Signals Inbox·August 22, 2026·Data Centers

Why is Nvidia investing hundreds of millions in Cloverleaf?

Nvidia is backing Cloverleaf because powered land is becoming part of the AI supply chain: the company wants earlier access to the sites, electricity and facility decisions that determine how many AI systems can actually be deployed.

Q1Did Nvidia really put hundreds of millions into Cloverleaf?

Yes: Nvidia has confirmed a minority investment in Cloverleaf Infrastructure, while The Wall Street Journal reports that the check is several hundred million dollars.

The exact amount is still private. Reuters confirmed that the companies did not disclose financial terms, so “hundreds of millions” should be treated as well-sourced reporting rather than an official figure from Nvidia.

The size is striking for a company founded only in 2024. Cloverleaf started with a $300 million commitment from Sandbrook Capital and NGP, meaning Nvidia may now be investing an amount in the same broad range as the capital that originally launched the business.

There is another useful detail in the timing. The Wall Street Journal reported that Cloverleaf had recently hired JPMorgan to explore strategic options, including a possible sale of the company. Nvidia ultimately took a minority stake instead. That gives us the first clue about what Nvidia wants here: access and alignment with Cloverleaf, without having to own the whole company.

Q2What does Cloverleaf actually do?

Cloverleaf turns ordinary land into sites where someone can realistically build a huge data center.

That sounds simple until we look at what has to happen before a 500 MW or 1 GW AI campus can exist. Cloverleaf finds land, studies nearby transmission, works with utilities on electricity supply, helps arrange grid upgrades, navigates local approvals and thinks through generation, substations, water and other infrastructure.

Its own description is “shovel-ready” powered sites. The Wall Street Journal gives an even clearer picture of the business: Cloverleaf signs power agreements with utilities and develops land around that power so data center companies can eventually take over.

The distinction is useful. Cloverleaf does not generally need to spend $10 billion building every final campus itself. It can do the difficult early work, improve the site's value enormously, then sell or transfer the development to a company with the balance sheet to build the actual data centers.

So when Nvidia invests in Cloverleaf, it is moving several steps upstream from the GPUs. It is backing the company trying to create the places where those GPUs can eventually go.

Q3Why does Nvidia care about powered land so much now?

Powered land has become one of the clearest physical limits on AI growth today, and Nvidia cannot sell unlimited AI systems if customers have nowhere to run them.

The latest Berkeley Lab update makes the scale unusually concrete. Its researchers now estimate that U.S. data centers could consume 11.8% of all American electricity by 2030, with scenarios ranging from 9.5% to 15.3%. The previous national forecast was already aggressive, yet the newer estimate moved higher again as planned AI hardware shipments increased.

At the individual-site level, the industry has also jumped into a different class of project. CBRE says the old 12-to-18-month development cycle for sub-50 MW facilities no longer describes the market. AI campuses are increasingly 500 MW or larger, and projects needing new transmission or generation can take 24, 36 or more than 48 months to connect.

CBRE now says the ability to secure more than 300 MW within 36 months can matter more in site selection than connectivity, which used to dominate the discussion. That is a remarkable change for the data center industry.

Nvidia's own language has moved in the same direction. The company now talks explicitly about “land, power and shell” as foundations of an AI factory. Its new DSX platform even includes the physical facility layer, covering land, electricity, cooling and building design alongside compute and networking.

Clean power is useful to Cloverleaf because large customers care about emissions and long-term energy economics, but we would not reduce Nvidia's investment to a green-energy bet. The immediate prize is dependable power at very large scale.

Q4Is Cloverleaf part of a much bigger Nvidia infrastructure push?

Yes, and lately Nvidia has been moving into AI infrastructure with a speed that makes the Cloverleaf investment look deliberate rather than opportunistic.

Nvidia recently invested $1.5 billion in SB Energy around a huge Ohio AI campus. It has invested $2 billion in Lancium and can add another $1 billion. It also struck a deal with IREN covering a possible five-gigawatt DSX-aligned infrastructure buildout, including a right to invest up to $2.1 billion.

Then there is the financing layer. Nvidia has partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on platforms designed to mobilize more than $500 billion of third-party capital for AI infrastructure.

We are watching Nvidia attack several bottlenecks at once: sites, power, buildings, financing and the design of the AI factory itself.

Recent Nvidia infrastructure moves

Q5How big is Cloverleaf's pipeline, really?

Cloverleaf's pipeline is enormous for a two-year-old developer, although the 10+ GW headline should never be confused with 10+ GW already built.

According to The Wall Street Journal, Cloverleaf has sold projects representing more than 7 GW of powered land and currently has more than 10 GW in its development pipeline.

