Abstract editorial illustration of interlocking glass and steel data-centre towers linked by glowing amber contract lines, symbolising the chain of leasing and financing behind AI compute deals

Anthropic’s $35bn Lambda Deal: Nvidia Becomes AI’s Landlord, Not Just Its Chip Supplier

Currat_Admin
14 Min Read
Disclosure: This website may contain affiliate links, which means I may earn a commission if you click on the link and make a purchase. I only recommend products or services that I will personally use and believe will add value to my readers. Your support is appreciated!
- Advertisement -

🎙️ Listen to this post: Anthropic’s $35bn Lambda Deal: Nvidia Becomes AI’s Landlord, Not Just Its Chip Supplier

0:00 / --:--
Ready to play

Abstract editorial illustration of interlocking glass and steel data-centre towers linked by glowing amber contract lines, symbolising the chain of leasing and financing behind AI compute deals

Last updated: 2 September 2026. Figures below are sourced to primary company filings and named publications, linked inline; where a number is self-reported by a company, that is stated explicitly.

The 60-second version

  • Anthropic has reportedly agreed to spend $35bn over six years with cloud provider Lambda for AI computing capacity, according to Bloomberg and the Wall Street Journal — the deal has not been confirmed by an official press release from any of the three companies involved.
  • The compute is understood to come from Beacon Point, a 1-gigawatt data-centre campus in Nueces County, Texas, built by Hut 8. Hut 8’s own filings describe the tenant only as a “high-investment-grade company” — the Financial Times has reported, but Hut 8 has not confirmed, that the tenant is Nvidia.
  • That creates a four-party chain — Hut 8 leases the building to Nvidia, Nvidia supplies chips to Lambda, Lambda resells capacity to Anthropic — that is unusual because it puts the world’s biggest chipmaker in the landlord’s seat, not just the vendor’s.
  • It is Anthropic’s second mega compute deal within a week, following a $45bn, six-year commitment to British firm Nscale on 26–27 August, which we covered in detail here.
  • Headline “total contract value” figures in these deals — as high as $50.2bn for the Hut 8 campus alone — include optional lease renewals nobody has yet committed to exercising, which inflates the numbers reported in most coverage.

Key numbers at a glance

Date Development Figure
20 Jul 2026 Hut 8 fully commercialises its 704MW Beacon Point campus with a second 15-year lease $19.6bn combined base-term contract value (up to $50.2bn with renewals)
17 Aug 2026 Anthropic’s annualised revenue run rate, self-reported ahead of a planned IPO $65bn, per Bloomberg and Axios
26–27 Aug 2026 Anthropic signs compute deal with Nscale (West Virginia, Nvidia Vera Rubin chips) $45bn over six years
31 Aug–1 Sept 2026 Anthropic reportedly signs compute deal with Lambda (Beacon Point, Texas) $35bn over six years

Two mega deals, worth a combined $80bn on paper, landed within roughly a week of each other — and both run through Nvidia-linked infrastructure.

What actually happened

According to Bloomberg, whose reporting was followed by the Wall Street Journal and outlets including Forbes, Anthropic has committed $35bn over six years to Lambda, the GPU cloud provider Nvidia has backed since taking part in its $480m funding round in February 2025. The capacity is expected to be drawn from Beacon Point, the Texas campus Hut 8 is building in Nueces County — the same site at the centre of a separate, already-public leasing arrangement with Nvidia.

None of the three companies — Anthropic, Lambda or Nvidia — has issued its own statement confirming the deal at the time of writing. That matters, because it means every figure in this story sits somewhere on a spectrum from “filed with a stock exchange” to “reported by people familiar with the matter,” and readers deserve to know which is which.

- Advertisement -

The chain nobody drew until now

Strip away the company names and the structure is straightforward, even if the reporting has arrived in fragments over several months:

  • Hut 8 builds and owns the Beacon Point data centre and signs the underlying real-estate lease.
  • Nvidia — per Financial Times reporting that Hut 8 itself has not confirmed by name — is the tenant on that lease, paying to secure the physical building and power capacity.
  • Lambda, an Nvidia-backed cloud reseller, deploys Nvidia GPUs into that leased capacity and sells compute time to customers.
  • Anthropic buys that compute time to train and run Claude.

Hut 8’s own 20 July 2026 press release is the one document in this chain that is unambiguous: a 15-year, triple-net lease with three five-year renewal options and a 3% annual rent escalator, covering 352 megawatts of IT capacity, worth $9.8bn in base-term value alone. Combined with an earlier, matching lease at the same campus, the release puts total base-term value at $19.6bn across 704MW — and notes that with every renewal option exercised, the figure could reach $50.2bn. Hut 8 does not name its tenant. The Nvidia identification comes from the Financial Times and has been repeated, not independently re-reported, by most of the outlets covering the Anthropic-Lambda story since.

This is the second time in a week that Nvidia’s fingerprints have shown up as something other than a chip supplier. It licensed its “model factory” technology to Poolside for $6bn in a deal we covered here, and its Vera Rubin chips are the substrate for Anthropic’s separate Nscale commitment too. The common thread across both deals is a company that used to sell hardware now underwriting the real estate, the financing and, per this reporting, even the lease that makes the hardware usable.

