Nvidia's $3 Billion Power Bet Signals Where the AI Bottleneck Actually Is

Nvidia is reportedly preparing to invest $2 billion upfront — with an option for another $1 billion — in Lancium, a Blackstone-backed power infrastructure developer behind OpenAI's Stargate data center campus in Texas. The deal values Lancium at roughly $10 billion and would give Nvidia an initial 20% stake, rising to about 30% if grid interconnection milestones are met.

Neither Nvidia nor Lancium confirmed the report, which originated from The Information on Friday. But the deal structure — and what comes before it — tells a story about where this company is heading that goes well beyond chip sales.

The bottleneck has moved upstream

Jensen Huang described AI as driving "an industrial transformation" last week, noting that demand for data center power is putting the U.S. energy sector under severe pressure. That's not rhetoric for his audience. It's the signal layer.

The real constraint on AI compute in 2026 isn't whether Nvidia can manufacture enough Blackwell or Rubin chips. It's whether those chips can get power.

The numbers make the bottleneck obvious. Texas' ERCOT grid operator is tracking approximately 474 gigawatts of interconnection requests — more than five times the state's all-time peak demand record. About 90 percent of those requests come from data centers. On August 3rd, Governor Greg Abbott ordered a pause on new data center grid connections pending a comprehensive audit of power and water usage. ERCOT has halted its first batch of approvals, with no clear timeline for resumption.

This is what separates the current cycle from the last one. In the dot-com era, the bottleneck was network bandwidth and fiber. In the AI era, it's watts. And Nvidia has decided to buy into the bottleneck directly.

Lancium: the asset Nvidia is buying

Lancium operates the 1,000-acre Lancium Clean Campus in Abilene, Texas — the first operational site of Stargate, the joint venture between OpenAI, Oracle, and SoftBank announced at the White House in January 2025 with a stated $500 billion investment target. Lancium's role is to develop and secure the power infrastructure that makes these campuses possible.

The company has secured 4 gigawatts of power resources on the Texas grid across multiple projects:

• 1.2 GW allocated to the core OpenAI/Oracle Stargate campus in Abilene
• 900 MW to Crusoe Data Center for Microsoft
• 1.0 GW to QTS Data Centers
• 1.0 GW to a Crusoe project in Childress, currently under construction

Beyond that, Lancium has land banks positioned for an additional 15 GW of projects awaiting grid interconnection permits. The Abilene campus itself relies heavily on on-site natural gas generation — Crusoe and Lancium are building a campus that can operate largely off-grid using over 1 GW of gas turbines. That's relevant because Abbott's interconnection pause exempts data centers using purely behind-the-meter on-site generation. Lancium's strategy is already adapted to the regulatory environment.

Sources indicate Lancium is considering an IPO in 2027. A week before this deal was reported, Lancium was in talks with major tech companies about a minority stake, with Anthropic holding exploratory discussions. Nvidia appears to have won that race.

This isn't a one-off

The Lancium deal is the latest in a rapid sequence of equity investments that marks a structural shift in how Nvidia operates. During the quarter ended April 2026 alone, Nvidia invested $18.6 billion into private companies and infrastructure funds — surpassing the entire volume of the previous year.

Look at what Nvidia bought in the first half of 2026:

• CoreWeave: $2 billion for roughly a 9% stake, giving Nvidia a near-10% position in the neocloud that's building 5 GW of capacity by 2030
• Nebius: $2 billion to expand AI cloud infrastructure through 2030
• Coherent and Lumentum: $4 billion split evenly, a "photonics push" to secure optical technologies as data centers shift from copper to fiber for high-speed transmission
• Marvell: $2 billion centered on NVLink Fusion, a platform integrating Marvell's custom silicon directly into Nvidia's ecosystem
• Corning: up to $3.2 billion to secure priority access to fiber optic cable

That's more than $15 billion across seven companies in 2026 so far, not counting the potential $3 billion Lancium commitment.

