The AI Capex Cycle's Real Winners Own Kilowatts, Not Code

Three things happened on May 7, 2026 that, read in isolation, look like unrelated corporate news. Read together, they constitute a single argument about who captures value in a world spending north of $700 billion annually on AI infrastructure.

First: IREN Limited (NASDAQ: IREN) announced a five-year, $3.4 billion AI cloud contract with Nvidia — not a colocation deal, not bare metal, but fully managed GPU cloud services for Nvidia's own internal AI and research workloads. Simultaneously, Nvidia received a five-year right to purchase up to 30 million IREN shares at $70 per share, a potential $2.1 billion equity investment. Stacked against the $9.7 billion Microsoft deal signed in November 2025, IREN now sits on more than $13 billion in contracted AI revenue — anchored by two of the four most powerful technology companies on earth. The company is targeting $3.7 billion in annualized run-rate revenue by end of calendar 2026.

Second: Hut 8, another former Bitcoin miner, closed a 15-year, $9.8 billion lease at its Beacon Point campus in Texas, bringing its total contracted AI data center capacity to 597 MW with aggregate base-term contract value of approximately $16.8 billion. Third — and most quietly devastating — Cloudflare reported record first-quarter revenue of $639.8 million, up 34% year-over-year, and simultaneously announced it was eliminating 1,100 jobs, roughly 20% of its entire workforce, because internal AI usage had risen 600% in three months. My thesis: the kilowatt is now the kingmaker input of the AI economy, ex-Bitcoin miners hold it in bulk, and the Cloudflare moment is the clearest signal yet that the companies who do not control physical infrastructure are being structurally deflated even as their revenues climb.

The IREN Blueprint: Power + Land + Managed Stack = Irreplaceable Counterparty

The framing that IREN is a Bitcoin miner that stumbled into AI undersells what actually happened strategically. IREN is a company that spent years solving the hardest, least glamorous problems in high-density computing: sourcing grid-connected land, negotiating utility interconnects, building and operating data centers in power-constrained geographies at scale. Those capabilities were originally applied to mining Bitcoin. They are now being redirected to the highest-margin, highest-demand workload on the planet.

The May 7 Nvidia deal makes this concrete. Per the official Nvidia Newsroom announcement, the two companies announced a strategic partnership to accelerate deployment of next-generation AI infrastructure, with NVIDIA and IREN intending to support deployment of up to 5 gigawatts of NVIDIA DSX-aligned AI infrastructure across IREN's global data center pipeline over time. Per the GlobeNewswire press release on the cloud contract, the agreement will be serviced by air-cooled Blackwell platform systems deployed within approximately 60 MW of IREN's existing data centers at its Childress, Texas campus. This is not a speculative pipeline announcement. Nvidia is writing a five-year, $3.4 billion check to consume IREN's own managed services for its internal AI research workloads — meaning Nvidia itself, the company that designs the chips everyone is fighting over, trusts IREN to run them.

The equity warrant structure is the part analysts are underweighting. As part of the partnership, IREN issued to NVIDIA a five-year right to purchase up to 30 million shares of ordinary stock at an exercise price of $70 per share, resulting in a right to invest up to $2.1 billion. This converts Nvidia from a supplier into a co-investor with direct financial upside tied to IREN's equity appreciation. Nvidia now has a strong economic incentive to route Blackwell GPU allocation, DSX architecture support, and customer referrals to IREN ahead of competing operators. That is a structural competitive advantage that no amount of capital can simply replicate — you cannot buy your way into a $2.1 billion equity alignment with Jensen Huang's company on short notice.

The Q3 FY26 results filing shows IREN reporting $3.1 billion in ARR under contract — comprising expected average annual revenue under the Microsoft deal, expected average annual revenue under the Nvidia contract, and $0.5 billion ARR from GPU deployments at Prince George — with a stated target of $3.7 billion ARR by end of calendar year 2026. The company reported $2.6 billion in cash as of April 30, 2026. Read our full breakdown of IREN's $3.4B Nvidia deal for the granular mechanics.

