Nvidia's $3 Billion Lancium Bet: Why Power Is the New Bottleneck — and What It Means for Its Balance Sheet

Nvidia will invest up to $3 billion in Lancium, a Blackstone-backed power infrastructure developer, for a roughly 20% stake in the company that owns the electricity behind the Stargate AI data center campus in Texas. The initial $2 billion locks that stake immediately. An additional $1 billion is contingent on Lancium completing grid interconnection milestones for its campuses. The deal implies a $10 billion enterprise value for a company that pivoted from Bitcoin mining to becoming one of the most strategic power plays in the AI buildout.

The market is reading this as a Stargate expansion move. It is something deeper. Nvidia is positioning itself at the new constraint layer in AI infrastructure — electricity — and the Lancium deal is just the latest node in a $40+ billion web of supply chain investments Nvidia deployed in 2026 alone.

The bottleneck shifted

For most of the AI boom, the constraint was chips. Nvidia's CUDA-locked GPUs were the scarce resource, and whoever had access to Blackwell or Vera Rubin capacity could deploy AI models while competitors waited on backlogs. That bottleneck is closing. TSMC's multi-patterning expansion and Nvidia's own capacity ramp are pushing GPU supply toward the demand curve.

The new bottleneck is power. Building one gigawatt of data center capacity costs between $50 billion and $60 billion, and securing the electrical grid connection is the hardest part. Hyperscalers can write checks for chips; they can't write checks for grid capacity that doesn't exist.

Lancium has already locked in 4 gigawatts of contracted power on the Texas grid — 1.2 GW for the OpenAI/Oracle Stargate campus in Abilene, 900 MW for a Microsoft-backed Crusoe project in the same area, 1 GW signed by QTS Data Centers in Turkey, and 1 GW under construction by Crusoe in Childress. On top of that, the company has secured land parcels for an additional 15 gigawatts of pending power projects, waiting on grid interconnection.

That 15 gigawatt backlog is the real asset Nvidia is buying. Texas Governor Greg Abbott recently ordered a suspension of new data center grid connection applications for review. But Lancium's partner utility, American Electric Power, has already completed the necessary line studies for those pending projects, giving Lancium a first-mover advantage that doesn't come from owning power today — it comes from being in line ahead of everyone else.

Nvidia is becoming the AI infrastructure bank

The Lancium deal is the latest in a pattern that accelerates with every quarter. In the three months ending April 26, 2026, Nvidia invested $18.6 billion in private companies and infrastructure funds — more than it invested in the entirety of fiscal 2025, which itself totaled $17.5 billion. As of late July 2026, the company has committed more than $40 billion in equity stakes and investment rights across seven public companies and roughly two dozen private deals.

The ledger reads like a supply chain grab: $2 billion each in CoreWeave and Nebius (neocloud data center operators), $3.2 billion in Corning (optical connectivity), $2.1 billion in IREN (data center deployment), $2 billion each in Marvell, Lumentum, and Coherent (silicon photonics), and a $5 billion equity stake in Intel that is now worth over $25 billion. Non-marketable securities on Nvidia's balance sheet swelled from $3.39 billion at the end of fiscal 2025 to $43.36 billion by April 2026.

Total investment commitments subject to contingencies stood at $27 billion as of the first quarter of fiscal 2027, expected to be funded through the rest of the year. This is not a company that finished its supply chain bets. It is a company that treats capital deployment as a competitive strategy.

Put plainly: Nvidia is no longer just selling chips. It is financing the companies that buy them, the companies that make their components, and now the companies that power the buildings those chips live in. The chipmaker has become the capital allocator for the AI infrastructure stack.

What this is — and isn't

The Lancium deal is structured as a pure equity investment. Nvidia provided no credit guarantees covering data center construction or leasing. That distinction matters. Unlike the reported $100 billion commitment to OpenAI — which Jensen Huang himself later described as "likely not in the cards," with $30 billion already delivered and no guarantee of more — the Lancium investment has defined terms, defined thresholds, and no off-balance-sheet construction risk.

It also carries a potential upside beyond strategic positioning. Lancium is exploring an IPO in 2027. At a $10 billion enterprise value today, a public listing following the trajectory of peers like Switch and CyrusOne could reprice those assets significantly if the company hits its grid interconnection milestones. Nvidia's $2 billion entry at 20% would then carry both a strategic moat and a financial return.

