IREN Hands Microsoft Horizon 1, Secures Elite NVIDIA Status
IREN Delivers First AI Supercomputer to Microsoft in Record Time
What Happened? (The Big News)
Imagine building a massive, super-powered computer brain for artificial intelligence—and handing over the keys to Microsoft in just a few months. That’s exactly what IREN Limited (a company listed on NASDAQ under the ticker IREN) just did.
On August 13, 2026, IREN announced that Horizon 1—the first of four planned AI "supercomputer campuses"—has been delivered to and accepted by Microsoft. This isn’t just any delivery; it’s part of a five-year, $9.7 billion contract signed in November 2025.
Think of it like this: IREN built a brand-new, state-of-the-art AI factory. Microsoft just moved in, turned on the lights, and said, "Yep, this works perfectly."
What Is Horizon 1? (Breaking It Down)
Horizon 1 is the first building block of a much bigger plan. Here’s what makes it special:
- Location: Childress, Texas — a campus with lots of land and access to renewable energy
- Power: 50 megawatts (MW) of IT load — that’s enough electricity to power roughly 37,500 homes, all dedicated to running AI chips
- Cooling: Direct-to-chip liquid cooling — instead of blowing air (like a fan on your laptop), cold liquid runs directly over the chips to keep them from overheating
- Purpose: Built specifically for AI training and inference (teaching AI models and letting them do work)
- Timeline: Delivered in 2026 — the first of four identical 50MW deployments planned for this year
ELI5: What is "direct-to-chip liquid cooling"?
Imagine your gaming laptop gets hot. Now imagine instead of a fan, tiny pipes carry cold water right onto the processor. That’s what this is — but at massive scale. It lets the chips run faster, longer, and harder without melting.
Why Is This a Big Deal? (The "So What?")
Three huge reasons this matters:
1. Speed of Execution
Most data centers take years to build. IREN did this in months. That’s because they control everything — from buying the land, to designing the building, to installing the pipes and plugging in the chips. This is called being vertically integrated.
2. NVIDIA Gold Star: "Exemplar Cloud" Status
After testing IREN’s setup with their newest NVIDIA GB300 NVL72 chips (the Ferrari of AI processors), NVIDIA gave IREN Exemplar Cloud status.
Translation: "This facility runs our hardest AI workloads flawlessly." It’s like getting a Michelin star for AI infrastructure.
3. Microsoft Trusts Them With $9.7 Billion
Microsoft didn’t just sign a check — they’re using the facility. That’s the ultimate vote of confidence.
How Did They Do It So Fast? (The Secret Sauce)
IREN’s vertically integrated model means they don’t wait on contractors, architects, or utility companies. They do it all in-house:
| Step | What They Control |
|---|---|
| 1 | Land & Power — They own land in renewable-rich areas (Texas, Europe, APAC) with grid connections already in place |
| 2 | Design — Custom-built for AI, not retrofitted from old server rooms |
| 3 | Engineering — Liquid cooling, power distribution, networking — all optimized for GPUs |
| 4 | Construction — Their own teams build it, no middlemen |
| 5 | Deployment — They rack, stack, cable, and test the GPUs themselves |
| 6 | Software — Their own cloud platform runs on top |
3,000+ people worked on Horizon 1 — engineers, electricians, pipefitters, software devs, project managers. Daniel Roberts (Co-Founder & Co-CEO) thanked them personally.
What’s Next? (Future Plans)
IREN isn’t stopping at Horizon 1. Here’s the roadmap:
| Year | Target Capacity | What That Means |
|---|---|---|
| 2026 | 480 MW (gross) | 9 more Horizons (4 more for Microsoft + 5 for other customers) |
| 2027 | 1.2 GW (gross) | Massive scale — enough for huge AI models like GPT-6 or beyond |
Gross vs. IT Load: "Gross" includes cooling, lighting, losses. "IT Load" is just the chips. 1.2 GW gross ≈ 900+ MW for actual AI compute.
IMPORTANT POINT: FORWARD-LOOKING STATEMENTS
The plans for 2026 and 2027 are targets, not promises. IREN’s press release includes a legal disclaimer: things like supply delays, permitting issues, or market changes could shift timelines. Always check their SEC filings (like the 10-K from August 2025) for risk factors.
Summary
- Horizon 1 delivered to Microsoft at Childress, Texas
- 50 MW of liquid-cooled AI compute — first of 4 in 2026
- Part of $9.7B / 5-year contract with Microsoft
- Earned NVIDIA Exemplar Cloud status on newest GB300 NVL72 chips
- Built fast thanks to vertically integrated model (land → power → build → software)
- 3,000+ workers made it happen
- Next: Horizons 2–4 this year, then 480 MW by end of 2026, 1.2 GW by 2027
FAQ
What does "50 MW IT load" actually mean?
A: It means the facility delivers 50 megawatts of electricity directly to the AI chips (the "IT" part). The total power used by the building (including cooling, lights, losses) is higher — that’s "gross" power.
Why does liquid cooling matter for AI?
A: Modern AI chips (like NVIDIA’s GB300) generate extreme heat — far more than air cooling can handle. Liquid cooling lets them run at full speed 24/7 without throttling or failing.
What is "NVIDIA Exemplar Cloud status"?
A: It’s a certification from NVIDIA saying: "We tested our hardest workloads on your hardware, and it passed with flying colors." Only top-tier partners get it.
What does "vertically integrated" mean in plain English?
A: IREN doesn’t hire outside companies to design, build, or run their data centers. They do it all themselves — like Apple making their own chips, phones, and stores.
Is this only for Microsoft?
A: No! Horizons 2–4 are for Microsoft (per the contract), but IREN is building additional capacity for other customers too. Their 480 MW / 1.2 GW targets include all deployments.
Want to learn more? Visit iren.com or contact investors at ir@iren.com / media at media@iren.com.