Anthropic strikes $35 billion cloud deal with Nvidia-backed Lambda

Anthropic agreed to a $35B deal to secure additional computing capacity via Nvidia-backed cloud provider Lambda, per the WSJ. The capacity will be supplied from a Texas data center operated by bitcoin miner Hut 8, with Nvidia retaining the underlying lease. The announcement signals large-scale AI infrastructure buildout and could be incrementally positive for AI compute supply chains, though details are limited given it’s sourced from unnamed sources.

Analysis

The signal here is not the nominal deal size; it is that immediately usable power and real estate are becoming the scarce inputs in AI infrastructure. That matters more for HUT than for a generic miner, because energized capacity can be repriced as a strategic asset rather than a commodity balance-sheet line item. For NVDA, the read-through is supportive but indirect: it reinforces demand durability for accelerated compute, yet most of the economic capture still sits with the operator and site host, not the chip vendor.

Second-order winners are companies with transferable megawatts, interconnects, and long-duration leases; second-order losers are miners and cloud operators without a power advantage. If this becomes a template, HUT’s multiple could migrate toward infrastructure/utility-style valuation, while pure hash-rate names risk compression as capital rotates toward AI hosting optionality. The catalyst path is months, not days: the market will need evidence on contract tenor, utilization, and margin per MW before paying up sustainably.

Contrarian view: the market may be overpricing the immediacy of monetization. A take-or-pay headline does not guarantee attractive cash economics, and if GPU supply loosens or AI demand normalizes, the scarcity premium can fade quickly. Falsifiers include weak follow-on disclosure on lease economics, softer hyperscaler capex commentary, or a failure to see HUT hold above the post-news range after the initial enthusiasm.

Key Decisions for Investors

• HUT: tactical long only on weakness, with a 1-3 month horizon. Best entry is a pullback that still holds the post-news VWAP; thesis is that the market will re-rate energized power capacity as AI infrastructure optionality. Falsify if disclosure shows low-margin hosting economics or no meaningful contracted duration.
• HUT vs. miner basket (RIOT/MARA): initiate a relative-value long HUT / short RIOT or MARA only if management later confirms multi-year AI utilization and favorable power pricing. The spread is attractive if investors start paying for megawatts and lease position rather than hash-rate.
• NVDA: treat as a validation trade, not a primary long. Add only on market weakness over the next 1-2 weeks if peer commentary confirms continued GPU scarcity; upside is steadier demand perception, but the news is likely already partly capitalized.
• Options alternative on HUT: use a 1-3 month call spread instead of common if you want exposure to rerating without taking full balance-sheet and execution risk. Exit if the stock loses the initial breakout level or if subsequent filings suggest the AI revenue stream is incidental.