Lando

$IREN New Contract Economics I have seen a lot of discussion on $IREN's new $2.8b of AI Cloud contracts, and I believe most people are underestimating just how attractive these contracts are. The most common assumption I've come across is that the new contracts have an average duration of roughly 3 years and generate around $3.6/GPU-hour. I believe the available evidence points elsewhere. Management didn't disclose the exact contract economics. However, they provided enough breadcrumbs throughout the announcement that we can reverse-engineer them with a surprisingly high degree of confidence. Here is the new information we got: - Combined value of new contracts: $2.8b - New year-end ARR guidance: over $4b - 85% of the new guidance has been contracted - Weighted average contract duration across $IREN's portfolio: ~4 years - Prepayments of new contracts amount to 45% of GPU related CAPEX Previously known information that remains useful: - New contracts are for a portion of the 50k B300 units that are being installed at Mackenzie & Childress this year - The recently signed $3.4b contract with $NVDA has a 5 year duration, paying ~$3.77 per GPUh - $MSFT accounts for $1.93b of year end’s ARR - PG accounts for $500m of year end’s ARR The reason many investors arrive at roughly $3.6/GPU-hour is because they assume $IREN's year-end ARR will be exactly $4b, even though management guided to over $4 billion. Starting with a $4b ARR target, they subtract Microsoft's ~$1.93b ARR and Prince George's $500m ARR, leaving roughly $1.57b of ARR attributable to the 50,000 GPUs being deployed across Childress and Mackenzie. $1.57b / 50,000 GPUs / 8,760 hours = ~$3.6/GPU-hour. The same assumption also leads to the commonly cited ~3-year contract duration. If total ARR is $4b, then 85% contracted equals $3.4b of contracted ARR. Subtract Microsoft's ~$1.93bn ARR and Prince George's $500m ARR, and you're left with roughly $970m of ARR for the new $2.8b contract cohort. $2.8b / $970m = ~2.9 years, which gets rounded to ~3 years. The issue is that both conclusions depend on treating “over $4b” as though it were exactly $4 billion. Once that assumption is removed, both the implied contract duration and the implied $/GPU-hour would sit higher than that. All we know for certain is that the new contracts are collectively worth approximately $2.8 billion. This means that the longer their duration, the thinner the attributable ARR becomes. There is another variable that most investors appear to be ignoring: management stated that the average contract duration across $IREN's entire portfolio is now “approximately 4 years”. Importantly, management did not limit that statement to capacity commencing this year. It was framed as applying to the entire contracted portfolio, which should therefore include the recently signed $NVDA contract, as well as $MSFT and Prince George and the newly signed $2.8 billion cohort. That allows us to work backwards and estimate how long the newest contracts would need to run for the overall portfolio average to land at ~4 years. We know the value and duration of the Microsoft and NVIDIA contracts precisely: Microsoft is worth $9.7 billion over 5 years, while NVIDIA is worth $3.4 billion over 5 years. We also know that Prince George represents $500 million of ARR, although IREN has not disclosed the average contract duration of that cohort. Fortunately, my good friend @_Sgr_A_Star recently analyzed $IREN's latest 10-Q and used the quarter-over-quarter change in remaining performance obligations to estimate that the newly commenced Prince George contracts have a duration of roughly 3 years. I independently verified the analysis and believe that is a reasonable assumption. Including the older contracts still running at Prince George, I estimate the entire Prince George portfolio carries a weighted-average remaining duration of ~2.7 years. At $500 million of ARR, that equates to ~$1.35 billion of remaining contract value. Microsoft, NVIDIA and Prince George therefore represent ~$14.45 billion of total contract value against roughly $3.12 billion of annualized revenue. That implies a weighted-average duration of approximately 4.6 years before including the newest $2.8 billion cohort. For the entire portfolio to then land at management’s stated average of approximately 4 years, the newest $2.8 billion cohort would need to average ~2.35 years in duration. That would imply ~$1.19 billion of attributable ARR ($2.8 billion / 2.35 years). Adding that to Microsoft’s & PG’s ARR results in