Hut 8 (NASDAQ: HUT) announced on July 20, 2026, that it has signed a second 15-year lease at its Beacon Point campus in Nueces County, Texas, bringing total committed contract value for the site to $19.6 billion. The company describes the transaction as fully commercializing the 1 GW facility, a characterization that reflects the signing of leases across both development phases rather than the metering of every megawatt: the two agreements together cover 704 MW of the campus's 1,000 MW total capacity, with the remaining 296 MW not yet contracted.

On the same day, Australia-founded IREN Limited (NASDAQ: IREN) disclosed $2.8 billion in new multi-year cloud contracts with AI developers including Microsoft, NVIDIA, Perplexity, and Figure AI, while raising its 2026 annualized run-rate revenue (ARR) target above $4 billion.

Hut 8 shares rose roughly 16 percent to above $106. IREN gained approximately 10 percent in pre-market trading, partially recovering from a 41 percent decline over the prior month.

---

Beacon Point Now Fully Leased

The second Hut 8 lease covers 352 megawatts of IT capacity, matching the first deal signed in May 2026. Both leases involve the same investment-grade tenant, whose identity has not been disclosed. The second lease carries a base-term value of $9.8 billion with a 3.0 percent annual escalator built in. Together, the two agreements cover 704 MW of the campus's 1,000 MW total capacity.

If the tenant exercises all three available five-year renewal options, the total potential value of the Beacon Point contracts reaches $50.2 billion.

Hut 8 CEO Asher Genoot described the outcome as direct validation of the company's development model. "The real test of our power-first approach is what partners commit," Genoot said. He added: "Beacon Point underscores why we start with power and maintain flexibility across end markets. Operating across multiple applications lets us underwrite assets that single-use-case developers cannot, then redirect them toward higher-value commercialization pathways as demand evolves."

The second phase of the campus is built to NVIDIA's DSX reference architecture, with initial energization targeted for Q1 2027 and full delivery expected in Q2 2028. The campus is served by an AEP Texas grid interconnection, a structural asset that underpins the power-first sequencing Genoot describes.

Across its full portfolio, which includes the River Bend campus in addition to Beacon Point, Hut 8 now holds 949 MW of contracted AI capacity backed by $26.6 billion in aggregate base-term contract value. The company projects average annual net operating income exceeding $1.75 billion once sites stabilize, with Beacon Point alone contributing $1.31 billion per year.

---

IREN's Customer Base Widens

IREN co-CEO Daniel Roberts framed the new contracts as evidence of broadening demand across frontier applications. "In the past 12 months we have expanded from approximately 3 MW of self-built AI cloud capacity to 480 MW being delivered this year," Roberts said, describing a customer base that now spans hyperscalers, enterprises, and AI developers working across design, physical AI and robotics, generative media, AI search, and model development.

The $2.8 billion announcement covers new multi-year cloud contracts with Microsoft, NVIDIA, Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, and one undisclosed party. Microsoft and NVIDIA are also IREN's longest-standing major customers: the company previously signed five-year contracts valued at approximately $9.7 billion with NVIDIA and $9.7 billion with Microsoft for GB300 GPU infrastructure. Today's announcement represents additional or expanded commitments from both companies alongside a broadening roster of AI developers.

IREN said roughly 85 percent of its revised $4 billion annualized run-rate revenue (ARR) target is now under contract.

The scale-up has been rapid. Twelve months ago the company operated approximately 3 MW of self-built AI cloud capacity. It expects to reach 480 MW by the end of 2026 and is targeting 1.2 GW for 2027. Customer prepayments cover about 45 percent of the associated GPU capital costs. As of June 30, 2026, IREN held $7.6 billion in cash and equivalents, according to its SEC filing. These contracts are being fulfilled through IREN's Horizon campus in Childress, Texas.

---

The Pivot in Context

Both companies were primarily Bitcoin miners as recently as 2023. Their current contract scale reflects a broader sector shift. By October 2025, formerly mining-focused companies had collectively announced roughly $65 billion in AI infrastructure contracts. According to analysts tracking the sector, mining revenue, which represented around 85 percent of total revenue for these companies in early 2025, could fall below 20 percent of revenue by year-end 2026 for firms that have secured major AI deals. Those same analysts estimate operating margins on AI colocation agreements in the range of 80 to 90 percent, substantially above typical mining margins.

The competitive advantage these companies brought to the transition was not technology but physical infrastructure: large power procurement agreements, thermal management expertise, and experience running bare-metal server operations around the clock.

---

What This Means Outside the United States

The concentration of contracted AI compute in Texas creates practical constraints for developers and researchers in other regions. India is building toward a target of 100,000 publicly accessible GPUs by December 2026, yet that entire national target represents a fraction of the capacity locked into a single Hut 8 lease. That comparison spans different units of measurement, since GPU counts and megawatts of IT capacity are not directly equivalent, but the directional gap is significant regardless of the conversion method applied.

South Asian developers accessing AI services built on US infrastructure pay USD-denominated API pricing, which carries a significant currency premium relative to local incomes. Several of the IREN customers announced today, including Perplexity and Together AI, operate inference services accessible globally, meaning South Asian developers are among the end users of the capacity being built at Horizon in Childress, Texas.

The gap is wider still for Africa. The continent holds less than one percent of global data center capacity, according to the LSE Africa Blog and AI Hub for Development, and researchers estimate that only about five percent of African AI practitioners have access to compute sufficient for complex model development. Africans represent less than 0.5 percent of global participation in machine learning and large language model development, figures that reflect both the infrastructure deficit and its downstream consequences for research output. The multi-billion-dollar contracts announced today involve US-listed companies operating US-based infrastructure. None of this capital is directed toward African AI capacity.

Decentralized compute networks such as Akash Network, io.net, and Render Network are increasingly positioning themselves as accessible alternatives for developers priced out of hyperscale cloud markets, though their current scale remains considerably below that of centralized builds such as Beacon Point or Horizon.

---

What Comes Next

The July 20 announcements reinforce that power access has replaced capital as the binding constraint for AI infrastructure. Analysts widely identified Q1 2026 as the moment this shift became definitive across the sector.

Hut 8 secured its AEP Texas grid interconnection at Beacon Point before signing tenants, a sequencing that has proved decisive in attracting investment-grade commitments at this scale. For the broader sector, the question is how many similar transactions remain available as ERCOT capacity tightens and interconnection queues grow, though that forward-looking question remains a matter of active analyst debate rather than settled consensus.

For developers outside the US, the more immediate question is whether sovereign compute initiatives in India or sub-Saharan Africa can reach contracted, grid-connected scale before the current generation of US-built capacity sets the pricing and access conditions for the next decade of AI development.