Building the physical backbone of AI: inside a 2.1-gigawatt data center campus and the economics that drive it
Chase Lockmiller frames AI infrastructure through a Cobb-Douglas economic lens: AI effectively creates digital labor for the first time in history, a factor of production previously constrained by birth rates and 20-year lead times. That unlocks an unprecedented GDP growth opportunity, which is why hyperscalers and sovereigns are committing capital at a scale the host compares to the Manhattan Project and the highway system combined.
The physical build-out Lockmiller describes is anchored in Abilene, Texas — chosen because stranded wind and solar energy was trading at negative prices due to a lack of transmission and local demand. Crusoe's strategy is to move data to the power rather than move power to the data. The Abilene campus now totals 2.1 gigawatts, hosts Oracle/OpenAI (Project Stargate) and Microsoft, includes a 1 GW privately owned substation (claimed to be the largest in the US), a 350 MW gas plant, and 9,000 workers on site daily in a city of 120,000.
The all-in capital stack runs roughly $60M per megawatt: ~$20M for the data center shell and power plant, and ~$40M for the compute cluster (of which $30M is GPUs, $4M networking, $3M CPUs/storage). On the revenue side, bare-metal GPU rental yields ~$15M/MW/year; adding a managed inference layer can push that to $25–30M/MW/year, implying a 2–4 year payback. Key inflation pressures are real: gas turbines have tripled to ~$3M/MW, labor is $4.7M/MW, and skilled tradespeople — electricians, welders, plumbers — are in acute shortage.
Several emerging themes surfaced beyond the core build: - Water: Lockmiller pushes back on the narrative that AI data centers are heavy water consumers, claiming closed-loop liquid cooling uses roughly the same annual water as a single-family home after initial fill — a contested claim. - Power electronics: Traditional transformer incumbents (Eaton, Schneider) face long-term disruption from solid-state transformers and the shift toward 900V DC rack power, even as they benefit in the near term from surging demand. - Open source AI is seen as winning against closed-source models over time. - Space data centers: Crusoe has partnered to put H100s in orbit, but Lockmiller is candid that thermal management, GPU failures, and economics make this unlikely to be material for 10 years — contingent on Starship reducing payload costs by two orders of magnitude.
The bottom line is that power capacity — not chips — is now the binding constraint on AI scaling, and whoever controls energized, permitted, connected land holds the strategic asset. Lockmiller sees the current moment as a once-in-a-generation infrastructure build, but flags that cost inflation across gas generation, labor, and chips is real and compressing margins for anyone who doesn't control the full stack or find stranded energy arbitrage.
Chase Lockmiller, CEO of Crusoe, walks through the full stack of AI infrastructure — from stranded renewable energy in West Texas to GPU racks, gas turbines, and space data centers. He breaks down the capital costs (~$60M/MW all-in), revenue potential (~$15–30M/MW/year), and the hard constraints — power, tradespeople, water — that are now the real bottlenecks. The conversation frames AI infrastructure as a macro-economic force comparable in scale only to the US defense budget.
the reason these investments are taking place and the reason it's so broad-based is that there's an opportunity to completely transform the economy by really accelerating the growth in GDP and fundamentally up-leveling and improving people's quality of life by just seeing an unprecedented level of growth.
over time margins do probably come down to more standard stabilized silicon margins, call it like 60% gross margin
Focus less on the what and focus more on the how, and leverage AI tools to run and live your life
This is one of the biggest investments that we are making. This is bigger than space, bigger than our highway system, our uh the Manhattan Project, second only to the US defense budget.
Chase is a very avid mountaineer. Five out of the seven summits across the world including Everest.
for the first time in history, what we're able to do is we're able to actually create this sense of digital labor
historically, labor is this thing that you could really only change via the birth rate, right? And it's got a 20-year lead time
today, the core bottleneck is like powered like energized data centers, right? Like power shells where you can plug in chips and and start operating a big GPU compute cluster.
Chips is kind of softened as the bottleneck. You know, I think access to chips has become more available. It's really finding places where you can put those chips and turn them on.
in Abilene, Texas, many folks had never heard of until we put a shovel in the ground in Abilene. So, why do we go to Abilene? Abilene is this area of West Texas that is consistently very windy windy and very sunny.
a lot of renewable energy developers had gone there to build out uh large-scale renewable energy production because they were incentivized by uh something called production tax credits, where they basically get paid by the government to produce clean electrons.
power prices were actually negative, right? Because there was no marginal buyer for this power. And there wasn't enough transmission to get the power to somewhere where it actually was useful.
the top one is a 200 MW substation. The second larger one is a 1 GW substation.
that gigawatt substation, first of all, that's the largest privately owned substation in the United States.
a gigawatt is basically what powers the whole city of Denver, right?
those first eight buildings are all for Oracle and OpenAI. Um so, it's basically what was uh known as Project Stargate
we also built down here a natural gas power plant. This is roughly a 350 MW natural gas power plant um to support the development and energize uh this giant computing cluster.
all of the chips across all of the data centers were interconnected on the same high-performance back-end network to be able to operate as one coherent workload. So, you could run one training job that runs on all of the chips on the entire data in the on all the data centers together
we had to build this parking lot over here. Like that parking lot is like a it's a 5,000-car parking lot, and you can see it's totally full because we have, you know, we have roughly 9,000 people on site every single day that are working to bring this campus to life.
That is for Microsoft. So, the campus is 2.1 gigawatts in aggregate.
