By Krystal Hu

Nov 26 (Reuters) - (Artificial Intelligencer is published every Wednesday. Think your friend or colleague should know about us? Forward this newsletter to them. They can also subscribe here.)

Happy early Thanksgiving!

Beyond the neon glow of the Las Vegas Strip, a different kind of high-stakes game was playing out last week. I was there with a few hundred tech investors at Goldman Sachs' annual Private Innovative Company Conference, where the mood was decidedly split. I kept hearing about the sentiment gap between public software companies and private AI names in conversations with bankers and investors.

Public software firms such as Salesforce CRM.N and Adobe ADBE.O, once the anchors of tech valuations, are looking shaky in investors’ eyes. Many are still trying to figure out how to coexist with tools like AI coding, autonomous workflows, and emerging agentic systems — all of which threaten to break the traditional software model. Meanwhile, in the private markets, the energy feels very different. Multi-billion-dollar AI fundraisings — from Databricks to Anthropic — are still pulling in capital at hundred-billion-dollar valuations, buoyed by rapid enterprise adoption and revenue growth.

The IPO chatter is getting louder, too. More companies are prepping for next year, polishing their AI narratives for S-1 filings — the registration document required to go public, by detailing how AI cuts costs and how it’s helping land customers faster and cheaper. Then you have the hottest intersection between tech and finance right now — chips and data center financing — making its way into Goldman’s deal pipeline.

The financial maneuvering behind this AI gold rush is both what excites some investors and what worries others. This week, we dive into how Google is borrowing from Nvidia’s financial-engineering playbook to claw away a slice of the AI hardware market from the dominant player, and why bond market investors are getting nervous about AI spending. Scroll on.

Email me or follow me on LinkedIn to share any thoughts.

OUR LATEST REPORTING IN TECH & AI:

• As AI reshapes shopping, US retailers try to change how they're seen online

• Dozens of state attorneys general urge US Congress not to block AI laws

• Robinhood CEO's AI startup Harmonic raises $120 million to solve math problems

• Warner Music settles copyright case with startup Suno for licensed AI music

• Trump aims to boost AI innovation, build platform to harness government data

GOOGLE RUNS THE NVIDIA PLAYBOOK

The race for AI hardware today is as much about financial firepower and creativity as it is about technology.

For years, Nvidia NVDA.O has reigned supreme as the supplier of choice for AI compute. But now, Google GOOGL.O is taking a page from Nvidia’s playbook by using its balance sheet and credit rating, alongside its chips, to win customers and secure precious data center capacity.

It began when the tech giants invested in AI startups — OpenAI and xAI by Nvidia; Anthropic and Safe Superintelligence by Google — which, in return, ran their models on the investors’ hardware and cloud services.

Now, both companies increasingly resemble lenders. To get more customers to use its proprietary Tensor Processing Unit chips, Google has realized it needs more TPU-ready data centers. That means financing them, creating a circular loop where chipmakers bankroll the infrastructure that boosts demand for their chips, and take on the risks if that loop ever breaks.

Nvidia set the tone when it began backstopping leases for its GPUs—the powerful chips that serve as the workhorses of AI— helping customers finance huge compute clusters. Google answered with something bolder: a complex deal last month with bitcoin miner TeraWulf WULF.O and cloud provider FluidStack that quickly became the talk of financiers and is now seen as a new playbook for AI infrastructure funding.

Google wasn’t just a customer of the data centers — it effectively acted as both bank and insurer, which made the whole project viable. TeraWulf and FluidStack could never have raised billions to build hundreds of megawatts of AI capacity because few lenders would sign a 20-25-year lease based on the startups’ credit. So Google stepped in with a lease backstop of up to $1.8 billion, swapping out FluidStack’s risk for Google’s AA+ rating.

That guarantee unlocked cheap debt financing, and Google received warrants convertible into roughly 8% of TeraWulf — a fee for 「renting」 its balance sheet. By acting as guarantor, Google gets the megawatts secured for TPU deployments in return.

These deals allow Google to place TPUs directly in client-operated facilities. If a startup fails, Google can assume the lease. It’s a high-stakes version of vendor financing — the kind of credit engineering once reserved for aircraft manufacturing.

As TPUs are finally getting traction outside Google, investors are starting to recognize the value of having a real alternative to Nvidia in the market.

Of course, Nvidia isn’t watching quietly. In a statement responding to The Information’s reportthat Meta was exploring a multibillion-dollar TPU deal, Nvidia said: 「We’re delighted by Google’s success — they’ve made great advances in AI, and we continue to supply to Google. Nvidia is a generation ahead of the industry.」 The chip giant also used its latest earnings call to highlight Anthropic’s GPU usage following a $5 billion investment, which is notable because Anthropic is one of the largest TPU users outside Google.

Both are doubling down not just on technology, but on financial engineering to keep AI’s most valuable customers close. The GPU vs TPU battle is not just about whose chips run faster — it’s about who can finance the future of intelligence.

CHART OF THE WEEK:

Wall Street is struggling to digest a wave of new debt from the companies building the world’s AI infrastructure. The chart shows hyperscalers Amazon, Alphabet, Meta and Oracle have sold over $75 billion in investment-grade bonds since early September, according to Bank of America Global Research. That’s more than what they issued over the previous three years combined.

While stock market investors are already nervous about the sky-high valuations of AI names, bond investors, who are usually the steady, risk-averse crowd, are also starting to reset their expectations. U.S. investment-grade credit spreads climbed to their highest point since June, meaning investors are asking for a bigger premium over Treasuries than they did just a few months ago. It’s a subtle but clear indication that the market is feeling the pressure of the new debt issued by the AI hyperscalers.

(Reporting by Krystal Hu; Editing by Lisa Shumaker)

((krystal.hu@thomsonreuters.com, +1 917-691-1815))