Dear $IREN (-1,49 %) Investors,
last night I read through the next 40 pages, this time there is a small summary, although of course many interesting details/connections are left out - but better than nothing :)
1. key facts of the deal (should be known)
- Term: 5 years, annual turnover: USD 1.94 billion.
- Location: Childress, Texas, 200 MW IT capacity, 76,000 NVIDIA GB300 GPUs.
- IREN had to massively upgrade the data centers (Tier 3 redundancy, 200 kW rack density), which almost doubled CAPEX.
- Microsoft pays 20% of the total volume as prepayment - one of the most important components of the deal.
2. why the deal makes economic sense despite high capex
- The prepayment of USD 1.94bn vehemently lowers the effective GPU CAPEX (net only USD 3.86bn).
- The reimbursement of this prepayment only starts from year 3 - this improves the early cash flows and raises the IRR.
- IREN owns its data centers → no rental costs → higher long-term margins.
- Financing terms are extremely favored by Microsoft's credit rating.
3. returns: Significantly higher than we/the market think/thinks
At first glance, the key deal data looks more expensive than that of competitors such as Nscale or $NBIS (-0,3 %) but:
- Unlevered IRR (i.e. return on the project itself excl. debt): ~10-12% (without residual value), up to ~19% (with residual value).
- Levered IRR (return on equity after debt has been utilized): ~25-40% in the base case scenario, depending on debt.
- Long-term (2nd & 3rd 5-year cycle): Return on equity >60% possible once DC debt is reduced.
Background: The expensive data center infrastructure only needs to be built once - after that the net earnings per MW increase significantly.
4. industry comparison: $IREN (-1,49 %)
is in a better position than many believe
Although some competitors report higher revenue-per-GPU figures, the Irish deal delivers better results thanks to prepayment and infrastructure ownership. better returns on capital.
Nebius, for example, has no prepayment and unfortunately provides far fewer details about the deal, while Iren discloses everything.
5. the long-term picture
If IREN utilizes the Childress infrastructure over multiple deal cycles, each subsequent deal becomes
- each subsequent deal significantly more profitable,
- the free cash flow remains positive,
- The profit per MW is significantly higher than with traditional colocation providers.
Over 15 years, this results in a model in which:
- the cumulative cash balance rises above USD 1.3 billion,
- the return on equity over the entire pipeline becomes exceptionally attractive.
6 Strategic importance
- The deal is (hopefully) not an isolated case, but the starting point for scaling across the entire 3 GW pipeline.
- The technical architecture (supercluster, variable rack density, etc.) is a key differentiator.
Also important: Instead of installing the usual 22-23K GB300 per 50 MW IT (i.e. like the competition), IREN uses approx. 19K GPU per horizon = free electrical reserve.
There may be 2 possible explanations for this:
1) Microsoft already has an eye on Horizon 1-4 for a possible Rubin deal --> so halls are already technically designed so that MS can book follow-up capacity without IREN having to rebuild big again
2) Preparation for Rubin/Vera-Rubin cluster
The next GPU generation (Rubin) needs higher rack densities. Horizon Design is already designed for 200 kW per rack. This speaks in favor of a data center that now works with Blackwell chips, but is apparently built in such a way that future chip generations can be accommodated immediately.
Thesis 1 would also fit with the prepayment fact --> 1.94 billion prepayment is more likely to be made if a partnership is to be formed. In other words, not a classic customer, but a "main tenant". So maybe IREN is building one of the possible "compute parks" for Microsoft.
Let's see what 2026 brings.
Lg Max
