Google's AI infrastructure push for Anthropic is not just a story about chips. It is also a story about balance sheets, guarantees, leases, data centers and power.
According to the source article, Google has worked with Broadcom, Apollo, Blackstone, Morgan Stanley and several crypto mining companies on one of the largest infrastructure financing programs in history, as described by the Financial Times. The goal is to give Anthropic access to Google's Tensor Processing Units, or TPUs, while avoiding a straightforward hardware purchase that would sit heavily on corporate books.
Why the TPU deal needed a financing workaround
Anthropic needs large amounts of AI hardware, but the source says it cannot buy the chips itself because it has no credit rating. Banks would not lend the startup that level of money, and the other companies involved also have reasons to avoid directly carrying the hardware on their balance sheets.
Google is already spending record sums and does not want more pressure on its own balance sheet. Broadcom, which resells the Google chips, also does not want to tie up capital in the hardware. That leaves the parties with a practical problem: Anthropic needs compute, Google wants to sell TPUs, and no one wants the full hardware burden to appear plainly on their books.
The answer is a special-purpose financing vehicle. Morgan Stanley helped arrange a structure that buys the chips and leases them to Anthropic. Outside investors, mainly Apollo and Blackstone, provide the money.
The first use of the structure comes in June, when a special purpose vehicle called Compute SPV buys about one gigawatt of TPU hardware for $35 billion. The Financial Times says that equals roughly one million TPUs. Broadcom provides a backstop covering about $30 billion of the purchase if Anthropic stops making lease payments.
How Google's TPU strategy reaches beyond its own data centers
The chips at the center of the arrangement are Google's Tensor Processing Units. Google has developed TPUs with Broadcom since 2016. They were originally built for Google's own data centers, but they are now sold to outside customers and compete with Nvidia's position in the AI processor market.
Google sells the chips in Pods, which are server racks that connect thousands of TPUs into one computing system. For an AI company like Anthropic, that kind of infrastructure is central to scaling model work. For Google, selling TPUs outside its own walls gives the company another way to shape the AI hardware market.
The financing structure is already being used as a template for more deals. The largest deal to date is an April agreement covering Google's sale of another 3.5 gigawatts of TPU hardware to Broadcom for Anthropic. Broadcom's financial filings list $128 billion in purchase commitments through 2028, and Financial Times sources say nearly all of that amount is tied to Google TPUs.
That means the arrangement is not a one-off workaround. It has become a repeatable model for moving large volumes of AI hardware through a chain of chip supply, financing, guarantees and leases.
Power and data centers are the other half of the problem
Buying chips does not solve the whole infrastructure challenge. The TPUs also need data centers with enough electricity to run them. The source says Google is turning to crypto miners because those companies have already secured access to large amounts of power.
TeraWulf is the first crypto miner to receive a Google guarantee. That guarantee covers a 360 megawatt data center in New York. Morgan Stanley packages the guarantee into a $3.2 billion construction bond, and Google receives an ownership stake in TeraWulf in return.
The same model has expanded to other crypto miners, including Cipher Digital and Hut 8. According to Financial Times sources cited in the article, Google has backed ten projects with a combined capacity of 2.4 gigawatts so far.
This part of the structure matters because AI infrastructure is constrained by more than silicon. Chips, financing, buildings and electricity all have to line up. In this arrangement, crypto miners provide a route to power and data center capacity, while Google's guarantees help move construction financing forward.
The balance-sheet exposure has not disappeared
The core tension is that risk has been moved around, not eliminated. The Financial Times reports that Google could face up to $44 billion in obligations if every lease defaults. Yet Google records only $815 million of that liability on its balance sheet, leaving most of the exposure off its books.
That gap explains why the structure is drawing attention. The companies involved can lower visible balance-sheet pressure while still depending on Anthropic to keep paying for the leased infrastructure. If those payments continue, the model supports a massive buildout. If they stop, guarantees and obligations become much more important.
The financing model also appears to affect borrowing costs. The source says Google-backed data center projects borrow at a median interest rate of 7.1 percent, compared with 9.3 percent for neocloud operators that rely on Nvidia chips. Jefferies analysts describe that gap as "a structural cost-of-capital disadvantage" for companies in Nvidia's ecosystem.
Why Anthropic's growth is the key variable
The arrangement depends heavily on Anthropic's ability to grow revenue and keep paying for capacity. The source says $200 billion in contracts are tied to that outcome. It also notes that Google is on both sides of the relationship as an investor in Anthropic and as the supplier of its chips.
An earlier report from The Information says Anthropic has committed to spending about $200 billion on Google Cloud over five years in exchange for five gigawatts of server capacity. That deal accounts for more than 40 percent of Google's committed future cloud revenue.
The source also says Anthropic and OpenAI together account for roughly half of the $2 trillion in cloud backlogs at Amazon, Microsoft, Google and Oracle. Both startups are counting on their revenue growing 20 to 30 times by 2029.
That makes the infrastructure boom highly dependent on startup growth assumptions. If Anthropic's revenue keeps expanding as expected, the financing structure could help Google place enormous TPU capacity into the market. If growth slows or stalls, the article says the entire structure could unravel.
For now, Google's Anthropic TPU strategy shows how AI infrastructure is being built through financial engineering as much as technical engineering. The chips matter, but so do the leases, guarantees, investors, power agreements and accounting treatment that make the buildout possible.