Jarron Jackson retweeted
NVIDIA: โDSX POD gets AI factories live in 14 days at prepared sites.โ
$IRENโs already-energized Sweetwater is a flagship DSX campus: ๐
When you say something without saying it. BULLISH $IREN ๐ฅ๐
What if an AI factory arrived with much of the complex integration already done?
NVIDIA DSX POD brings Vera Rubin infrastructure pre-integrated and tested in factory-built containers to prepared sitesโreducing onsite deployment from six months to as little as 14 days.
Learn how DSX POD accelerates the path to productive AI capacity: nvda.ws/3UQEnaT
Jarron Jackson retweeted
AI Drug Discovery Is Becoming a Bottleneck Trade
I get excited when an industry starts going through a real regime change. AI drug discovery (โAIDDโ) increasingly looks like one of those moments.
Two things are happening:
1/ Upstream: the frontier AI labs are piling in.
2/ Downstream: The supply chain is clearly moving.
Supply-chain checks suggest the upstream picks-and-shovels of discovery and early preclinical R&D are starting to feel the increase in experimental volume.
DNA โ protein โ assays โ sequencing โ automation โ preclinical testing
Names across that stack include $TWST , @GenScript , $ILMN , $TXG , lab-automation vendors and CROs.
$TWST expects triple-digit percentage growth in AI-enabled drug-discovery orders in FY26, and another year of triple-digit order growth in FY27.
@GenScript's AIDD business doubled YoY in 1H26. Its current platform advertises industrial-scale validation of 4,000+ designs/day, with integrated sequence-to-data workflows. Our channel checks suggest the ramp is moving even faster: roughly 8,000 designs/day currently, with a path toward ~16,000/day by YE26.
---
Why does AI drive more wet-lab demand?
1/ AI makes hypothesis generation almost free โ way more shots on goal.
The bottleneck is moving from expert-driven design to biological validation.
2/ AI models need continuous experimental feedback โ and both good and bad data are useful.
Traditionally, only the highest-conviction A+ candidates might get pushed into expensive validation. With AI, even the B/C candidates can be valuable because failed experiments generate training data.
@GenScript has said its sequence-to-binding workflow can return data in 4โ7 days, and that faster cycle times matter because AI models depend on continuous experimental feedback.
@Anthropic is a clean example. @claudeai designed 1,320 protein binders. @adaptyvbio converted those digital sequences into DNA, expressed the proteins and tested binding. Only 354 actually bound. And the 966 failures are not wasted. They are useful negative labels: what does not express, what does not bind, what has poor affinity. Those results help train the next model iteration.
3/ Wet labs are no longer just making drugs. They are making training data.
$TWST / @GenScript increasingly look like biological data foundries.
$TWST explicitly talks about generating model-ready data from AI-designed sequences. In some workflows, the customer may care less about receiving the physical protein than about getting structured experimental results back into the model.
Traditional drug discovery asks: โDoes candidate X work?โ
AI drug discovery also asks: โWhat can this experiment teach the model?โ
---
TAM of AIDD
If AI is simply a better R&D tool, the relevant spending pool is the $300โ400B of annual global pharma R&D. If AI meaningfully increases the number of viable drug programs, the opportunity is larger because it expands downstream demand for DNA synthesis, protein production, assays, and preclinical work.
Near term, we can also size demand from AI-company spending. If Anthropic reaches $80B of ARR in 2026 and spends just 1% on AIDD, that alone would imply ~$800M of annual investment.
---
Trade setup
This is a trade that could have long legs. Itโs hard to really stop working until PhaseI/II results (2028+)
It smells a lot like the bottleneck trade we just saw in semis: GPUs โ HBM โ networking โ power/cooling. In biology, It basically follows the drug discovery process downstream: AI models โ designs โ DNA/protein โ assays โ preclinical capacity.
After the upstream picks-and-shovels, animal testing could become the next bottleneck. Monkey prices are already near prior highs and CRO capacity is tight. AIDD pushing more candidates into preclinical development would only add demand.
?? But clinical trials are still the bottleneck?