For scale, OpenAI originally framed Stargate around a 10 GW U.S. infrastructure target backed by as much as $500 billion. Cloverleaf's reported pipeline is therefore in the same power range as the original headline target for one of the largest AI infrastructure programs ever announced.

That comparison does not mean Cloverleaf owns a second Stargate. A project sitting in Cloverleaf's pipeline can still fail to secure approvals, lose local support, run into utility constraints or change hands long before construction.

Still, 7+ GW reportedly sold is much harder to dismiss as a PowerPoint pipeline. Cloverleaf has already managed to move several gigawatts far enough through development for customers to take them over.

For Nvidia, that combination is attractive: a very large pool of future sites, plus evidence that some of those sites can actually be monetized.

Q6Has Cloverleaf actually turned a site into a real AI campus?

Yes, and the Port Washington project is the strongest proof we found that Cloverleaf can take a site from early development into the top tier of AI infrastructure.

Cloverleaf assembled land around Port Washington, Wisconsin through a development known as Project Red Granite. It worked on the site, utility relationship and development framework before Vantage Data Centers took control of the project.

What happened next is unusually important. Vantage announced a planned investment of more than $15 billion for a four-building campus delivering close to 1 GW of AI capacity. OpenAI and Oracle then identified the Wisconsin campus as part of Stargate.

So Cloverleaf helped originate a site that ultimately attracted Vantage, Oracle and OpenAI.

That tells us much more than a generic claim about “site development.” We can follow the conversion from farmland and power access to one of the largest AI campuses under development in the United States.

The project also shows where Cloverleaf creates value. By the time OpenAI needs the compute, much of the ugly early work around land, utilities and local development has already been handled by other people.

Nvidia increasingly wants that process to happen faster and more often.

Q7Why can't Microsoft, Oracle or OpenAI just find the power themselves?

They can, and the biggest tech companies already employ large teams to do exactly that, but the amount of new infrastructure being requested now is stretching even the hyperscale model.

Microsoft, Google, Amazon and Meta have spent years building internal expertise in real estate, electricity procurement and utility negotiations. Brian Janous, one of Cloverleaf's founders, actually spent roughly a decade leading Microsoft's cloud energy strategy.

The pressure becomes more obvious when we look beyond the established hyperscalers. Frontier AI companies can suddenly require several gigawatts of infrastructure without having spent 20 years building utility teams, power portfolios and investment-grade relationships across dozens of electricity markets.

OpenAI is the obvious example. It has moved from buying cloud capacity to participating in infrastructure programs measured in hundreds of billions of dollars and multiple gigawatts. Oracle, Vantage, SoftBank, SB Energy and others are doing large parts of the physical development around that demand.

Cloverleaf fits into this increasingly specialized chain. A customer does not have to discover every promising transmission corridor, negotiate every parcel of land and start every utility conversation from zero.

We would expect hyperscalers to keep developing many sites themselves. The interesting change today is that AI demand has become large enough for an independent powered-land specialist to build a 10+ GW pipeline alongside them.

Q8Why back Cloverleaf instead of another big data center owner?

Cloverleaf gives Nvidia exposure earlier in the process, when a relatively small amount of capital can influence a much larger future development.

Consider the difference. A company building and owning a multi-gigawatt campus can eventually require tens of billions of dollars. Cloverleaf works earlier: find the site, assemble the land, secure a credible route to electricity, move through approvals, then hand the opportunity to the companies prepared to finance the full buildout.

Port Washington gives us a useful real-world example. Cloverleaf originated the project, while Vantage is now talking about more than $15 billion of investment in the finished campus.

That is the leverage Nvidia can potentially get from this model.

Several hundred million dollars invested in a company originating many projects can give Nvidia exposure to far more potential capacity than spending the same money directly on the concrete, steel and electrical equipment of one finished data center.

Nvidia is already willing to make much larger commitments downstream, as the SB Energy and Lancium deals show. Cloverleaf gives it another entry point near the beginning of the funnel.

Q9What does Cloverleaf know that Nvidia wants?

Cloverleaf's real expertise is getting enormous electricity demand from “interesting location” to “credible project,” which has become a valuable skill these days.

Its leadership explains a lot. Brian Janous previously ran Microsoft's energy portfolio. David Berry co-founded Clean Line Energy and ConnectGen, businesses built around large energy and transmission projects. Jonathan Abebe spent years working on grid issues, including as a senior technical adviser at the U.S. Department of Energy. Chief Development Officer Aaron Bilyeu has worked on land acquisition and site development at Meta and Microsoft.

That mix is unusually well matched to the current AI bottleneck.

Nvidia knows chips, systems, networking and increasingly facility design. Cloverleaf's team has spent careers dealing with utilities, transmission, energy projects, land and local development.