Why the circularity matters

None of this is illegal or even unusual by the standards of how infrastructure gets built — utilities, telecoms and railways have long used vendor financing and anchor-tenant leases to get capital-intensive projects off the ground. What’s new is the scale and the concentration: one company (Nvidia) is simultaneously the equity backer of a cloud reseller (Lambda), the anchor tenant securing the building that reseller operates from, and the chip supplier those operations depend on. If Nvidia is paying to secure buildings that will only be profitable if AI labs keep buying vastly more compute, its incentive to keep those labs’ growth story alive is not incidental to its chip business — it is now load-bearing for its property exposure too.

That is the “circularity” concern that analysts have started flagging around AI infrastructure spending more broadly: money moving from Nvidia’s balance sheet into data-centre leases and cloud-provider equity, which then flows back to Nvidia as chip revenue and lease income, all before an end customer like Anthropic has necessarily proven the economics of the workloads those chips are running. Anthropic’s own numbers offer some reassurance here — Bloomberg and Axios both reported the company’s annualised revenue run rate reached $65bn as of mid-August 2026, a figure Anthropic disclosed itself ahead of a reported autumn IPO process, and which is not yet independently audited in public filings. Real revenue growth of that magnitude is a genuine counterweight to circularity worries. But it doesn’t resolve the separate question of whether the Hut 8/Nvidia/Lambda chain specifically is being sized to Anthropic’s actual near-term compute needs, or to the much larger renewal-inclusive numbers that make for bigger headlines.

- Advertisement -

What most coverage is getting wrong

Most write-ups of this deal have led with the $35bn headline figure and treated it as directly comparable to the $45bn Nscale number from the same week, implying Anthropic has committed roughly $80bn in new spending in a matter of days. That framing skips two important qualifications the underlying filings and reporting actually support: first, that “total contract value” in deals like these routinely bundles in optional renewals that may never be exercised — Hut 8’s own base-term figure for its full campus is $19.6bn, not the $50.2bn headline some coverage has used; and second, that the Lambda deal itself is sourced entirely to unnamed people familiar with the matter, not to a filing or joint statement, which is a meaningfully different evidentiary standard than the Nscale deal, which Nscale has discussed on the record in the context of its own IPO plans.

Practical takeaways for builders, publishers and operators

If you plan around the cost or availability of frontier-model compute, three things are worth doing differently after a week like this one. Read “total contract value” as a ceiling, not a spend commitment — when a headline cites a renewal-inclusive number, look for the base-term figure in the underlying filing and use that for any planning assumption. Separate confirmed capacity from reported capacity — Anthropic’s Nscale compute doesn’t begin flowing until late 2027 by the company’s own account, and the Lambda capacity, if the deal is accurate as reported, is tied to a Texas campus whose phases energise across 2027 and 2028; neither materially changes GPU availability or API pricing this quarter. And track the financing structure, not just the dollar figure, when you’re assessing how durable a given cloud provider’s capacity commitments are — a reseller sitting on leased-not-owned infrastructure, several layers removed from the underlying chip supplier, carries different counterparty risk than a hyperscaler with its own balance sheet and data centres, which is worth knowing before you build a product roadmap around a specific vendor’s promised capacity.

What we still don’t know

  • Whether Nvidia is, in fact, the Hut 8 tenant. Hut 8 has never named it; the identification rests on Financial Times sourcing that other outlets have repeated rather than independently confirmed.
  • Whether the Anthropic-Lambda deal is a signed, binding contract or a preliminary commitment. No company has published terms, a start date for billing, or a termination clause.
  • What Lambda actually pays Nvidia for chip access and, per this reporting, sub-leased capacity — those commercial terms are undisclosed, which makes it impossible to independently verify how much margin sits at each layer of the chain.
  • How much of Anthropic’s roughly $80bn in compute commitments announced this week is incremental new capacity versus options and renewals it may never draw down.
  • Whether any regulator is looking at the competitive dynamics of a chip supplier taking financial positions this deep into its own customers’ supply chains — a question that sits alongside the broader debate we’ve tracked in our look at whether AI governance can keep pace with the industry.

FAQ

Is the Anthropic-Lambda deal officially confirmed?
No. It is reported by Bloomberg and the Wall Street Journal, citing people familiar with the matter. None of Anthropic, Lambda or Nvidia has issued a public statement confirming the $35bn figure or the deal’s terms as of this article’s publication.

- Advertisement -

Is Nvidia really Anthropic’s landlord?
Not directly. The reported chain is Hut 8 (owner/developer) leasing to Nvidia (tenant, per Financial Times reporting, unconfirmed by Hut 8), which supplies chips to Lambda, which sells compute to Anthropic. Nvidia is not reported to have any direct lease or contract with Anthropic itself.

How does this compare to the Nscale deal covered here last week?
They’re structurally different. Nscale is a UK-based infrastructure developer that has discussed its Anthropic deal publicly in the context of its own IPO plans; see our earlier coverage. The Lambda deal, by contrast, rests entirely on anonymously sourced reporting so far.

Does this affect the price or availability of AI tools people use today?
Not immediately. The capacity involved in both deals is scheduled to come online from late 2027 into 2028, so neither changes GPU supply or API pricing in the near term. For the medium-term picture, see our explainer on the next wave of AI hardware and infrastructure investment.

Sources

Please follow and like us:
Pin Share
- Advertisement -
Share This Article
Leave a Comment