Coatue's Max Cook on the trillion-dollar race to power AI – and why it's cool to be a nuclear engineer again

In 2023, Max Cook, sector head at Coatue, stayed up until 3 a.m. tracing AI's infrastructure demands — and came up with projections that shocked his boss.

Coatue was betting big on the chips powering AI, with investments across semiconductors. Cook took the assumptions that had informed his team’s investments and ran the numbers: How many data centers would these chips require? And how many gigawatts of power would those data centers need?

The result: US electricity demand, which had grown by 0-1% annually for the past 50 years, would surge 5-10% per year for the next two decades.

“Those kinds of growth rates can make sense in fast-scaling technology markets,” Cook said. “But when you translate them into physical infrastructure at trillion-dollar scale, the magnitude becomes very real, unbelievable even.”

Over the following three years, industry forecasts have crept closer to Cook's curve. The question — where will the power come from? — now guides Cook’s day-to-day professional life.

But Cook’s interest has a personal dimension to it, too. His grandfather was a nuclear engineer in the 1970s, and helped to build America's nuclear plants while working with the Atomic Energy Commission (the federal agency that preceded today’s Nuclear Regulatory Commission), the NY State Atomic and Space Development Authority, and engineering firms like Pratt and Whitney. He started at the dawn of the Atomic Era commissioning plants and designing new nuclear technologies, and spent the end of his career helping to decommission some of those very same plants.

"He's 90 now." Cook says. “And is a pragmatic former nuclear engineer who always said shutting down those nuclear plants was the silliest thing we'd ever done. It seems that everyone is finally starting to agree with him."

A new generation is now taking up this work. Cook spends his days tracking the nuclear renaissance, following in his grandfather's footsteps. In his office, he keeps a mockup from the 60s of a small modular reactor design his grandfather worked on at General Electric. It's a reminder that some of today's breakthrough ideas have been waiting decades for their moment.

Max Cook

Follow the bottleneck

Cook sums up his job succinctly: follow the bottleneck.

AI's infrastructure boom has made Cook’s focus on public markets especially valuable, though it required a shift in his thinking about opportunities. Cook started hearing about grid equipment constraints in 2022, after pandemic-era semiconductor shortages. His initial reaction was skepticism. “Coming from a background of investing in fast-moving technology cycles, I initially assumed these bottlenecks would be short-lived.”

But then he realized hyperscalers weren't just dealing with tight supply: they were buying grid equipment directly from manufacturers, adding massive demand on top of already-constrained supply.

Cook’s investment thesis crystallized: listen to what developers and technology companies consistently lack. “Bottlenecks in the physical world tend to drive sustained demand and confer strategic importance,” he explains. “When supply has been underbuilt and demand accelerates, those constraints can persist for years.”

This thesis gained validation over time. In 2023, increased attention across the market focused on companies such as Eaton and Schneider Electric, manufacturers of electrical transformers and switchgear.

“No one knew what a transformer was back then," Cook says. "And now there are articles weekly in mainstream news outlets about three-phase transformer shortages."

“Transformers are massive pieces of equipment that step down electricity from the grid to voltages usable by buildings and data centers,” Cook elaborates. “Because electricity demand growth had been essentially zero for decades, manufacturers had scaled back production. Then, AI created explosive demand.”

Over the years, Cook has followed the thread from chips to data centers to the ultimate constraint: building new generation. He now researches gas turbine manufacturers, solar panel producers, and nuclear equipment makers, as the infrastructure challenge reveals itself layer by layer.

The manufacturing constraint

To Cook, America’s AI buildout surfaces the tradeoff between being self-reliant and maintaining cost-competitiveness.

He explains that the US is in a race with the world to build as many data centers as possible, but many key inputs are produced much faster and cheaper overseas, including batteries and solar panels. “In certain areas of clean energy manufacturing, the US currently faces higher production costs compared with some overseas suppliers,” Cook says. “This reflects differences in scale, technology, and established supply chains.”

Cook notes that the challenge is less about raw material availability and more about manufacturing and processing capacity. “Key inputs such as silicon and lithium are widely available globally, but processing and refining capabilities have developed unevenly over time,” he says. Similarly, while uranium resources exist in multiple countries, enrichment and processing infrastructure is concentrated in a few regions, reflecting decades of investment patterns.

Since these supply chains were built over 50 to 100 years, they don’t reverse quickly. “Even if we had unlimited uranium in the US, it wouldn't necessarily mean that tomorrow we would be self-sufficient for nuclear fuel,” Cook explains. "There are multiple steps to take uranium out of the ground and have it be ready to be a nuclear fuel rod that goes in a nuclear plant.” This would require rebuilding the entire infrastructure. A few companies are now pursuing US domestic enrichment with government subsidies. “It’s a reindustrialization the US hasn’t seen for a while,” Cook says.

Meanwhile, the question comes down to how the US weighs geopolitical priorities: sponsoring US business and keeping low-cost foreign imports out, while winning the AI race, which requires as much electricity as possible.