"AI factories are becoming foundational infrastructure for the global economy. Deploying these systems at scale requires deep integration across the full stack — compute, networking, software, power and operations." — Jensen Huang, CEO of NVIDIA, May 7 2026 strategic partnership announcement

Hut 8 Proves the Template Travels: Power-First Is the Repeatable Model

If the IREN deal were an anomaly — one unusually well-positioned former miner catching a lucky break — you could dismiss it as sector noise. But Hut 8's $9.8 billion Beacon Point lease, announced the same day, shows the template is repeatable and that the market for power-first AI infrastructure landlords is broadening fast.

Per the official PR Newswire release, Hut 8 is delivering a 352 MW AI factory designed to NVIDIA's DSX reference architecture for gigawatt-scale AI infrastructure — a 15-year lease with a confidential, high-investment-grade tenant, including a 3% annual base rent escalator. With three optional five-year renewal terms, the potential contract value could rise to roughly $25.1 billion. Hut 8's CEO Asher Genoot described the company's approach as a "power-first, innovation-driven development model" — language that is nearly interchangeable with how IREN's Daniel Roberts frames his company's edge.

Critically, Beacon Point is Hut 8's second major AI data center deal. The first was the $7.0 billion River Bend lease with Fluidstack, backstopped by Google (a subsidiary of Alphabet) — with Google providing financial support covering the lease payments and related pass-through obligations. Together, the two campuses give Hut 8 597 MW of contracted IT capacity and approximately $16.8 billion in aggregate base-term contract value. This is a company that was operating Bitcoin miners eighteen months ago. It is now one of the largest contracted AI infrastructure landlords in the Western hemisphere. See our news brief on Hut 8's $9.8B AI lease and the 30% stock surge for the full deal terms.

The pattern is unmistakable: hyperscalers and frontier AI labs need enormous amounts of power-dense, purpose-built compute capacity immediately, and they are finding that the fastest path to it runs through former Bitcoin mining campuses — sites that already have utility interconnects, permitted electrical infrastructure, and operational teams experienced with high-density, always-on compute. The alternative — greenfield hyperscale construction — takes five to seven years and faces a permitting and grid interconnect backlog that is not clearing in the near term.

The Cloudflare Paradox: Record Revenue, Vanishing Headcount — and What It Means for Infrastructure

The Cloudflare announcement on May 7 is the third leg of this argument — and in some ways the most important, because it tells you what is being destroyed in the same moment that kilowatt-backed infrastructure is being created.

Cloudflare (NYSE: NET) reported $639.8 million in Q1 2026 revenue, up 34% year-over-year — a record quarter, above analyst estimates, with full-year guidance raised. On the same day, the company announced it was cutting 1,100 employees, approximately 20% of its total workforce — the largest layoff in the company's 16-year history. The stated reason, per the official Cloudflare blog post titled 'Building for the Future', co-authored by CEO Matthew Prince and co-founder Michelle Zatlyn: Cloudflare's usage of AI has increased by more than 600% in the last three months alone, with employees across engineering, HR, finance, and marketing running thousands of AI agent sessions each day.

Prince and Zatlyn wrote that the cuts are explicitly "not a cost-cutting exercise or an assessment of individuals' performance" — they are "about Cloudflare defining how a world-class, high-growth company operates and creates value in the agentic AI era." The market punished the stock anyway, sending shares down roughly 18–24% in the session following the announcement. The market's reaction is informative: investors are not sold that a software company's value proposition is enhanced by replacing people with agents, particularly when the productivity gains are self-reported and the Q2 guidance was only marginally above consensus.