This is the biggest pushback I keep coming back to. No matter how fast discovery becomes, drugs still need to go through preclinical โ Phase I โ Phase II โ Phase III โ approval. You still need patients, time and capital.
But that doesnโt mean the bottleneck trade wonโt work. More viable candidates โ especially with higher success rates โ still means more demand throughout the development process.
And who knows: clinical trials themselves may eventually be optimized by AI.
?? What breaks the trade?
Near term, the picks-and-shovels trade breaks if experimental budgets stop growing, AI-generated designs donโt translate into useful wet-lab hits, or capacity catches up too quickly.
Longer term, the thesis breaks if AI drugs look great in discovery / Phase I but fail at normal rates in Phase II/III. That is why Phase II matters so much.
?? Milestones
Late 2026โ2027: first Isomorphic-designed drugs enter human trials; more AI-native programs move into IND-enabling work / tox.
2028โ2030: clinical trial results start telling us whether AI-designed drugs actually perform better than conventional drugs.
Calling all the "bottleneck bros". :) @jukan05 @zephyr_z9 @aleabitoreddit @ParadisLabs
+++
More comprehensive analysis: robonomics.substack.com/p/aiโฆ
Jarron Jackson retweeted
Goldman's CommTech conference, San Francisco. 48 hours of @IREN_Ltd investor meetings, and the mood has shifted meaningfully.
Key takeaways:
Jarron Jackson retweeted
Replying to @ChaseLochmiller @OpenAI
GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years.
AGI has arrived. Congratulations @OpenAI team.
400K GPUs coming online next.
Jarron Jackson retweeted
$IREN: The Good, The Bad, & The Ugly
We just released our new $IREN deep dive into last week's earnings results.
I've spent the past week going through everything; the call, the deck, the 10-K and the filings around it, and I can say this with full confidence....
I haven't seen a single sell side report or X post that covers these earnings anywhere near as comprehensively as this piece does.
As some of you may already know from my recent post, the biggest takeaway from $IREN's Q2 (FYQ4) earnings was the company's emerging financing flywheel.
$IREN has quietly built a structure of GPU financing, customer prepayments, and soon data center refinancing, all working together to fund the build-out without leaning on the equity market the way the company used to.
This report goes several levels deeper than that post, breaking down $IREN's evolving financing strategy piece by piece and showing why it's turning into a genuine moat that very few competitors can replicate. It fundamentally changes the dilution math this stock has been punished over for years.
I firmly believe this is the most important development since the company committed to building its own cloud, and most investors still haven't fully grasped what it means.
The entire first section covering the financing flywheel is free to read (no paywall). If you're on the fence, start there & get a sense of the depth you can expect from the rest.
The report also covers what genuinely disappointed me this quarter, from the Q4 2027 liquid-cooled timeline and what it says about $IREN's standing with $NVDA, to the soft AI Cloud print, to an IR track record that keeps undercutting one of the strongest stories in the sector.
Throughout it all, I'm reflecting on my broader $IREN thesis at large, so this is as much an assessment of where the company stands today as it is an earnings breakdown.
We also made a dozen custom graphics for this report, because topics like the flywheel are far easier to grasp visually, with supporting imagery, than through text alone.
As always, it's written so that virtually every investor can follow it. Whether you hold $IREN, cover it, or are considering a position, I'm certain you'll derive great value from this one.
Once you've read it, I'd love to hear your feedback in the comments. Thank you very much!
Enjoy! โ๏ธ
agrippa.investments/p/the-goโฆ
Jarron Jackson retweeted
RAISE Day 1 Summary $IREN:
โฃ Horizon 1 handover confirmed 19th July
โฃ Commisioning H2
โฃ MSFT very happy with IREN so far, very hands off. Great partners.
โฃ IREN on call with NVDA basically everyday
โฃ Deals signed today are even better than 2 months ago.
โฃ Pre payments sweetspot now between 25-40 roughly. Some even offering 100%
โฃ IREN GB300 NVIDIA EXEMPLAR Cloud certified. Likely announced next week. Only THREE have this right now.
