Investors Question Oracle’s Data Center Financing

This is Bloomberg Tech. Coming up: questions over Oracle’s data-center financing, OpenAI in talks to raise ten billion dollars from Amazon and to use its Trainium chips, and Waymo exploring a raise that could top fifteen billion dollars. First, Oracle shares sink after Blue Owl declines to back a ten billion dollar Michigan site; Oracle says the project stays on track with a different equity partner. Is funding these massive builds getting tougher?

Oracle’s pursuing a huge, fast build-out that could remake the company if everything lines up, but the FT’s report on financing snags at one big site rattled investors who wonder if lenders see risks they don’t.

You’ve flagged the surge in leasing. How is this off–balance sheet approach different from classic capital spending?

Instead of owning facilities outright, companies are renting capacity, spreading costs over time and keeping it off the books. Oracle’s future commitments are enormous—on the order of hundreds of billions over about two decades—and Michigan likely fits that bucket, underscoring the scale.

Oracle’s newer to this game than the hyperscalers. Does that inexperience factor into the market’s nerves?

Yes. Microsoft can carry similar obligations with less fuss because of its cash flow. For Oracle, the spend is large relative to the business, so flawless execution matters more.

OpenAI is in early talks to raise at least ten billion dollars from Amazon and to use Trainium. Is this a circular deal?

It certainly feels that way. OpenAI already committed roughly thirty-eight billion dollars of capacity over seven years with Amazon, mostly on Nvidia. Now Amazon may invest back into OpenAI.

And if OpenAI trains on Trainium, that’s a fresh stamp of approval for Amazon’s silicon, right?

Exactly. Few companies train gigantic models, so getting GPT to run well on Amazon’s chips would be a big win for AWS’s in-house hardware.

What does that mean for Nvidia?

It highlights growing demand for alternatives. Nvidia is powerful but pricey and constrained, so AMD, Google’s TPUs, and Amazon’s accelerators are all pushing to grab share.

Let’s bring in T. Rowe Price’s Tony Wang. Are you worried about the financing loop between cloud providers and model makers?

Creative structures always draw scrutiny, but OpenAI needs more compute and multi-sourcing makes sense. The bigger point is that AI adoption and usage keep climbing, and these tools are reshaping work.

You own Alphabet and Nvidia in size. Are you concerned Nvidia cedes ground to house-made chips?

The pie is expanding. It’s not winner-take-all. TPUs are tuned to Google’s stack, while Nvidia offers a broad merchant platform, and companies like AMD and Broadcom are also well positioned.

Back to Oracle’s bet. Do you have concerns about its leasing push and sheer capacity build?

Oracle sees strong demand signals. Leasing now and owning later is a reasonable path. If three to five year demand materializes, the strategy can work despite today’s volatility.

Does the anxiety create a buying window?

We rotate toward the best risk-reward, and AI remains a powerful multi-year theme. Back businesses with durable advantages, knowing not every rocket reaches orbit.

Most gains so far have come from infrastructure. Do you shift toward applications next?

Profit pools are moving up the stack. Foundation model players and incumbents that layer in AI could capture more value as costs fall and models get more domain-specific.

Memory has been a standout. Micron reports after the bell. Are higher prices sticking?

HBM demand soaks up capacity, tightening DRAM and even NAND, which lifts pricing and spreads benefits across component suppliers.

You also invest privately. What about Lambda?

We invest across public and private. We partnered with Lambda on an AI-focused cloud and data-center approach. They’ll go public when the time is right.

Warner Bros. Discovery urges shareholders to reject Paramount–Skydance’s offer and stick with its Netflix deal. What’s the argument?

They say the thirty-dollar tender is less certain than their signed pact with Netflix. The Paramount group lined up debt but still needs a large equity check. Warner Bros. wanted a firmer personal backstop from the Ellisons, not a revocable trust. With Affinity pulling out and other equity unclear, they view financing risk as elevated, while regulatory risk looks similar for both paths.

Google rolls out a more efficient Gemini model. Tesla faces a potential thirty-day sales ban in California tied to Autopilot marketing. And Waymo is talking about raising more than fifteen billion dollars at a valuation that could top one hundred billion. Sarah, why the big raise now?

Waymo has a real service on city streets and needs capital to scale into more markets, road types, and even overseas. Revenue is reportedly north of three hundred fifty million on an annualized basis. It sits within Alphabet’s Other Bets, where Ruth Porat has pushed outside capital and more independence to foster financial discipline.

Markets are uneasy about AI spending and Oracle is down again. Mandeep, what’s behind the skepticism on Oracle’s data-center push?

Backers are being asked to fund capacity that ramps in 2028 and beyond, with Oracle heavily tied to OpenAI. If model leadership is now more distributed across Gemini, Anthropic, XAI and others, it argues for customer diversification. Training-heavy exposure raises ROI questions, which is why Microsoft passed on certain OpenAI training loads that Oracle took.

How are investors weighing off–balance sheet financing and the path forward?

Many big tech firms use off–balance sheet structures to avoid adding debt. Oracle might need to lean more on equity or moderate ambitions if funding costs rise. Near-term capacity looks tight, but the three to five year payoff depends on who leads in models and whether the next training runs deliver enough return.

Micron is up big year to date heading into earnings. Kim, what’s the setup?

Micron is cyclical. When the cycle runs hot, it can be very strong, and the stock often rallies into results and cools afterward.

What will you watch after the bell?

HBM DRAM is the must-have product and supplies are limited. I’ll watch how quickly they pivot from older, consumer NAND toward DRAM and why management is confident that demand stays elevated.

Do newer cloud players give you pause?

AI will succeed, but pure brute-force scaling may give way to smarter designs that temper hardware growth over time.

And China exposure?

Export limits matter, but for the next eighteen to thirty-six months, scaling likely drives strong data-center demand.

Can Nvidia, TPUs, and Trainium all win?

Different phases of model development can favor different hardware. Competition benefits builders, and it’s notable how quickly Google and Amazon have produced usable alternatives.

Relativity Space burned cash chasing fully 3D-printed rockets and now turns to Eric Schmidt and fresh funding. Lauren, is the strategy changing?

Yes. After prolonged struggles with 3D printing, the company has shifted toward conventional manufacturing, with printing now playing a minimal role.

Why would Schmidt take the helm and invest so heavily?

He’s hinted at building data centers in space, an idea others have floated. He’s injected substantial capital already and is courting more high-net-worth backers to grow the business.

That’s it for Bloomberg Tech. Our podcast is on the Terminal, Apple, Spotify, and iHeart. We’ll continue to track Oracle’s data-center financing across the show. From New York, this is Bloomberg Tech.