But here is the second-order point that matters for our thesis: every dollar that Cloudflare — and the dozens of software companies that will follow its lead this earnings cycle — does not spend on human headcount, it will redirect toward AI compute. That compute has to run somewhere. It runs in data centers. Those data centers require power. The power is scarce. The entities controlling the power are IREN, Hut 8, Core Scientific, Terawulf, and a small cohort of grid-connected operators who spent the Bitcoin cycle building exactly this asset base. The more aggressively the software layer automates itself, the more desperately it needs the physical layer that ex-miners built. This dynamic is not yet priced into the broader software sector's valuations, which remain elevated against shrinking headcount assumptions that haven't yet been stress-tested against genuine AI productivity audits.

For context on how agentic AI is reshaping commerce and payments infrastructure in parallel, see our analysis of AI agents driving stablecoin's second wave — the same agent proliferation powering Cloudflare's internal efficiency gains is the force building autonomous payment rails.

Supply Constraint Is the Thesis: Blackwell Is Still Scarce, and So Is the Land to Run It On

The bull case for ex-Bitcoin miners as AI infrastructure landlords rests on a specific supply/demand asymmetry: Nvidia Blackwell GPUs remain supply-constrained through at least 2027, and the data center capacity to house them is even more constrained. You cannot separate these two bottlenecks. A Blackwell GPU sitting in a warehouse without grid power, cooling infrastructure, and an operational team is worth nothing to the hyperscaler that bought it. The GPU is the headline; the kilowatt is the actual binding constraint.

IREN understood this before most. Per its own corporate description, IREN's platform is underpinned by its expansive portfolio of grid-connected land and power in renewable-rich regions across North America, Europe, and APAC — a characterization that appeared in both the Nvidia Newsroom release and IREN's own Q3 filing. The company has spent the past decade assembling that portfolio for Bitcoin. The infrastructure is now being redirected to Blackwell clusters. That redeployment required no new land acquisition, no new utility negotiations, no new permitting battles. It required swapping ASIC mining rigs for GPU racks — operationally complex, but nowhere near as hard as building from scratch.

The IREN deal structure — managed cloud services rather than bare metal colocation — signals another upgrade in the value chain. As IREN Co-CEO Daniel Roberts stated in the press release: "This contract demonstrates our ability to deliver fully managed cloud solutions, not just bare metal, to a broad and growing customer base." Moving up the stack from landlord to managed service provider expands margin and stickiness simultaneously. And the acquisition of Mirantis — an orchestration and cluster management software company — announced the same week, completes the vertical integration picture: IREN now controls power, land, data center buildout, GPU deployment, and software orchestration. That is a meaningful moat at a moment when Blackwell deployments are failing not because of hardware but because of software integration complexity.

The financial market is beginning to catch up. Bernstein has placed a $100 price target on IREN; Compass Point sits at $105. The stock trades near $61, implying roughly 65% upside to the lower of the two targets — against a contracted revenue backlog that dwarfs the current market cap. The bear case is real: IREN posted a $247.8 million net loss in Q3 FY26, driven primarily by decommissioning costs and unrealized losses on convertible note hedges during the Bitcoin-to-GPU transition. Execution risk on a 5 GW global deployment at scale is not trivial. But contrast that operational risk with the demand risk facing pure-software peers — IREN's problem is delivering on signed contracts; Cloudflare's problem is convincing the market that agents can replace the people it just fired. I know which problem I'd rather have.

The same repricing logic applies across the cohort. Core Scientific and Terawulf have each signed multi-billion-dollar compute deals in prior quarters. The deal pace and deal size are accelerating, not decelerating — and the addressable market, against ~$700 billion in projected 2026 global AI infrastructure spend, has barely been touched. For a parallel view on how infrastructure control translates to payment layer dominance, see our piece on Solana as the AI payment layer — the same infrastructure-first logic applies in compute and in settlement. And for how Coinbase is navigating its own infrastructure pivot away from transaction-fee dependency, see Coinbase's next infrastructure bet.

Second-Order Consequences: What This Repricing Means Beyond the Miner Sector

Let me be direct about the second-order consequences, because this is where the analysis goes beyond a simple "buy the miners" call.