Nvidia's own financial profile makes this level of investment possible without strain. The company generated $119.1 billion in free cash flow over the trailing twelve months, with $72.1 billion in net cash on the balance sheet after debt. Revenue growth of 70.7% year-over-year and an operating margin of 64% give it a cash engine that most capital allocators only dream of. The $3 billion Lancium commitment represents roughly 2.5% of Nvidia's trailing free cash flow. That is not a stretch.

However, the scale changes the risk calculus

Demand is not the issue. The issue is whether the speed and breadth of Nvidia's investment commitments introduce new risks that a $5.42 trillion company should account for.

Three concerns stand out.

First, circularity. Nvidia finances companies that buy Nvidia products, then those purchases show up as Nvidia revenue. The company's own SEC filings acknowledge that "one AI research and deployment company" contributes a meaningful amount of revenue by purchasing cloud services from Nvidia's financed customers — but the amount is unquantified. Three direct anonymous customers represented 21%, 17%, and 16% of revenue in the first quarter of fiscal 2027. When Nvidia is also the primary investor in several of those customers' infrastructure, the line between organic demand and balance-sheet-supported demand blurs.

Critics, including analyst Jordan Klein at Bernstein, have compared the strategy to the vendor financing that inflated the dot-com bubble — pre-funding the purchase of your own products and calling it growth. The concern isn't that Nvidia is fraudulently inflating revenue. It's that when the AI capex cycle eventually normalizes, a portion of Nvidia's demand base will be companies whose survival depends on Nvidia continuing to invest. That dependency cuts both ways.

Second, concentration of capital at the infrastructure layer. Nvidia has $27 billion in contingent commitments outstanding, plus the $100 billion OpenAI letter of intent (even if only $30 billion has materialized so far). The company's non-marketable securities grew nearly tenfold in one year. If the AI infrastructure buildout slows, these private investments — many in pre-revenue or early-revenue companies — could mark down significantly. A $40 billion investment book that was meant to secure supply chain dominance could become a drag on earnings if the demand curve flattens faster than the deployment timeline.

Third, the power bottleneck itself may not resolve as cleanly as Lancium's pipeline suggests. Texas grid interconnection timelines have historically run years behind initial estimates, even for projects with completed line studies. The Abbott administration's suspension of new applications adds regulatory uncertainty. Lancium's 15 gigawatts of pending projects are in a favorable position relative to competitors, but "favorable position" is not the same as "guaranteed timeline." Nvidia's additional $1 billion is explicitly contingent on grid interconnection milestones — which means that money is not guaranteed and may not be deployed if those milestones slip.

Where capital goes from here

Nvidia's stock trades at $224, with a market cap of $5.42 trillion. Revenue growth of 70.7% and free cash flow margins of 47% justify the scale of the valuation in absolute terms. The company is reporting earnings on August 26, and consensus expects $179.3 billion in quarterly revenue — a number that would continue the growth trajectory that has already been priced in.

I still believe Nvidia will reach multi-trillion-dollar valuations well beyond today's level as the Vera Rubin platform rolls out and the company's software monetization layer matures. But much of that return is likely back-half weighted in 2028 and beyond.

The Lancium deal itself tells a clear story: Nvidia is hedging against its own supply chain risk by owning pieces of the infrastructure that enables its chips to run. That is smart strategy in a market where power has become scarcer than silicon. But the pattern of $40 billion in annual investments, layered on top of contingent commitments that could reach $27 billion more, changes the risk profile of a position that already carries concentration risk.

The debate is not whether Nvidia remains the dominant force in AI compute. It is whether a company that has become the bank, the builder, and the vendor for the same industry has taken on enough off-balance-sheet risk to warrant active management of the position rather than a set-it-and-forget-it approach.

For investors with a multi-year horizon, the long-term thesis remains intact. Nvidia is on the right side of the transition from chip supply to infrastructure control. But for those holding large allocations today, the question that matters is opportunity cost: with the stock up 20% year-to-date and the next leg of returns likely back-half weighted, is capital better deployed here — or elsewhere in the AI stack where the market hasn't yet fully priced in the upside?

What would change my view on this position? If Nvidia's next two earnings reports show revenue growth decelerating below 50% year-over-year while investment commitments remain in the $40 billion annual range, the circular financing risk becomes harder to ignore. If the Lancium grid interconnection milestones slip beyond 2027, the power-first-mover thesis loses its time advantage. And if OpenAI's deployment timeline for the 10-gigawatt Vera Rubin project extends past the second half of 2026 target, the entire Stargate ecosystem — and by extension Lancium's near-term revenue case — gets pushed further into the back half.

Until those signals appear, the Lancium investment is a smart move. But smart moves at this scale require active oversight, not passive conviction.