#AnthropicSigns35BCloudDeal

On August 31, 2026, Anthropic — the AI company behind Claude — signed a cloud-computing agreement worth $35 billion with Lambda, an Nvidia-backed cloud provider, according to Bloomberg and The Wall Street Journal. The infrastructure: a data center in Nueces County, Texas, developed by Hut 8, a bitcoin miner turned data-center developer — and the striking part: Nvidia itself will hold the lease. It is the latest in a stunning series of compute deals showing how AI consumes capital, electricity and chips like never before.

First, what is a cloud deal and why is it worth $35 billion? It means Anthropic is not building its own data centers; it rents massive computing power from a cloud provider running servers filled with AI chips. Models like Claude are trained and run on special chips — Nvidia GPUs, Google TPUs, custom accelerators — which need enormous data centers, cooling and electricity. Industry estimates put a one-gigawatt data center at around $50 billion, with roughly $35 billion, about 70%, going to the chips themselves. So $35 billion is essentially the chip bill for one massive data center — that is how expensive AI infrastructure has become.

And this deal is not isolated. Last week Anthropic committed $45 billion to Nscale for six years of Nvidia capacity (460 megawatts) in West Virginia. Before that came Google (up to one million custom TPUs, over a gigawatt of capacity), Amazon (up to five gigawatts), Google plus Broadcom (five gigawatts of TPUs), Microsoft ($30 billion of Azure compute plus one gigawatt of Nvidia hardware) and SpaceX (GPU capacity in its Colossus centers). Total: roughly 12 to 13 gigawatts across six providers — one of the largest compute procurement programs ever assembled.

The numbers behind Anthropic explain the urgency. Its annualized revenue run rate went from roughly $1 billion at the end of 2024 to about $5 billion by mid-2025, around $7 billion by October 2025, and then $30 to $40 billion by spring 2026 — driven mainly by the coding tool Claude Code and enterprise API usage. A 30- to 40-fold expansion in sixteen months. Valuation followed: $61.5 billion in March 2025, about $350 billion in November 2025 after a $15 billion Microsoft and Nvidia investment, then $965 billion post-money in May 2026 when it raised $65 billion in a Series H round led by Altimeter, Dragoneer, Greenoaks and Sequoia. That is a 15.7x increase in just over a year, and secondary trading already implies near $1 trillion. With an IPO expected, Anthropic reportedly plans to tell investors its potential revenue opportunity exceeds $30 trillion, topping SpaceX's $28.5 trillion.

Now zoom out to the industry, where percentages get wild. Dell'Oro projects worldwide data-center capex will surpass $3 trillion by 2030, with AI accelerators about a third of that. McKinsey estimates AI-related infrastructure needs $5.2 trillion by 2030, or about 156 gigawatts of capacity. UBS forecasts hyperscalers will spend $4.1 trillion on AI infrastructure from 2026 to 2028 — more than triple the $1.29 trillion of the previous six years — with capex rising from $1.009 trillion in 2026 to $1.619 trillion in 2028, a 60% jump in two years. Goldman Sachs expects data-center power consumption to rise 165% from 2023 to 2030; AI-optimized servers grow 30% a year versus 9% for regular ones, and one AI hyperscaler already uses electricity like 100,000 homes.

On the stock side, the deal is another tailwind for Nvidia, which closed Monday at $220.78, up 1.48%, after announcing a $20 billion quarterly buyback — about 0.38% of its market cap. Nvidia now sits near $5.25 trillion, the world's most valuable company, with a 52-week range of $164.07 to $236.54 and a price-to-earnings multiple in the high 30s. Yet Nvidia has gained only about 16% in 2026 versus the PHLX Semiconductor Index's 62% — the market questions whether even Nvidia can grow into this capex cycle. Analysts still see upside: average targets around $308 to $323, with Raymond James recently raising to $352. And note the structure: Nvidia invests in Lambda, leases the Texas site from Hut 8, and supplies the hardware — turning its balance sheet into a demand engine that locks in multi-year chip orders.

Now the risks, because honest analysis names them. First, circular financing: Nvidia invests in Lambda, Lambda buys Nvidia chips, Anthropic leases capacity — revenue circulates inside an ecosystem partly financed by Nvidia itself. Second, capex-to-revenue: Amazon, Microsoft and Alphabet are expected to spend about 102% of their cloud revenue on capex in 2026, recycling nearly all cloud income into AI infrastructure — sustainable only while revenue compounds. Third, the depreciation cliff: chips obsolete in two to three years mean brutal write-offs. Fourth, power: electricity, not chips, may be the real bottleneck. Fifth, valuation: even bulls admit Anthropic at 22x to 25x revenue is rich, and "run rate is a press-release number" — the $30 to $40 billion figure is company guidance, not audited reality. Finally, dependence: its multi-cloud spread across AWS, Google, Microsoft, Lambda and Nscale is partly a hedge against any single choke point.

Now my own opinion — numbers only tell half the story. In my view, this $35 billion deal is a structural signal, not just a headline. AI compute is becoming the electricity of the 21st century — no AI without chips, power and data centers, and whoever controls that layer controls the industry. Anthropic committing roughly $80 billion to compute in a single month ($45 billion to Nscale plus $35 billion to Lambda) while still private shows that compute, not just model quality, is the primary battlefield — a land grab for capacity years in advance. The smartest move in the whole story is Nvidia holding the lease itself: it monetizes its balance sheet, locks in demand for its own chips for years, and makes itself indispensable — strategic control no chipmaker has ever had. For investors, my take is this: the AI infrastructure theme is real and probably still early, but it is priced for perfection. Nvidia's 16% gain versus the sector's 62% tells you the market already fears the peak. Do not buy on headlines; watch utilization, cloud margins, and whether Claude's revenue keeps compounding. The winners will be those converting capex into durable recurring revenue, not the biggest spenders. Expect high volatility — power, capital and execution are the bottlenecks, and no industry grows capex 60% a year forever. But over the next decade, companies at the intersection of AI models, chips and data centers are building the infrastructure of everything. This deal is one small but very loud proof of that.

In short: AI is not slowing down, it is doubling down. Anthropic spends like a company that believes the future is compute-hungry, Nvidia is financing the shovel sales of the whole gold rush, and the only question is who turns the biggest spending into the biggest profits. That is the story behind the headline.

$NVDA ‌$SNDK ‌$TSLA ‌$MU ‌$AMZN ‌