Why Lenders Trust Google's Chips More Than Nvidia's

The Headline You Almost Missed

SpaceX had a rough day. The stock was down 13%, sitting at a record low just above $108 a share, roughly 52% below its post-IPO peak, with more than 900 million insider shares set to unlock the very next morning. Short sellers already controlled 34% of the float. It was the kind of chart that makes portfolio managers reach for antacids.

But the SpaceX drama, as vivid as it is, was almost a distraction from a quieter and more consequential story unfolding a few miles away in the same industry.

Two Companies, Two Bets

On the same day SpaceX was melting down, Elon Musk stepped onto his first earnings call and announced that SpaceX would be doubling down on Nvidia as the exclusive AI chip supplier to the company. Nvidia's stock popped 4% on the news. The market loved it.

Meanwhile, Google and Anthropic were finalizing something far more elaborate: a roughly $200 billion financing architecture designed to supply Anthropic exclusively with Google's TPU chips, co-developed with Broadcom over nearly a decade. The structure pulls in Morgan Stanley, Apollo, Blackstone, and even some crypto miners, all connected through a web of contracts that Neeta described as getting "very, very sticky very quickly."

The contrast is almost cinematic. One billionaire publicly pledged loyalty to the dominant chip incumbent. A sprawling consortium quietly built an alternative ecosystem and priced it more cheaply.

The Sale-Leaseback You Did Not See Coming

The most interesting piece of engineering in the Google-Anthropic deal is not technical. It is financial. None of the parties wanted tens of billions of dollars of depreciating AI chips sitting on their balance sheets. The solution, arranged by Morgan Stanley, is a private credit vehicle that allows outside investors to purchase the chips and lease them back to Anthropic.

The model borrows from aircraft financing, the kind GE Capital and Boeing used to move planes off balance sheets for decades. The analogy is imperfect: a 737 depreciates on a different curve than an Nvidia H100 or a Google TPU. But the structural logic holds. You move the asset, retain the revenue relationship, and distribute the risk across investors who are betting on the lessee's survival.

Google, for its part, is marking just $815 million of its potential $44 billion liability on its balance sheet. That is a gap worth watching.

220 Basis Points Is Not a Rounding Error

Here is the number that should stop any infrastructure investor mid-sentence. According to reporting from the Financial Times, data center projects backed by Google's TPU ecosystem are borrowing at a median rate of 7.1%. Comparable operators building around Nvidia chips are paying approximately 9.3%.

That 220-basis-point gap is what a Jefferies analyst called a structural cost of capital disadvantage for entities in Nvidia's orbit. Over a multi-gigawatt buildout, across dozens of facilities, that spread compounds into a meaningful competitive moat. It is not just about chip performance anymore. It is about who can finance infrastructure more cheaply over time.

Google has also committed to backstopping ten data center developments totaling up to 2.4 gigawatts of power capacity. That is not a sideline investment. That is a bet that the TPU ecosystem needs to be the default for serious AI workloads.

Anthropic's Quiet Insurance Policy

There is one more layer to this. Even as Anthropic sits at the center of a $200 billion deal premised on its exclusive use of Google TPUs, Reuters has reported that Anthropic is also quietly working to develop its own in-house chips in partnership with Samsung.

This is not contradictory. It is rational. When a single supplier relationship underpins a $200 billion obligation, you build a backup. Anthropic is the load-bearing wall of this entire structure, and it is not going to let one cracked foundation bring the building down.

What Comes Next

The broader picture here is unsettling in a way that goes beyond any single deal. A Nikkei Asia study cited in a prior Good Revenue episode estimated approximately $1.6 trillion in AI infrastructure liabilities across five major tech companies, with Meta topping the list. The Google-Anthropic scheme is being layered on top of that existing exposure.

The $200 billion structure works if Anthropic keeps growing, keeps paying, and keeps the lights on. If it slows, the circular financing web Neeta described does not gently unwind. It snaps. And with SpaceX betting on Nvidia at a cost premium while Google builds a cheaper alternative ecosystem around a rival, the industry is quietly splitting into two camps with very different financial physics.

Which camp survives a downturn intact is a question worth asking before the next unlock date.