$IREN Dispelling this most recent @theinformation + @SemiAnalysis_ FUD.
Just do a quick LLM check and you’ll see that the payments are just moving to the next quarter, and full year CAPEX is getting recategorized to OPEX by the CFO. I agree that semianalysis must have ulterior motives and/or a vendetta against IREN, since his post are extremely misleading. He no longer has any credibility left in my view:
(LLM response)
The Push-Out: CFO Amy Hood confirmed this is entirely a matter of timing. The money wasn't canceled; the spending has simply pushed out to the next quarters. Microsoft expects its CapEx to sharply accelerate again as that capacity finally comes online.
2. The Accounting Shift: Moving CapEx to OpEx
The other reason Microsoft lowered its full-year CapEx forecast (cutting its 2026 outlook from $190 billion down to $170 billion) is a structural accounting change.
•The Rule: Under standard accounting (US GAAP), if a company leases a building for a huge chunk of that building's expected lifespan, it must be classified as a Finance Lease and reported as CapEx.
•The Change: Microsoft officially extended the estimated "useful life" of its data centers from 15 years to 25 years.
•The Result: Because the buildings are now mathematically expected to last longer, the 5-year lease deals Microsoft signs with providers like IREN no longer cover a "major part" of the building's total life. Consequently, these agreements are reclassified from Finance Leases (CapEx) to Operating Leases (OpEx).
Summary
When you see headlines saying Microsoft "cut" or "lowered" CapEx, it means:
1Physical construction delays pushed billions in planned spending into future quarters.
2Paperwork reclassification shifted billions from the CapEx bucket straight into the OpEx bucket, even though the total cash flowing to data centers remains unchanged.
BREAKING NEWS🚨: According to @theinformation, Microsoft's $5B less in capex in Q3 than it had planned was in part due to delays by $IREN.
We would not be surprised by this, considering that IREN continues to come up in our conversations with neocloud customers, usually for the wrong reasons. Multi-day power outages, network upgrades, storage failures, air quality controls leading to link flaps, XIDs everywhere. The #1 worst site in the industry, according to users.
Update: I’m down about -50% on these $IREN options, dated Feb ‘27, Jun ‘27, Jan ‘28. (Options are only a very small amount of my investments, mostly to take advantage of extreme price swings since I am already properly allocated into common shares)
I have been DCA’ing mostly on my Jun ‘27 and Jan ‘28 calls as the price continues to go lower. I don’t believe I am a “loser averaging losers”, I just feel as though the current macro noise and negative catalysts will be a distant memory in 2027 when investor focus will be about $IREN $ XXB in 2027 ARR and remaining healthy secured power backlog
I just wanted to provide full transparency on how it’s been in the casino. I’m not done yet - I’m not selling, and willing to ride these down to zero. However, let’s check back on how they are doing in a few months!
Be greedy when others are fearful!
I now have a healthy amount of $IREN calls (OTM calls spread across Feb ‘27, Jun ‘27, and Jan ‘28).
I haven’t been this loaded up on calls since 2025! Take the gifts 🎁 you are given
Best post I have seen on the Florida Learjet 60 crash
Long post worth the read. I have a close friend who flies private jets. He flew Lear 60 for over 15 years and now flies a Gulfstream. This is a copy and paste of his response to me asking him if I thought Ice was the issue.
“We always fly in ridiculous cold temperatures.”
“Usually you get the bad ice when the temperatures are like 0 to negative 20. Believe it or not, there's a point where it gets too cold for ice. At negative 40, you pretty much can't get ice. Besides that, the airplane is equipped with hot wings and hot tail.”
“I flew the Lear 60 for like 15 years. It's a no joke airplane and probably the most dangerous airplane I've ever flown. It has a super critical wing. The airflow has to be perfect. When it gets into rough air and turbulence, it will easily stall. It has stalled on me at 41,000 feet just because we got into some turbulence. (On the way to Palm Beach as a matter of fact. ) we recovered and got her flying again around 30,000 ft. Scary
But here's the problem. This airplane has a terrible coffin corner. (At high altitudes) what that means is, you fly in a really narrow window where if you lose 20 knots, it stalls. And if you gain 20 knots, it goes into overspeed. The air goes supersonic over the wing and it goes into a high speed stall, or worse, a Mach tuck.
So if you stall, the normal recovery is to push the yoke forward to lower the nose and gain air speed again. At low altitude, you can push really hard and recover quickly. But in the 60... if you push hard, it will gain air speed and go into a high speed stall, it doesn't take much. It's very sensitive. The 60 has super short wings that don't make much lift. So then you high speed stall, and the computer is yelling at you... OVERSPEED! So then you pull back and enter a low speed stall. And you go back and forth trying to keep it in the air speed window.
