The $35 billion number landed Monday morning. By itself, it is large. Paired with what Anthropic signed three weeks ago, it is something else entirely.

Bullet Summary

• Anthropic reportedly agreed to a $35 billion computing deal with Lambda, a cloud provider backed by Nvidia, to expand its AI capacity.
• Earlier in August, Anthropic reportedly signed a $45 billion agreement with Nscale, another AI infrastructure provider, for computing capacity in West Virginia. That brings Anthropic’s neocloud commitments to roughly $80 billion inside one month.
• Under the reported Lambda arrangement, Hut 8 is developing the Texas data center site, Lambda deploys Nvidia systems and provides cloud services to Anthropic, and Nvidia is reported to hold the site lease.
• Lambda has raised multiple billions in financing recently, including about $1 billion in short-dated debt arranged to fund GPU purchases that will be leased to Microsoft, according to Bloomberg.
• Major Wall Street institutions have become active lenders in AI and compute-linked credit, and Goldman Sachs Research has estimated nearly $500 billion of AI-related debt issuance globally so far in 2026.
• The trend is drawing scrutiny. The Bank for International Settlements has warned that AI-related financing is becoming more concentrated and more circular within the ecosystem, and it has highlighted the risk of a sudden pullback in financing if returns disappoint.
• Spot rental prices for Nvidia H100 GPUs have been widely tracked falling into the $2 to $4 per hour range by 2025 as newer architectures arrived. That depreciation curve is the collateral risk embedded in every loan backing these deals.

The Structure Is the Story

Most coverage is treating this as a simple cloud capacity agreement. It is not. The unusual structure highlights Nvidia’s expanding role beyond selling AI chips. The company has invested in Lambda and is increasingly helping cloud providers obtain the financing and infrastructure needed to deploy Nvidia-powered systems. Nvidia is reported to hold the underlying lease on the Texas facility. Lambda deploys the chips. Anthropic pays Lambda. Lambda services debt backed by those same chips.

Mizuho chip analyst Jordan Klein has described the broader financing model as concerning: using Nvidia GPUs as collateral, renting them back into the AI supply chain, and relying on customer lease payments to repay loans can look like a form of vendor-supported demand. That is the circularity active traders need to price.

What the Hut 8 Layer Adds

Hut 8 has disclosed two 15-year leases at its Beacon Point campus in Nueces County, Texas. The base value of the two leases combined sits at $19.6 billion, and if renewal options are exercised, that number climbs to $50.2 billion. The tenant has not been publicly named by Hut 8, though Reuters has reported that the Financial Times identified Nvidia as the customer behind the leases and that Reuters could not independently verify the report. Lambda now plugs directly into that same facility, adding Anthropic’s reported $35 billion demand signal on top of an infrastructure layer Nvidia is reported to be anchoring.

Lambda is reportedly in talks for up to a $3 billion pre-IPO round at around a $12 billion valuation. The Anthropic deal changes that valuation math materially, but it also concentrates Lambda’s counterparty risk in a single customer relationship worth multiples of the company’s last widely reported equity mark.

Trading Framework

Three variables define the risk spectrum here. First, GPU depreciation: Lambda’s debt financing structure more closely resembles traditional infrastructure project finance, with debt secured by GPU systems and contracted cash flows from those assets. If Blackwell-generation pricing follows the H100 depreciation pattern, the collateral deteriorates faster than the loan matures. Second, Anthropic’s credit: the company is not publicly rated, and neither deal discloses Anthropic’s payment obligations. Third, Nvidia’s concentration: the chip giant simultaneously holds the lease, backs the lender, and supplies the hardware. A single supply disruption travels all three channels at once.

For traders, NVDA’s +1.48% response Monday is a sentiment read, not a valuation one. The real signal sits in HUT, which carries direct lease exposure, and in the neocloud credit complex broadly. Watch Lambda’s pre-IPO terms if they surface. The spread between that equity mark and the debt stack it is carrying will tell you what the market actually believes about AI demand durability.

Preparation over prediction. The structure is novel, the collateral is depreciating hardware, and the counterparty chain runs through a single chip supplier. Know those variables before the position is sized.