approximately $3.62 billion of contracted ARR that counts toward this year’s year-end ARR guidance. Management stated that roughly 85% of its expected year-end ARR was already contracted. If $3.62 billion represents that contracted 85%, it would imply a total year-end ARR of ~$4.26 billion. We can then subtract the $1.93 billion attributable to Microsoft and the $500 million attributable to Prince George, leaving $1.83 billion of implied ARR for the broader 50,000-GPU fleet. Spread across 50,000 GPUs and 8,760 hours per year, that translates into ~$4.18 per GPU-hour. This is notably above the $3.60 per GPU-hour figure most people are anchoring towards. The honest caveat is that management described the 4 year weighted-average contract duration as an approximation. How much leeway they allowed themselves is impossible to know. Under the lower-yielding, roughly 3 year assumption for the latest cohort, the weighted-average duration across the portfolio comes out to ~4.22 years. Technically, that still rounds to four years, so it could reasonably be viewed as fitting management’s description. Here is why I believe the $4.18/GPU-hour and 2.35-year outcome is closer to reality than the alternative. First, as I have repeatedly noted, the $3.60/GPU-hour / 3 year estimate is derived by treating management’s “over $4 billion” year-end ARR guidance as exactly $4 billion. We already know that is not what management said. If anything, $3.60/GPU-hour should therefore be viewed as the hard floor. Another factor pointing toward a materially higher GPU-hour rate is a recent note from Needham analyst John Todaro, published after a call with $IREN's IR team shortly after the latest contract announcements. In that note, he stated that the newest cohort of contracts was priced above the NVIDIA agreement, which came in at ~$3.77/GPU-hour. We cannot know whether that was his own inference or whether management gave him an additional piece of information. But it is difficult to see how he could state that with confidence without some degree of management confirmation. Purely from a commercial perspective, it would also make little sense for the NVIDIA contract to yield more than the newer deals. The NVIDIA agreement is at least 2 years longer in duration and is backed by one of the most creditworthy counterparties in the world. Lower counterparty risk and longer duration should result in lower pricing, all else being equal. The NVIDIA contract is also significantly larger than any individual agreement within the $2.8 billion cohort, and economies of scale should place additional downward pressure on pricing. It is true that the NVIDIA contract includes managed services, while we do not know how much of the newest cohort consists of bare-metal capacity versus Mirantis-managed services. But that incremental service component should not outweigh the combined impact of materially longer duration, lower counterparty risk and greater scale. TLDR 👇 I believe management deliberately chose to sign a mix of higher-yielding, shorter-duration contracts. That would make strategic sense given that GPU-hour pricing is rising across the market and $IREN's previous contract mix was heavily skewed toward lower-yielding, long-duration agreements. Management has also made it clear that it wants a diversified customer base and a broad mix of contract structures. This latest cohort appears to fit that strategy. A roughly $4/GPU-hour rate on the newest contracts would allow the underlying GPUs to be fully paid off and even generate a modest hardware profit by the time the contracts expire approximately 2.5 years later, even after assuming gross operating expenses equal to 15% of revenue, or roughly 85% project-level EBITDA margins (previously guided). This also raises the question of why management chose to guide to “over $4 billion” rather than disclose the precise year-end ARR figure it is currently targeting. My view is that management intentionally left room to come in comfortably above the $4 billion level investors will naturally anchor to. The remaining contracts still need to be finalized, so maintaining a buffer instead of publishing the exact figure they are trending toward is prudent risk management. It reduces the chance of missing a closely watched year-end target due to timing or execution slippage. My current estimate is that year-end ARR will land between $4.1 billion and $4.3 billion. That would imply the recent cohort of contracts is yielding roughly $3.80 to $4.30 per GPU-hour. Thanks for reading, cheers! ✌️