This whole thing is pretty scary. And you should recover basically by using your finger tips using super small corrections. You can't over do it. If you try to muscle the recovery, she gets mad and will become unrecoverable. If she enters a mach tuck. You're done.”
“I taught the Lear 60 to other pilots and mentored them. I told them... this airplane has a soul and it's trying to kill you. Do not ever try to top the weather. Don't go over a thunderstorm. Go around it. If it means a half hour longer, then do it. Going over the thunderstorm will go into turbulence that you might not recover from.
These pilots went right over a thunderstorm. I think it stalled which caused them to drop down into the thunderstorm and ended up in mach tuck. They picked up so much speed that it ripped off the horizontal stabilizer. At that point... chuck Yeager can't fix it.
These same pilots did something similar a couple months ago. ATC got mad at them. They poked the bear one too many times.
This is not a beginner airplane. The Gulfstream would never do this. It has HUGE margins of error and won't stall like that. But then again... it's got HUGE wings. It's a pussy cat.
The Gulfstream is like driving a 4 wheeler. The Lear 60 is like riding a uni cycle.”
Buy low sell high seems easy until you are actually in market. This is what happens with a ‘typical’ retail investor and why it’s hard to implement these words of wisdom right now when we are clearly in a near bottom on AI DCs:
A) when stocks are falling, you believe the bleeding will never stop, so you hold off on buying because you “don’t want to catch a falling knife”
B) when stocks start to turn you aren’t sure if this is a “dead cat bounce” so you still hold off buying
C) when the stock really starts to run up, you wait till “the next dip”
D) when the dip finally comes, your mindset immediately goes back to either A or B above and either don’t buy or panic sell
E) eventually after the stock has already made a major run up, there is enough FOMO that you start to buy the stock. Typically this is when there is already consensus on the stock
F) once exuberance finally hits, you are feeling good about recent gains on the stock and you “ride your winners”, but unexpectedly it starts to turn on you when your conviction is highest.
G) you panic sell once the stock is maybe at / near your average, leaving you with little gains or even a loss.
Rinse and repeat…forever
Don’t over think it!!
**Buy these stocks now, hold for 1-2 years, make money.**
That simple!
One year from now, it will be hard to remember all the fear and FUD that currently is in the market.
Great, now that Elon is talking about neoclouds, we really are getting a new wave of chart squigglers and kid analysts coming in thinking they know and understand the space after researching it for an hour and drawing squiggles on a chart.
We might still go down on $IREN and others, but they won’t be my shares that they are buying. I will keep stacking options if the price further declines and ignore the Elon smear campaign that is in his own interest.
Those of us who have been here for a while know what we own. Hold strong!
Seems like every major run up is preceded by an extreme dip down, and followed by a sharp reversal. I’ve seen it in crypto, BTC miners, and now with AI DCs.
Makes sense because all the weak lettuce hands get shaken out.
I think we are witnessing that extreme dip and reversal right now.
Hope you didn’t give up right before Valhalla
You don’t have to hit the absolute bottom, just get in what feels like “close enough” (I think we’re there now $CIFR $IREN
Shifted funds around to take advantage of this downturn in AI DCs - they have been doing this for years, and they always bounce back.
Any excuse will do for a sharp turnaround very soon (mid terms elections complete, etc)
If the market keeps going down due to 10 yr interest rate yields going ballistic, just be sure to hold enough $BTC and $ETH for to inevitable YCC coming
phads retweeted
Today I will be educating you with some construction progress at $IREN sites in Texas.
Additionally, I will try to explain why the current strategy is the optimal one, and attempt to clarify some misunderstandings around the company and its trajectory.
$IREN while others worry about the price decline, I am making moves.
With a lot of positive catalysts likely coming, I am comfortable buying a few Jun 2027 call options ($75 -$85 range for ~$3), with the strategy that within a month or two I will be able to sell $140 covered calls at about this same price, which essentially would bring my cost for the time separated call spread to $0. If this happens, I would essentially be getting a up to $6k option (best case) going into 2027 that only requires low time exposure risk given my confidence in near term catalysts.
NFA and I realize that these can all go to zero, but great risk reward IMO at these levels.
Cleanest explanation I have seen on why $IREN made what I consider a mistake at earnings when showcasing estimated 2027 CAPEX, but not updating their ARR guidance along with it. This doesn’t mean IREN leadership was right to do this, it just helps one understand why it likely happened.
@_Sgr_A_Star is also doing a summary of each analyst’s valuation framework and how they get to their IREN valuations, which is worth checking out on TGP’s timeline
Jaja! 🤝
Personally, I think IREN is struggling operationally to meet its own deployment timelines.
