hour and a fraction of the particulates. Every operator has made the same calculation: better to burn gas and train today than wait for carbon-free electrons that do not exist yet. The bridge is real. The question is how long it runs before hydrogen blending, carbon capture, or small modular reactors take over.
What this means for the Hidden Alpha thesis
Litman’s Hidden Alpha newsletter is pitching this as a stock-picking opportunity. The idea is that the companies building and owning behind-the-meter gas generation are undervalued compared to the hyperscalers they serve. The market is watching NVIDIA and Microsoft while ignoring the firms that actually deliver the electrons.
That framing is consistent with the rest of Litman’s methodology. His background is forensic accounting. His firm Altimetry runs the Uniform Accounting model that corrects for GAAP distortions in company financials. His team of over 100 analysts covers 6,566 companies. The institutional arm, Valens Research, counts all ten of the world’s largest asset managers as clients, paying up to $100,000 per month for access. Hidden Alpha repackages that research at $79 for the first year.
The Dark Energy marketing campaign is the current wrapper. The research process underneath it has been consistent for years. The question for investors is whether the wrapper matters. If the underlying analysis is sound, the packaging is noise. If the packaging distracts from evaluating the thesis on its merits, that is a risk.
The underlying thesis — that AI data center load growth will strain the grid for years, that behind-the-meter gas generation is the most scalable near-term solution, and that most investors are not pricing this into the infrastructure stocks — does not depend on the name. It depends on the actual buildout. And the actual buildout is happening. SEC filings, interconnection requests, and state tax waivers are better sources than marketing copy.
The technology behind the name is real. SpaceX installing gas turbines in Tennessee is a fact. 99.7% uptime is a verified operating metric. 1.2 GW of permitted capacity at Colossus is a public number. The branding is a device to get attention. The attention is earned because the underlying development is genuinely large.
Dark energy makes up 68% of the universe and nobody knows what it is. Dark Energy the power play makes up a significant fraction of the AI buildout and it is very well understood. Jet engines on trailers, burning gas, running compute. The name sells. The technology delivers.