T2C

$MSFT: Who Third Party Delays The Neocloud news today was $MSFT's capex miss & CY outlook cut ~$175B to $170B attributed to third party GPU-as-a-service delays. Number Sense Respect Dave here, but $IREN alone cannot be responsible for 5B cut. IREN's 9.7B contract is only 1.94B a year and that only fully counts towards MSFT Capex if all tranches were expected to be billing at from Jan 1, 2026 which is not the case at all. If MSFT expected take deliveries from June to Dec, average delivery time is Aug/Sept so the billed amount is 1/4 of the 1.94B ARR in CY 2026. That's 500m Capex and the shortfall was 5B, a 10x difference. If we take delivery schedule of June-Dec then Aug/Sept is the average billing start means a CY 2026 5B Capex miss is 20B ARR. If take an average of Jan-Dec deliveries, then June is the average biling start and a 5B Capex miss is 10B ARR. Over a 5-year contract that's 50B-100B. That's bigger than $IREN, $NBIS, and Nscale's 2026 responsibilities combined. As we will see most of NScale's commitments are for 2027+. Who Is It If you have sense of numbers you will realize a insight: if $NBIS, $IREN, Nscale, all delivered on Dec 31, 2026, technically no one would miss and MSFT would have almost 0 Capex from these 3. That's obviously not the case but I'm illustrating that the Capex miss may not be attributed to a complete miss but also a combination deliveries back weighted towards end of the year. In fact, these three combined cannot make up a 5B Capex or 10B-20B ARR shortfall in 2026. This magnitude of a miss means that MSFT has more third party bare metal suppliers besides these three. I keep saying there's some sh** Neocloud we haven't heard of. Nscale Situation NScale is confirmed 6 months delayed at Glomfjord (1) but that's only 30MW max = 300m ARR = 75m 2026 Capex. Nscale scratched their UK datacenter completely but it doesn't seem any meaningful amount was scheduled for 2026 anyways. @SMASIMHO says that SIN02 is a culprit at 200MW (3) but that's only 2B ARR = 500m CY 2026 Capex for average delivery tranche Aug/Sept. We see that Nscale has some definitely misses but alone cannot doesn't make up the 5B shortfall. IREN Situation IREN has delivered H1 for 500M ARR delivered in mid August that's only ~187.5m billed in 2026 for H1. After acceptance, IREN’s fees are “non-refundable, non-cancellable” through 2031 (4). H1 will be billing and count towards MSFT Capex and networking change will be done in maintence windows.. H2 was expected Oct and a complete miss in 2026 would only cut 125m Capex from MSFT. I expect H3-H4 will be delivered in Dec. H2 is likely the biggest risk for IREN as it was midway between old and new topology and if anything new needs to be ordered then Jan delivery is possible. Not saying that's the case but calling out what's possible. Yes there's a clause that says MSFT has right to refuse H3-H4 is H2 isn't delivered but fact is that MAI is desperate for capacity right now and will take whatever is up. If they don't accept H3-H4, they cannot use it either. NBIS Situation From the site design submitted to Vineland local government, NBIS is to build out 16 blocks of BE Fuel Cells. BE Fuel Cells are likely all on site and ready to go but the main time is building out the electrical aggregation and natural gas piping. None of the 2 or 3 story platforms even started from a recent picture in Sept and that's 12/16 of the blocks. Pros is that temporary generators provide 123MW of the 300MW. On Oct 22, BE announced that the site had 100MW but only the first block was delivering power to the DC. Assuming NBIS builds out the 4 blocks without the platforms up this year, that's 123+80MW=203MW which is only a 100MW miss. That NBIS alone doesn't make a 5B capex miss either. Conclusion It's clear that combining Nscale's big misses, IREN's zero to small miss, and possible NBIS medium sized missed cannot explain 5B CY 2026 shortfall in MSFT. MSFT must have other bare metal providers outside of these 3. The signal in this is that alot of sh** Neoclouds are missing for MSFT in 2026. In 2027, we will see more misses as the added center capacity to double from 2026. Any Neocloud who can put capacity up will have negotation power because alot of sh** Neoclouds will miss. Nscale isn't even the worst and will miss a sh** load in 2027 too.