To date, the company has provided three major time bound targets:
"$225M ARR by Q4 2025"
"$500M ARR by Q1 2026"
"Horizon 1 by Q2 2026"
All three were missed. Badly. So they stopped quaterly guidance and went to a year end number. They're gun shy now about providing precise timelines (and performance guidance).
Even now:
-NVIDIA capacity is vaguely targeted for “early 2027”
-Horizons 2 -4 are grouped into Q4
-All 2027 liquid cooled capacity is pushed to Q4 2027
-Even the remaining 2026 air cooled GPUs are not expected to be commissioned until Q4
Virtually every deployment is now assigned the latest possible delivery window.
Think about what that means - IREN is not currently guiding to commission a single additional GPU during Q3. Not one.
Management may be gun shy for understandable reasons, but the underlying execution problems and resulting lack of visibility are still theirs to own and resolve.
phads retweeted
@thinkymachines may not be a premier frontier lab yet but Inkling (open scale model) and Tinker (developer tool) reinforce the emphasis on data sovereignty that $nvda $iren @MirantisIT plan to champion and empowers enterprises and sovereigns to create their own proprietary data that remains private
⚡️ $IREN + Thinking Machines Lab Thesis Expanded ⚡️
After the Anthropic announcement today, my post on IREN + TML has started to get more attention. Here is an expansion on how / why I think TML might be IREN's "unnamed leading frontier AI Lab" and how their target customers and strategies overlap.
First, if you haven't heard of TML, they are a new up-and-coming frontier AI Lab, started by Mira Murati (ex OpenAI CTO). They just raised funding, and will need both inference and compute (specifically for their 975B parameter, open weight, multi-modal Inkling model - which can take in audio, visual and other inputs beyond text in multi-stream, micro-turn chunks). Open weight means that companies can fine tune for their specific custom use cases and host models on their infra [think finance, medical.. basically the AI enterprise and sovereign AI I mentioned in a previous post that IREN would go after!]
In March 2026, NVIDIA announced a 1GW Vera Rubin commitment to TML, but the language is careful. They said "deploy," not "host" or "build" (thanks @bitcoinbutcher1 for spotting this nuance and flagging). No site was named. NVIDIA doesn't operate gigawatt data centers itself, but they do have AI Factories which they power with their DSX platform, which IREN's Sweetwater is claimed to be the flagship site, which is 1GW+, can house Vera Rubins, and is available in the 2027-ish time frame referenced in the TML + NVIDIA investment announcement.
It also makes sense that TML would ask for anonymity, mainly because they are much earlier in their journey and are possibly in the middle of raising another (larger) round of funding. This feels more similar to how Prometheus was handled by IREN in the past (similar pattern).
I also see a clear target customer and strategy alignment between these companies forming here. They both are going after enterprise / sovereign AI, so I can see some future partnership and offering opportunities if TML is the anchor large tenant at Sweetwater. This would be a perfect fit IMO.
Note, this is all pure speculation at this point, but I would be thrilled if IREN announces that TML is their multi-year leading frontier AI Lab tenant.
nitter.gq/PhadsEth/status/209371…
$IREN My bet: the “unnamed AI Lab” from the Aug-27 $IREN earnings call = Thinking Machines Lab @thinkymachines (AI Lab founded by Mira Murati, previously CTO at OpenAI, who raised $12B in July 2026).
My guess is that this may start with a relatively small amount of initial capacity in early 2027 that will then likely turn TML into one of the first “large block” tenants for Sweetwater in H2 2027 if all goes well.
Similar pattern of being ‘unnamed’ like Prometheus was due to the proximity timing of recent funding. TML also has an NVIDIA relationship / backing (see article), and their existing model ‘Inkling’ has seen some initial traction, but now needs to aggressively scale up and compete in order to justify their recent high $50B valuation.
finance.yahoo.com/news/nvidi…
adding this link here for "Part 2" - nitter.gq/PhadsEth/status/209461…
⚡️ $IREN + Thinking Machines Lab Thesis Expanded ⚡️
After the Anthropic announcement today, my post on IREN + TML has started to get more attention. Here is an expansion on how / why I think TML might be IREN's "unnamed leading frontier AI Lab" and how their target customers and strategies overlap.
First, if you haven't heard of TML, they are a new up-and-coming frontier AI Lab, started by Mira Murati (ex OpenAI CTO). They just raised funding, and will need both inference and compute (specifically for their 975B parameter, open weight, multi-modal Inkling model - which can take in audio, visual and other inputs beyond text in multi-stream, micro-turn chunks). Open weight means that companies can fine tune for their specific custom use cases and host models on their infra [think finance, medical.. basically the AI enterprise and sovereign AI I mentioned in a previous post that IREN would go after!]