Those worlds do not naturally move at the same speed. Nvidia can launch a new compute platform on a regular product cycle. A transmission upgrade or new generation project can take years.

Cloverleaf's job is to narrow that gap.

Q10Does Nvidia now get first dibs on Cloverleaf's sites?

There is currently no public evidence that Nvidia has exclusive rights to Cloverleaf's whole pipeline.

That is worth keeping clear because Nvidia has made stronger arrangements elsewhere. In the SB Energy project, Nvidia's role comes with explicit commitments around Nvidia compute. We have not seen equivalent language saying every future Cloverleaf project must use Nvidia hardware.

The actual Cloverleaf announcement says Nvidia became a minority investor and that the companies will work together on AI factory infrastructure. It also says Cloverleaf customers will be able to engage with Nvidia across accelerated compute, networking, software and DSX.

That still gives Nvidia an interesting position.

If Cloverleaf is discussing power, cooling and site design with a customer years before the first server arrives, Nvidia can now enter that conversation much earlier. It can help shape the facility around the requirements of its computing systems instead of appearing after the electrical and cooling architecture has already been fixed.

So we would not assume contractual control over Cloverleaf's pipeline. What Nvidia has clearly bought is proximity to the decisions that eventually determine which AI systems a site can support.

Q11Can one good Cloverleaf site really justify a check this big?

Yes, because one multi-gigawatt AI campus can generate Nvidia hardware sales that dwarf a several-hundred-million-dollar investment.

We can see the economics in Nvidia's separate Ohio deal with SB Energy. Nvidia has estimated that one hardware generation at the first 4.25 GW of that campus could require around 1.5 million Nvidia GPUs and generate roughly $150 billion to $200 billion of Nvidia revenue.

We should not copy those numbers directly onto a Cloverleaf project. Power density, customer mix, accelerator choice and timing will vary enormously from site to site.

But the order of magnitude answers the question.

Nvidia's latest reported quarter produced $75.2 billion of Data Center revenue, up 92% from a year earlier. At that current revenue rate, several hundred million dollars represents less than one normal day of Data Center sales.

A powered site that eventually hosts hundreds of thousands of Nvidia GPUs can therefore be strategically valuable even if Nvidia's equity return on Cloverleaf itself turns out to be merely decent.

That changes how we should judge the check. Nvidia does not need every Cloverleaf project to become another Port Washington for the investment to make sense.

Q12Is Nvidia helping finance demand for its own GPUs?

Partly, yes, and the circularity concern around Nvidia's infrastructure strategy is becoming harder to ignore.

Nvidia now invests in AI companies, compute providers and infrastructure developers while also helping financing platforms channel outside capital into AI factories. Some of those companies then spend heavily on Nvidia hardware.

The Ohio arrangement pushes this particularly far. Nvidia is providing major financial support around a campus where it expects enormous future hardware sales. Critics can reasonably ask how much of today's AI demand is being accelerated by Nvidia helping the ecosystem finance the infrastructure that buys Nvidia products.

Cloverleaf is a milder version of that loop.

Cloverleaf itself is mainly developing powered sites. Nvidia's money does not simply go into an AI startup and come straight back as a GPU order. It helps create physical capacity that could ultimately be used by OpenAI, Oracle, hyperscalers or other customers.

Still, Nvidia is increasingly doing more than waiting for demand to appear. It is actively removing the financial and physical obstacles that could prevent that demand from turning into shipments.

That strategy can be extremely powerful while AI usage keeps expanding. It also concentrates Nvidia's exposure if the industry eventually builds far more capacity than customers can economically use.

Q13Why does Nvidia want Cloverleaf using DSX so early?

Putting Nvidia DSX into Cloverleaf's planning process lets Nvidia influence the data center before crucial power and cooling decisions become expensive to change.

This part of the announcement deserves more attention than it received.

Cloverleaf says it will use Nvidia DSX to bring together decisions around site, power, cooling, compute and the facility itself earlier in development. Nvidia's latest DSX facility design covers everything from substations and battery systems to cooling loops, data halls and network infrastructure.

The technical reason is simple: the building and the computer are increasingly becoming one system.

Nvidia's current reference designs include racks drawing hundreds of kilowatts, liquid cooling, detailed power architectures and software that can alter compute behavior in response to grid conditions. Nvidia also says its DSX MaxLPS approach can run up to 40% more GPUs at their most energy-efficient point within the same power envelope.

A developer that locks in the wrong electrical or cooling design early can therefore waste scarce megawatts later.

Cloverleaf gives Nvidia a chance to push its architecture into projects while they are still being shaped. If that works, Nvidia gains something more durable than a chip sale: future campuses arrive already designed around the way Nvidia wants AI factories to operate.