First, the equity warrant structure pioneered in the IREN-Nvidia deal will spread. When a $3 trillion chip company finds it advantageous to align its equity interests with a data center operator rather than simply charging it for GPUs, every other major GPU allocator will consider the same playbook. Expect AMD and potentially Intel to initiate warrant-based partnerships with power-rich operators in the next two to four quarters. This converts the data center operator sector from a transactional vendor category into a strategic equity ecosystem — a fundamentally different valuation framework.

Second, the triple-net lease structure that Hut 8 is executing — where an investment-grade tenant (backed by Google's credit rating) signs a 15-year NNN lease with 3% annual escalators — transforms Bitcoin mining real estate into investment-grade infrastructure bonds. The River Bend deal press release notes that Google is providing a financial backstop covering lease payments and related pass-through obligations. That credit support enabled Hut 8 to access the investment-grade bond market — the first single-sponsor data center project to do so at construction stage. This opens an entirely new capital formation pathway for the sector. Operators who control the right land and power can now securitize their contracted cash flows against hyperscale credit, dramatically reducing their cost of capital and widening the competitive moat against new entrants who cannot access that financing structure.

Third, and most broadly: the Cloudflare moment is a preview of what happens to the software sector's labor model over the next 18 months. Amazon, Microsoft, Alphabet, and Meta plan to spend approximately $725 billion on AI infrastructure in 2026 — a figure that flows almost entirely into data centers, custom chips, GPUs, and AI models. That capital has to touch physical infrastructure. The companies that own that infrastructure — the former Bitcoin miners who solved power at scale during the most brutal bear market in crypto history — are sitting at the intersection of the largest capital deployment cycle in technology history and a structural supply constraint that is not resolving quickly. The market is still in the early innings of repricing this reality.

Beyond the Headlines: The 2026 AI capex supercycle is producing a violent and underappreciated bifurcation. Companies with atoms — power, land, cooling, operational teams — are signing decade-long contracts with the world's most creditworthy tenants and issuing equity warrants to chip giants. Companies with only bits — however well-coded, however cleverly marketed — are finding that their most expensive input, human labor, is being automated away by the very AI systems they sell. IREN's trajectory from sub-$70M market cap in 2022 to over $20 billion today is not a meme stock story. It is the most accurate real-time signal we have of where the AI economy is concentrating its durable value: not in the model, not in the chip, but in the kilowatt that runs both. The miners who survived Bitcoin's worst years by optimizing every watt now find themselves holding the keys to the AI decade. That is not luck. That is asset-class transformation hiding in plain sight.

Sources

Primary sources and prior BlockAI News coverage referenced in this article.

Primary sources

• Nvidia Newsroom — NVIDIA & IREN 5GW strategic partnership announcement
• GlobeNewswire — IREN $3.4bn AI Cloud Contract with NVIDIA
• GlobeNewswire — IREN Q3 FY26 Business Update & Financial Results
• PR Newswire — Hut 8 $9.8B Beacon Point AI Data Center Lease
• PR Newswire — Hut 8 $7.0B River Bend Lease with Fluidstack
• Cloudflare Blog — 'Building for the Future' (layoff memo, May 7 2026)

From BlockAI News

• IREN's $3.4B Nvidia deal breakdown
• Hut 8's $9.8B AI lease surge
• Coinbase's next infrastructure bet
• AI agents driving stablecoin's second wave
• Solana as the AI payment layer

Stay close to BlockAI News.

• 📧 Subscribe → daily Web3 × AI brief
• 💬 Join the community on Telegram
• 𝕏 Follow @BlockAI_News on X

If this week's deal flow doesn't convince you that the kilowatt is the new kingmaker input of the AI economy, wait until Q4 — the next wave of mining-to-AI pivots is already in term sheets.

How we report: This article cites primary sources, regulatory filings, and on-chain data where available. BlockAI News uses AI tools to assist with research and first-draft generation; every article is reviewed and edited by a human editor before publication. Read our full How We Report page, Editorial Policy, AI Use Policy, and Corrections Policy.