In March 2026, NVIDIA announced a 1GW Vera Rubin commitment to TML, but the language is careful. They said "deploy," not "host" or "build" (thanks @bitcoinbutcher1 for spotting this nuance and flagging). No site was named. NVIDIA doesn't operate gigawatt data centers itself, but they do have AI Factories which they power with their DSX platform, which IREN's Sweetwater is claimed to be the flagship site, which is 1GW+, can house Vera Rubins, and is available in the 2027-ish time frame referenced in the TML + NVIDIA investment announcement.
It also makes sense that TML would ask for anonymity, mainly because they are much earlier in their journey and are possibly in the middle of raising another (larger) round of funding. This feels more similar to how Prometheus was handled by IREN in the past (similar pattern).
I also see a clear target customer and strategy alignment between these companies forming here. They both are going after enterprise / sovereign AI, so I can see some future partnership and offering opportunities if TML is the anchor large tenant at Sweetwater. This would be a perfect fit IMO.
Note, this is all pure speculation at this point, but I would be thrilled if IREN announces that TML is their multi-year leading frontier AI Lab tenant.
nitter.gq/PhadsEth/status/209371…
⚡️ $IREN + Thinking Machines Lab Thesis Expanded ⚡️
After the Anthropic announcement today, my post on IREN + TML has started to get more attention. Here is an expansion on how / why I think TML might be IREN's "unnamed leading frontier AI Lab" and how their target customers and strategies overlap.
First, if you haven't heard of TML, they are a new up-and-coming frontier AI Lab, started by Mira Murati (ex OpenAI CTO). They just raised funding, and will need both inference and compute (specifically for their 975B parameter, open weight, multi-modal Inkling model - which can take in audio, visual and other inputs beyond text in multi-stream, micro-turn chunks). Open weight means that companies can fine tune for their specific custom use cases and host models on their infra [think finance, medical.. basically the AI enterprise and sovereign AI I mentioned in a previous post that IREN would go after!]
In March 2026, NVIDIA announced a 1GW Vera Rubin commitment to TML, but the language is careful. They said "deploy," not "host" or "build" (thanks @bitcoinbutcher1 for spotting this nuance and flagging). No site was named. NVIDIA doesn't operate gigawatt data centers itself, but they do have AI Factories which they power with their DSX platform, which IREN's Sweetwater is claimed to be the flagship site, which is 1GW+, can house Vera Rubins, and is available in the 2027-ish time frame referenced in the TML + NVIDIA investment announcement.
It also makes sense that TML would ask for anonymity, mainly because they are much earlier in their journey and are possibly in the middle of raising another (larger) round of funding. This feels more similar to how Prometheus was handled by IREN in the past (similar pattern).
I also see a clear target customer and strategy alignment between these companies forming here. They both are going after enterprise / sovereign AI, so I can see some future partnership and offering opportunities if TML is the anchor large tenant at Sweetwater. This would be a perfect fit IMO.
Note, this is all pure speculation at this point, but I would be thrilled if IREN announces that TML is their multi-year leading frontier AI Lab tenant.
nitter.gq/PhadsEth/status/209371…
$IREN My bet: the “unnamed AI Lab” from the Aug-27 $IREN earnings call = Thinking Machines Lab @thinkymachines (AI Lab founded by Mira Murati, previously CTO at OpenAI, who raised $12B in July 2026).
My guess is that this may start with a relatively small amount of initial capacity in early 2027 that will then likely turn TML into one of the first “large block” tenants for Sweetwater in H2 2027 if all goes well.
Similar pattern of being ‘unnamed’ like Prometheus was due to the proximity timing of recent funding. TML also has an NVIDIA relationship / backing (see article), and their existing model ‘Inkling’ has seen some initial traction, but now needs to aggressively scale up and compete in order to justify their recent high $50B valuation.
finance.yahoo.com/news/nvidi…
Adding a new article on my timeline for other retail investors (similar to myself) to help explain the market's reasoning for $IREN 's current $15B valuation nitter.gq/PhadsEth/status/209408…
Article for other retail investors (like me) to better understand how $IREN could possibly be valued at $15B market cap
One nugget of information buried at the end of the article I discovered during my research which I haven’t seen a lot of people on X talking about (maybe the LLM was wrong):
The $25-$30B in CAPEX does not include the liquid cooled GPUs needed for Sweetwater or Childress for 2027, just a significant portion of the data center build for liquid cooled!
(From the article)
**Here's the part I don't think most investors have priced in. On the Q4 call, when directly asked by Goldman Sachs, management confirmed the $25–30B guidance covers 2026 deployments, air-cooled 2027 capacity, and only "a significant portion" of the data center (building/infrastructure) costs for the new liquid-cooled facilities at Childress Horizon 5–6 and Sweetwater 1, but explicitly excludes the GPU capex for those new liquid-cooled facilities.