This is no longer a chip company that occasionally makes strategic investments. Nvidia is acting as the balance sheet for the AI infrastructure stack. It's buying into power, optics, networking, cloud operators, and custom silicon — every layer between its GPU and the end user.

What this means for the thesis

Nvidia's core financials remain exceptional. The company generated $81.6 billion in revenue for its first quarter of fiscal 2027, up 85% year over year. Revenue growth sits at 70.7% year over year, with sequential growth of 19.8%. Gross margins are 74.1%, operating margins 64.0%, and free cash flow for the trailing twelve months is $119.1 billion. Return on invested capital is 89.4%.

The stock trades at roughly $224, with a market cap of $5.4 trillion. The trailing P/E is 34x, but the forward P/E stretches to 60x — meaning the market is pricing in continued acceleration.

What changes when Nvidia starts writing $3 billion checks to power developers isn't the revenue story. It's the risk profile of that revenue.

By taking equity stakes in power infrastructure, optical suppliers, and neocloud operators, Nvidia is effectively creating a circular financing loop that binds its customers and suppliers to its hardware. CoreWeave's 2026 capex is roughly $35 billion — Nvidia's $2 billion investment is only 5.7% of that, but it comes with a $6.3 billion capacity buyback agreement running through 2032. Nvidia generates roughly $2 billion in revenue every three days. These investments are small relative to cash flow, but their strategic function is large.

The move secures demand certainty. If Lancium secures power, its customers (OpenAI, Microsoft, Crusoe) need GPUs to fill that capacity. Those GPUs come from Nvidia. The power-to-compute chain becomes a closed loop where Nvidia sits at the center.

However

There are two risks embedded in this shift that deserve explicit attention.

First, the Texas regulatory environment is actively hostile to the pace of data center development. Abbott's pause is a delay of indeterminate duration. ERCOT's Batch Zero transmission planning study — the first phase of new large-load approvals — is suspended. The state legislature convenes in January 2027, and regulators are expected to seek expanded statutory authority over the industry. Lancium's 15 GW pipeline of projects awaiting interconnection permits faces real timeline risk, and a $10 billion valuation for a pre-revenue infrastructure developer assumes those permits get approved on schedule.

Second, the circular financing model that Nvidia is building carries a leverage counterweight. The company has invested $18.6 billion in a single quarter into companies whose entire business depends on continued hyperscaler AI spending. If the neocloud model falters — if CoreWeave or Nebius face financing stress, or if hyperscaler capex decelerates — Nvidia doesn't just lose GPU orders. It loses equity value across a portfolio of companies whose valuations are tied to the same thesis.

Demand remains robust. The risk is concentration disguised as diversification.

Where the capital goes

I believe Nvidia's move into power infrastructure is the right strategic play. It locks in the most binding constraint on its own growth and extends its moat beyond the chip layer. The CUDA advantage in training doesn't protect Nvidia from inference competition or custom silicon — but controlling the power-to-compute chain does.

The question isn't whether Nvidia will remain central to AI infrastructure. The question is whether the stock's return profile over the next two years justifies its current allocation relative to what's available elsewhere.

At $5.4 trillion, Nvidia trades at 21x trailing sales and 60x forward earnings. Those aren't bubble-era multiples, but they're not entry-zone multiples either. Much of the near-term upside from power security, photonics supply, and neocloud expansion is already priced into a company that's beating estimates on 85% revenue growth.

If the Lancium deal closes and the broader investment pattern continues, Nvidia is positioning itself to dominate not just the chip layer but the entire AI factory. That supports the long-term thesis — I still believe the company can extend its dominance through the end of the decade. But the near-term return curve may be back-half weighted.

The debate isn't about whether Nvidia stays important. It's about whether the remaining return from here is as compelling as what can be found elsewhere in the AI trade. Companies deeper in the power and optical supply chain — or smaller neoclouds still trading at lower multiples — may offer a better risk-reward setup for the next two years.

For existing positions, the Lancium deal is a reason to hold, not to add. For new allocations, the signal is clearer in the companies Nvidia is buying into than in Nvidia itself — at least until the market finds a better entry point.