Google's Power Play

The Energy Carve-Out

The story of the AI build-out is simple: you need more chips, which need more data centers, which need more power. An almost comical amount of power. The obvious solution, if you have infinite money, is to just go out and buy a power company.

And so, last week, Google's parent company Alphabet did just that, agreeing to buy energy developer Intersect Power for $4.75 billion. The tech giant, facing an insatiable demand for energy to fuel its AI ambitions, is vertically integrating. It is getting into the utility business. Or so it seemed.

To understand this deal, you have to look at what Google left on the table. The real story is buried in the fine print, a detail that reveals this isn't a straightforward infrastructure purchase so much as a piece of careful financial structuring. Here's Alphabet:

Intersect's existing operating assets in Texas, and its operating and in-development assets in California, will not be part of the acquisition. Those assets continue to operate as an independent company, supported by existing investors...

This is an important detail. It's like buying a real estate developer for their blueprints and their team of star architects, while telling them to keep the finished, rent-paying skyscrapers they already manage. Google looked at a thriving energy company and performed a surgical extraction, taking the future and leaving the present behind.

The logic here reveals a clear division of risk. Actually owning and operating gigawatt-scale power plants is a messy, capital-intensive nightmare. You have to deal with grid volatility, regulatory headaches, and the slow, grinding work of maintenance. That is a low-margin, high-risk business that Google has absolutely no interest in running.

This impulse—to get the upside of AI infrastructure without the balance-sheet hangover—is becoming the defining financial maneuver of the AI boom. Look at Meta. To fund its massive Hyperion data center in Louisiana, it didn't just borrow money. It formed a joint venture with private credit firm Blue Owl Capital. Blue Owl owns 80% of the facility, Meta owns 20%. Meta contributes its "land and construction-in-progress assets"—essentially, the half-finished site—while Blue Owl brings in billions in cash. And for its trouble, Meta walks away with a one-time cash distribution of $3 billion. They got paid to let someone else fund their project.

Meta then becomes the anchor tenant, signing a lease for the facility. But there's a sweetener in the fine print. To make the deal work, Meta provides a "residual value guarantee." This means if Meta decides not to renew its lease down the line and Blue Owl can't sell the data center for a certain price, Meta has to write a check to cover the difference. It's a clever way to offload the initial capital burden while still retaining some of the long-term risk to make the numbers work for its partners.

Google's deal is a different flavor of the same strategy. Instead of building a new financial vehicle from scratch, it just bought the part of the company it wanted and let the private equity guys keep the parts it didn't. They acquired the two things that are impossible to replicate: a proven development team and a pipeline of future projects. They are buying the intellectual property, the permits, and the human expertise to build the next generation of data centers, custom-built for Google's needs.

And who is taking on those boring, operational assets? The press release helpfully tells us: TPG Rise Climate, Climate Adaptive Infrastructure, and Greenbelt Capital Partners. These are the specialists you call when you want to own and operate stable, cash-flowing infrastructure. The carve-out was by design, completing the other half of the trade.

But avoiding capital risk is only half the story. The deal is also a talent acquisition of staggering proportions. The real bottleneck in building AI infrastructure isn't just power; it's the handful of people on earth who know how to build these projects on impossible timelines. By keeping Intersect's team intact as a subsidiary, Google did more than just buy a project pipeline. It effectively paid $4.75 billion to hire a pre-packaged, world-class energy skunkworks and pointed it exclusively at solving its number one problem.

The market thinks Google bought a power company. What it really bought was a growth option, leaving the stable, cash-generating assets for the private equity firms that specialize in them.

More on AI Infrastructure Financing:

• Hyperscalers: Hypergrowth, higher risk (Barclays Brief)
• Is Your AI Funded by Junk Bonds? (Tomasz Tunguz)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

The Foundry Flop

• The Headline: Nvidia has halted testing on Intel’s crucial "18A" manufacturing process, sending Intel shares tumbling and casting doubt on the chipmaker's ability to act as a domestic alternative to TSMC. (Bloomberg)
• ARPU's Take: This is the "canary in the fab" moment Intel feared. The recent $5 billion capital injection from Nvidia was a financial lifeline, but this halted test is a technical indictment. It reveals that despite the patriotic narrative and "gate-all-around" hype, Intel’s foundry yields are likely still too low or inconsistent to satisfy the industry's most demanding customer.
• The Operations Implication: The "China Hedge" just collapsed. Global supply chain leaders were banking on Intel 18A to provide geographic diversity away from Taiwan (TSMC). If Nvidia—who has every political incentive to make US manufacturing work—is walking away from the test, it signals that the performance gap is still too wide to bridge. For the broader market, this means AI hardware supply chains remain dangerously tethered to the Taiwan Strait, ensuring that GPU availability remains volatile and pricing power stays firmly with TSMC.

The Groq Loophole

• The Headline: Nvidia is executing a "reverse acqui-hire" of AI chip rival Groq, licensing its technology and poaching CEO Jonathan Ross to capture key inference capabilities without a formal acquisition. (Reuters)
• ARPU's Take: This is a ruthless application of the Microsoft/Inflection playbook. Nvidia is effectively creating a "zombie company" out of Groq—leaving the corporate shell alive to appease antitrust watchdogs, while extracting the brain trust and IP that made Groq a legitimate threat to Nvidia's inference dominance.
• The Regulatory Implication: This deal cements the "licensing plus talent" structure as the definitive loophole for Big Tech consolidation in a hostile antitrust environment. By labeling the deal a non-exclusive license, Nvidia maintains the "fiction of competition" required to sidestep the FTC, while functionally removing its most promising architectural rival. It forces regulators to either fundamentally redefine what constitutes a merger or watch as the semiconductor market consolidates via payroll rather than equity.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.