The short-term response has been to build with what America can make. Cook gives the example of GE Vernova, one of the largest suppliers of gas turbines in America, which is seeing a lot of growth. The company uses natural gas in generation, which is not as clean as renewables or nuclear, but is buildable with American manufacturing. Cleaner alternatives like solar and wind, Cook clarifies, are intermittent energy sources that require battery storage, which also isn't cost-competitive to manufacture in the US.

Cook sees reshoring as the inevitable long-term trend. As the deglobalization and decoupling from China continues, Cook predicts that more battery plants will be made domestically. Companies like First Solar, a major American panel provider, will expand US solar panel facilities. And foreign semiconductor companies will agree to build production facilities in the US with incentives. “This will likely be a big trend for the next 10 years.”

The affordability crisis

Aside from the question of who manufactures power, there’s the question of who pays for it.

Currently, there’s a tension between AI supremacy and consumer affordability. “All things equal, more electricity demand with no new supply likely equals higher electricity prices," Cook says. “But we need to keep the lights on and make sure consumers aren't paying higher electricity bills.”

This tension regularly plays out in news headlines. This month, the White House's National Energy Dominance Council called for an emergency auction to build power plants. The auction was in response to a forecast by the PJM, the major electrical region serving Pennsylvania, New Jersey, and Maryland, that they would be 6 gigawatts short of their reliability requirement by 2028. The auction sought to procure new generation capacity while ensuring that consumer prices wouldn't spike.

This is a difficult problem to address, but Cook identifies three emerging possible regulatory solutions.

Cost allocation and grid impact

As data centers scale rapidly, regulators and utility giants have examined how to allocate the costs of serving very large electricity loads while protecting consumers. In response, several states have adopted frameworks that more closely align infrastructure costs with large-scale users. Some hyperscalers have also made voluntary pledges not to increase costs for retail consumers, including Microsoft with its “community first” initiative.

“These companies are increasingly focused on their role in local electricity systems,” Cook says. “Many are willing to take on additional cost and work to sponsor new supply.”

Linking demand to new supply

There is growing attention on whether new electricity demand is accompanied by incremental generation capacity, a concept often referred to as “additionality.” In practice, hyperscalers have taken different approaches to matching new demand with new supply, ranging from pairing data center development with renewable energy and storage commitments to sequencing energy procurement alongside phased buildouts.

Permitting and infrastructure timelines

A key challenge is the pace of infrastructure development. Significant generation capacity remains in interconnection queues, reflecting the complexity of permitting and grid integration.

“There are different operating speeds across these groups: technology companies, energy companies, and government,” Cook observes. “How those timelines align, or are brought more in sync, will shape the path forward.”

Some progress is underway. Cook notes that “common sense regulatory reform” has started to accelerate permitting for critical energy assets and data centers that meet certain standards.

But the mismatch in speed between different players remains stark.

Gayatri Sabharwal (Ramp) in conversation with Max Cook (Coatue)

The nuclear renaissance

Today, Cook spends most of his time researching the nuclear bottleneck.

"From a first principles standpoint, nuclear is the best source of energy we have – most energy-dense and with the lowest emission," Cook argues.

The opportunity with nuclear is exciting to him. The US currently has about 100 gigawatts of nuclear capacity across 94 reactors, representing roughly 20% of the country's electricity generation. “This number is higher than some people think,” he adds. But it hasn't grown since the late 1980s. The last nuclear reactor to come online was in Georgia in 2023, almost a decade delayed and 3x over budget. Currently, nuclear has bipartisan support, and the startup world is responding. Cook estimates that there are roughly 50 companies working on small modular reactors (SMRs) right now.

SMRs are designed to be built in factories rather than constructed on-site as bespoke megaprojects. Smaller designs – 300 megawatts, 80 megawatts, even smaller – that are "walk away safe" (can shut down automatically without human intervention) can be built in a more repetitive, factory-line process. This approach targets the cost and regulatory problems of traditional gigawatt-scale plants.

Two paths have emerged within the SMR world. Generation 3 reactors, which are scaled-down versions of existing designs, are being developed by traditional industrial companies like GE Vernova, Mitsubishi Heavy, and Rolls-Royce. Generation 4 reactors represent an entirely new technology pursued by venture-backed startups – higher risk, but potentially transformational.

Cook notes that while outcomes will vary, the opportunity is significant. “These technologies have the potential to form an important part of the future energy mix and contribute meaningfully to addressing long-term supply constraints.”

There’s some promise already, with hyperscalers making nuclear bets. This year, Meta announced a partnership with Oklo for new nuclear capacity in Oklahoma. In recent years, Amazon partnered with X-Energy, and Microsoft made a deal with Constellation Energy to bring the Crane Clean Energy Center, formerly Three Mile Island, back online.

“There’s growing experimentation with new energy generation technologies in the US, which hasn’t happened at meaningful scale in some time. That has the potential to influence how the broader energy ecosystem evolves.”

As a product of these changes, there's something less tangible happening. "Lots of really smart people have realized it's cool to be a nuclear engineer again," Cook says.

Max Cook is Sector Head at Coatue, and specializes in AI infrastructure and energy investments. You can follow Cook on X, and Coatue on LI and X, and keep an eye out for C:\Takes.