Of Compute Deals, AI Infrastructure and Texas
Alongside Nvidia's $500-billion compute deal, projects in Texas reveal milestones
Playing the AI infrastructure trade comes in a variety of ways. The more long-lived infrastructure play can consist of the stocks of neoclouds, data center-connected REITs, or hyperscalers themselves. The recent news with Nvidia’s partnership with financiers with KKR, Apollo Global Management, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs for a $500-billion compute financing platform is a sign of the times.
Conceptually, the large-scale investment will become diversified across a pool of capital beyond the Nvidias, Microsofts and Metas. This trend was already starting to emerge through Meta’s financing arrangements in El Paso and Louisiana’s Hyperion for its large-scale projects.
Constructing compute capacity
Last October 2025 at the North America Blockchain Summit (NABS), top public crypto miners discussed the market shifts from crypto mining facilities’ capacity to AI and high-performance computing. At the “Capacity is the New Commodity” talk at NABS, representatives of Iren, Core Scientific, and Riot were present. In an article I highlighted still-relevant takeaways, which I’ll update too:
1) “Interconnection queues were a problem to get sites energized. In Texas’ ERCOT grid, 189 gigawatts were noted as being in the queue.”
Now, that queue is over 400 (much of which is not real). The state is pausing to better review these project. Still, plans are in full swing for large players.
2) “Speed to being energized is more valued now because capacity is in crunch mode. Notably, sites that have power online within, say, 24 months, are much more valuable than those sites that are to be energized in 2030. There’s a premium on existing assets, or capacity, and existing sites that can power up quicker.”
The recent announcements by RIOT with chipmaker AMD and Anthropic are new tenants with nearly $10-billion contracted revenue, most of it Anthropic’s. Their deal has a 20-year life, with nearly 100MW in late 2027 and the rest for June 2028. Though this is near-term capacity, it reflects this premium (as witnessed by RIOT’s shares up) and the need for capacity. I met Riot’s chief executive in November 2023 at the same Blockchain summit, and these strategies has shifted considerably. The value of the connected power is being allocated to high-performance compute.
3) “Quality of assets matters. Part of that quality is when power is going to be available; the size of the facility (eg., 100 megawatts), and then the tenant that fulfills the lease, with creditworthiness important.”
In just the last year or less, chipmakers, model makers and hyperscalers have scooped up spare capacity or expanded existing plans, especially where interconnection was in process. For example, Meta’s El Paso campus was originally a $1.5 billion project in 2024, growing to $10 billion invested. BlackRock formed a venture with $12.5 billion in debt part of the financing arrangement. Thus it follows the group of financiers with Nvidia are launching a more concerted next step beyond the pure AI infrastructure (land, power infrastructure and networking) into a chips and infrastructure-related play.
Other AI infrastructure notables
In October of 2025, IREN was on blockchain summit stage. Their flagship Texas site has 750 MW of grid-connected capacity, 650 of which was energized and operating at August 2025. The Horizon I building, 50 MW purpose-built AI data center, had an infrastructure project cost of $300-350 million. (Five to six million per megawatt is below othe costs I’ve heard.) The whole project is called a $9.7 billion AI cloud 5-year agreement with Microsoft as the offtaker, my words, for 200 MW ultimately.
Separately, Nvidia has another 5-year agreement for further cloud and hardware capacity expected. A $5.8 billion hardware supply agreement with Dell Technologies is to utilize NVIDIA GB300 (Blackwell Ultra) GPU clusters. Bottomline: this capacity found lucrative buyers and 1 GW does not come as fast as the velocity of news announcements.
At the first Stargate Abilene, Texas, Microsoft also stepped in to absorb released capacity of 600MW intended for OpenAI. Microsoft secured the 700 MW of leased capacity.
Equally important is another approach connected to Anthropic and Macquarie Asset Management of Australia. Their focus is tied to supporting a global infrastructure play.
Link to my post here about it, with more about Anthropic.
Initially, the entire Stargate projects were announced as $500 billion in January of 2025. Fast forward and much has changed. As we observe, the $500 billion new financing venture is part of the developments of funding an AI buildout that has expanded.
NB: Below I describe an infrastructure development finance model in the making…