The population of Abilene is 120,000.
in the long term there is like a a steady job population that's operating these large uh computing clusters and power plants that are operating the infrastructure. That staff is somewhere in the neighborhood like 2,000 people
Each building has about 1 million gallons of water in the building to cool these chips.
We use like zero water. We fill this system one time, and then on an annual basis, we use about the same amount of water as a single family home.
the full power plant plus building costs is 4.7 million dollars per megawatt for labor
for a gigawatt, you know, that becomes 4.7 billion dollars
the gas plant is probably two to three million dollars per megawatt
a gas turbine that used to cost a million dollars a megawatt now costs three million dollars a megawatt
gas turbine manufacturers are basically GE Vernova, Siemens, Mitsubishi Heavy Industries, Pratt & Whitney, Caterpillar has a company called Solar
today I think we have 3,500 people at this site. It's in a town called Quan, Texas, which is a town of 1,500 people
the IT CapEx per megawatt is roughly 40 million per megawatt
30 million of that is going to the GPUs
4 million roughly to the networking
there's 72 GPUs in that rack, they're all interconnected on the same NVLink domain
we're seeing a massive shortage of CPUs because with the boom in all of these agentic workflows, with the boom in Claude, you need a lot of CPUs to actually orchestrate those compute workloads
about 3 million a megawatt for CPUs and storage
the full stack upfront CAPEX is close to 60 million per megawatt — roughly 20 million for data center and power plant, and 40 million for the compute cluster
for each 100s that initially debuted in about 3 years ago, the pricing had come down, but with the boom in demand coming from agents, the price of H100s has actually come up and actually exceeded the price that folks were paying when these chips first came out
ongoing OPEX of this plant is roughly a little over 1 million per megawatt
revenue from renting chips is roughly 15 million per megawatt annualized
adding a managed services layer where you're actually serving the model improves revenue by anywhere from 5 to 15 million per megawatt, resulting in up to 30 million per megawatt per year and a 2-year payback
Crusoe Spark, our modular self-contained modular AI data center, achieves call it 30 to 50% savings depending on overall costs, so that 19 million a megawatt we can actually bring down pretty dramatically
each unit for our air-cooled architecture is 500 kilowatts and we have a liquid-cooled version that's 2 megawatts
today Nvidia's commanding like 80% gross margin
there's a new transmission line going in Texas that's 765 kV
not everything needs 100% five nines of reliability full backup, but the core storage and networking systems do
infrastructure costs — specifically gas generation and labor — are going up, not down, due to demand-driven price inflation
Solid state transformers and power electronics put in jeopardy companies like Eaton and Schneider that have not innovated much in the last 100 years
Companies like Eaton and Schneider will do very well in the near term
There are going to be huge shifts to like 900 V DC and all these different aspects of the electrical stack
Innovating on how to get power from 765 kV to like 900 V DC in the rack is a super super huge opportunity
Open source is winning and will take more from closed source model players
Crusoe has established a partnership with Star Cloud, which has actually launched the first H100s into space
Space data centers eliminate the need for concrete foundations, permitting, power-side approvals, and millions of strands of fiber
In space data centers, everything is basically optically interconnected
Natural GPU deprecation in space data centers will create challenging economics
Space data centers' viability depends on whether Starship brings payload cost down by two orders of magnitude
Space data centers will probably not be material in the next 5 years, and probably not material for 10 years
Over a longer period of time, data centers in space are going to play a major role in sort of the future of intelligent infrastructure
We all have the workforce of a million people at our fingertips with mass adoption and utilization of AI, which is fundamentally going to change work
there's this ongoing narrative that like AI is like taking all the water
We don't have enough of these tradespeople. We don't have enough electricians. We don't have enough welders. We don't have enough plumbers. We don't have enough construction workers.
Over the long term, if they don't innovate, that whole piece of the stack is going to dramatically come down in cost
Thermal management in space data centers is very challenging
GPUs in space data centers fail and have to be re-seated or RMA'd, and you're not sending an astronaut to space to take a chip and send it back to Jensen
AI is the combination of data, algorithms, uh so backpropagation, neural networks, you know, transformer architectures, all these different algorithms that people have come up with to to essentially statistically model, you know, the the data sets, compute large amounts of compute, and particularly this this high-performance computing infrastructure through GPUs, where you're able to parallelize a lot of the work load workloads and do a lot of this tensor math in a high-performance architecture. Energy that's required to run those GPUs. And then data centers, the physical buildings that house and operate all of this computing infrastructure.
There's there's sort of like this this economics model for production called the Cobb-Douglas model. So, like if you look at like the growth in GDP, the growth in GDP is fundamentally the sum of like three key components. The the change in labor, the change in capital. So, like, you know, and this capital sort of includes both like physical capital like buildings and plants and equipment as well as like capital that's invested into an economy. And then the change in technology.
four-year payback period on a 60 million per megawatt investment at 15 million per megawatt in annualized revenue
across the meter strategy: on-site generation through wind, solar, batteries, and gas; sell surplus to the grid; draw from the grid when needed to firm up power
One of our core values at Crusoe is actually living on the infinite growth loop — this notion that nobody's a finished product, everybody's a work in progress
we said, 'Can we find areas that aren't historical data center markets and go there and build uh data centers where we can instead of having to move the energy, we're actually moving data.'
“you know, I don't know what the world's going to look look like 5 years from now with like the mass adoption and utilization of AI where, you know, we all have the workforce of a million people at our fingertips”
Shared with Earmark · earmark-ai.com