Q14What could stop Cloverleaf's 10+ GW pipeline from becoming real?

Local opposition is currently one of the clearest threats to Cloverleaf's pipeline, and the latest events show that this is already killing or delaying projects.

Cloverleaf backed away from Greenleaf, Wisconsin after community resistance. It also stopped pursuing a site near the former Kewaunee nuclear plant after local leaders opposed the idea.

More recently, Wrightstown gives us an even stronger test. Cloverleaf had explored a roughly 1 GW project around the Wisconsin village. Local officials imposed a one-year moratorium, and when residents were asked whether the village should support public infrastructure for large data center developments, roughly 87% voted no.

That is 1,033 “no” votes out of 1,184 ballots cast.

The referendum was advisory, so it does not legally erase every possible future project. Politically, though, an 87% rejection is difficult for any developer to ignore.

As seen above, Cloverleaf has already shown it can turn a successful site into something enormous. The opposite pattern is now visible too: a technically attractive location can become worthless if the community refuses the development.

Main risks to Cloverleaf's pipeline

Q15Did Nvidia need to buy equity, or would a normal partnership have been enough?

Owning part of Cloverleaf gives Nvidia stronger alignment and earlier visibility than a simple technical partnership would provide.

We do not know whether Nvidia received special governance, information or commercial rights because the detailed investment terms remain private. So there is a limit to what we can claim here.

Still, equity changes the incentives.

If Nvidia simply supplied DSX technology, Cloverleaf could use it on individual projects while both sides remained ordinary commercial partners. A minority stake lets Nvidia participate financially if Cloverleaf becomes more valuable, while Cloverleaf gets a strategic investor whose products sit at the center of the AI buildout driving demand for its sites.

The timing makes the choice more interesting. Cloverleaf had reportedly been considering strategic options including a sale. Nvidia had the opportunity to remain outside the company, and apparently chose ownership instead.

The likelier reading is that Nvidia wants Cloverleaf independent enough to keep originating sites for many customers, but close enough that Nvidia can help shape how those sites turn into AI factories.

Q16Is Nvidia overpaying for Cloverleaf?

We cannot responsibly call the Cloverleaf investment cheap or expensive today because neither the exact check nor the valuation has been disclosed.

Anything more precise would be invented.

What we can assess is whether several hundred million dollars looks disproportionate relative to the strategic asset Nvidia is buying into. It does not.

Cloverleaf reportedly has more than 10 GW in its pipeline, has already sold over 7 GW of projects and has demonstrated that one originated location can become a $15 billion-plus Stargate campus.

Nvidia, meanwhile, generated $45.5 billion of non-GAAP net income in its latest reported quarter. It recently authorized another $80 billion of share repurchases and is making infrastructure commitments measured in billions elsewhere.

At Nvidia's current scale, the Cloverleaf check is financially small.

The real valuation question therefore sits inside Cloverleaf itself: how much of that pipeline eventually reaches a sale or construction, what profit Cloverleaf captures per developed site, and how repeatable Port Washington proves to be.

Without those private numbers, we can judge the strategic logic much more confidently than the financial price.

Q17So why is Nvidia investing hundreds of millions in Cloverleaf?

Nvidia is investing hundreds of millions in Cloverleaf because the next constraint on AI growth increasingly sits outside the GPU: finding places with enough land, electricity and infrastructure to install millions more GPUs.

Cloverleaf has already shown it can originate a site that becomes a major AI campus. Its reported pipeline exceeds 10 GW. Its team comes directly from the utility, hyperscale-energy and data center development worlds. And Nvidia can now bring DSX into those projects while decisions about power, cooling and facility design are still being made.

At the same time, Nvidia's behavior has changed. The company is currently putting billions into power and data center developers, working with financial institutions to mobilize hundreds of billions more and extending its technical stack from GPUs all the way into facility infrastructure.

The pattern tells us what Nvidia is worried about.

For years, Nvidia had to make sure enough advanced chips, memory and networking equipment existed to satisfy AI demand. Today it is also helping make sure there are enough powered buildings to receive those systems.

Cloverleaf gives Nvidia a relatively capital-efficient way to intervene at the very beginning of that process. Nvidia can help create future AI capacity before anyone spends the tens of billions required to finish the campus.

There are real limits. Cloverleaf's 10+ GW pipeline will suffer attrition, local opposition is already killing projects, and we found no public evidence that Nvidia now controls or exclusively reserves Cloverleaf's sites.

Nvidia is backing Cloverleaf because powered land is becoming part of the AI supply chain, and Nvidia no longer wants to leave such an important bottleneck entirely in other people's hands.