Nvidia's $3 Billion Power Bet Isn't About Fear — It's About Control

The easy narrative here is that Nvidia is losing control of the AI stack. The chips are just silicon — the real bottleneck is power, the argument goes, so Nvidia has to go downstream into energy infrastructure to stay relevant. I've been very surprised that this framing has gained traction, because the data tells a different story.

Nvidia agreed on Friday to invest up to $3 billion in power infrastructure developer Lancium, a Blackstone-backed power infrastructure developer whose 1,000-acre campus in Abilene, Texas, anchors the Stargate AI data center project. The deal values Lancium at approximately $10 billion in enterprise value, with the additional $1 billion tranche conditional on Lancium hitting specific thresholds, including securing grid connections.

But let's look at the numbers before reaching for the narrative.

Nvidia generated $119.1 billion in free cash flow over the trailing twelve months. Its FCF margin is 47%. The company holds $13.2 billion in cash against $6.4 billion in total debt — a net cash position of roughly $72 billion. That $3 billion commitment to Lancium represents about 2.5% of Nvidia's annual free cash flow.

This isn't Nvidia scrambling for survival. This is the most profitable company in the semiconductor industry buying a minority stake in the power layer of its largest customer ecosystem. It's not a response to weakness. It's an assertion of control.

Who is Lancium, and why does a $10 billion valuation deserve a skeptical eye?

Lancium was founded in 2017 as a Bitcoin mining power infrastructure play. Its proprietary "Smart Response" technology matched flexible mining loads with intermittent renewable energy — essentially turning wind power into hash rate when the grid had excess capacity. In 2024, Lancium pivoted hard to AI, partnering with Crusoe Energy to build a 200-megawatt data center just outside Abilene, Texas, designed to house Nvidia's Blackwell GPUs. Blackstone invested more than $500 million in Lancium for roughly a 50% ownership stake.

The campus is designed to scale to 1.2 gigawatts of capacity — enough power to support approximately 400,000 Nvidia AI chips. That is the physical infrastructure layer of Stargate, the joint venture between OpenAI, Oracle, and SoftBank that was announced in January 2025 with a stated ambition of up to $500 billion in AI infrastructure investment.

A $10 billion enterprise value for a company that pivoted from Bitcoin mining to AI data center power less than two years ago, on land it acquired before the AI boom, requires enormous confidence that Stargate delivers on its stated scale. Stargate's initial expansion plans have already been revised once. Crusoe and Oracle are building the first two buildings — 200 megawatts — at a cost of roughly $1.1 billion for the infrastructure alone. The remaining 1 gigawatt is still drawings on paper.

In my opinion, Lancium's valuation front-runs a great deal of execution risk. But that's a Lancium problem, not an Nvidia problem. Nvidia is putting down 2.5% of its annual free cash flow for a non-controlling stake with milestone-based earn-outs. The upside is guaranteed GPU placement at the world's most ambitious AI campus. The downside is capped.

This is supply chain security, not strategic panic

The false narrative I see forming around this deal is that Nvidia is acknowledging the limits of its moat. The reality is the opposite. Nvidia's data center revenue hit $51.2 billion in the third quarter of fiscal 2026 and continues accelerating — 73% year-over-year revenue growth for the full fiscal year, with 70% year-over-year revenue growth continuing into the current period.

What Nvidia is doing here is the same logic that drives its cross-pollination model across the AI stack. Google's TPU demand creates revenue for Nvidia's networking chips. Amazon's custom silicon investments require Nvidia's software ecosystem and reference designs. Each player's capital expenditure reinforces the others' demand. Nvidia doesn't need to win every layer — it needs to be embedded in every layer.

By taking a stake in the power infrastructure at Stargate, Nvidia ensures that 400,000 GPUs at Abilene are its GPUs. Not AMD's. Not a competitor's. The company is effectively pre-booking its product into the physical plant that will house it. That's not panic. That's how dominant suppliers behave when they can afford to own their supply chain.

What about the risk?

There's one legitimate counterargument worth examining. If power infrastructure becomes the binding constraint on AI deployment, and companies that control power can dictate terms to chipmakers, then Nvidia's leverage could erode over time. Microsoft is already restarting Three Mile Island for its own data centers. Meta signed a 20-year nuclear power purchase agreement with Constellation Energy last year. Walmart just signed its own nuclear deal with Constellation in June. The hyperscalers are building energy independence.

That trend is real. But it doesn't threaten Nvidia — it reinforces Nvidia's incentive to be present in the power layer. Every AI data center that comes online needs GPUs. Whether it's powered by a natural gas turbine in Texas or a nuclear plant in Pennsylvania, the computational layer still runs on Nvidia silicon. Nvidia's 89% return on invested capital and 64% operating margin don't depend on who owns the generator.

What does matter is that Nvidia's earnings report on August 26th will show whether hyperscaler capex — the engine behind deals like Stargate — is sustaining the pace that justifies the company's current valuation. The stock trades at 34 times trailing earnings and 60 times forward earnings, above Microsoft at 28x trailing and Google at 18x. Nvidia is only cheap relative to its own growth trajectory: the PEG ratio of 0.31 implies the market is pricing in significantly slower growth than the 70%+ revenue acceleration the company has been delivering.

That being the case, I maintain my Buy rating on Nvidia. The Lancium investment is a small, well-structured allocation by a company sitting on a mountain of free cash flow. It's not evidence that chips are becoming commoditized or that Nvidia is losing its grip. It's evidence that Nvidia is doing what it does best — embedding itself deeper into every layer of the AI infrastructure stack so that even as the hyperscalers build out their own power plants, they still need Nvidia's silicon to make those plants profitable.

For investors who already hold Nvidia, this deal doesn't change the thesis. It confirms it. For those on the sidelines worried that the AI infrastructure buildout has shifted downstream, the structural data says the chips are still the bottleneck — they just now come with a power stake attached.