โฃ IREN selected with small number for VERA rubin for testing VERY SOON. Unofficially late EOY mass production. Likely early 2027
โฃ H5-H6, Sweetwater 1 Vera Rubin
โฃ Mirantis deal closing this month is the aim
โฃ Semi conductor shortage micron $MU seeing into 2031
h/t: @FinGigawatt
teapawt.net/LandoInvests/status/20โฆ
Jarron Jackson retweeted
New $IREN Deep Dive
Our new $IREN deep dive is finally live!
It's honestly the most comprehensive report we have ever released and something I'm firmly convinced will age like fine wine.
Even though it goes into great depth, it's written in a way that virtually every investor can understand. I purposefully went light on industry and finance jargon, and whenever I did use technical terms I made sure to explain them properly.
This time around I've also unlocked the entire first chapter for free Substack subscribers to read.
So if you're on the fence, I encourage you to read the first pages to get a sense of the depth and analytical quality you can expect from the rest of the deep dive.
I'm sure every $IREN shareholder, analyst, or investor curious about the company will derive great value from this deep dive.
I very much appreciate everyone's patience. This one took a while.
Enjoy! โ๏ธ
agrippa.investments/p/irens-โฆ
Jarron Jackson retweeted
๐๐ก๐ซ๐๐ ๐๐๐ฒ๐๐ซ๐ฌ. ๐๐ง๐ ๐๐จ๐ฆ๐ฉ๐จ๐ฎ๐ง๐๐ข๐ง๐ ๐๐๐ฏ๐๐ง๐ญ๐๐ ๐. ๐๐ก๐ ๐๐๐๐ ๐๐ก๐๐ฌ๐ข๐ฌ.
There's been a lot happening at IREN recently.
Expansion across North America, Europe and Asia-Pacific.
The NVIDIA partnership.
The Mirantis acquisition.
New GPU deployments.
New customer discussions.
A growing global footprint.
Underneath all of it is a fairly simple view of where the world is heading, and a deliberate strategy for how we position IREN within it.
That strategy is built on three layers. Together, they compound into a structural advantage that gets harder to replicate every quarter we execute.
Layer 1: Physical infrastructure. Power, land, substations, data centers, cooling. The foundation that everything else sits on.
Layer 2: Compute infrastructure. The GPUs, servers and networking that go inside those buildings. Deployed at scale. Generating revenue. Building execution track record.
Layer 3: Software and operational capability. The orchestration, deployment tooling and enterprise expertise that makes the first two layers work harder for customers, and opens the door to a broader, higher-value market over time.
Layers 1 and 2 are where the overwhelming majority of IREN's value is being created today. Layer 3 is where that advantage compounds further over time, but only because Layers 1 and 2 are built, owned and controlled at scale by IREN, not subscale nor contracted from a third party.
Think of Amazon. They didn't win e-commerce by building a great website. They won it by controlling the fulfilment infrastructure at a scale nobody else could replicate. The foundation you don't control becomes the ceiling on your business.
That is exactly how we think about IREN. The physical infrastructure - the land, the power, the substations, the data centers - is owned and controlled by us. The compute deployed into it generates the revenue and execution track record. And the software, orchestration and enterprise capability we are more methodically building on top is what turns the total product into a vertically integrated AI Cloud platform that compounds over time and deepens into a competitive moat.
AI is still early. The bottleneck is increasingly physical. And we have spent eight years building the foundations.
Jarron Jackson retweeted
There is no larger long-term strategic move than this โ NVIDIA joins forces with $IREN to build the flagship AI factory deployment for the DSX architecture
The market will continue to repeatedly reinterpret the deeper intent and long-term objectives behind the partnership between NVIDIA and IREN.
On May 7, 2026, IRENโs CEO reposted NVIDIAโs official announcement on X regarding the partnership between the two companies: NVIDIA and IREN Limited today announced a strategic partnership to accelerate the deployment of next-generation AI infrastructure.
NVIDIA announcement
nvidianews.nvidia.com/news/nโฆ
At the same time, IREN also released another announcement on its own website: IREN signs a US$3.4 billion AI cloud services agreement with NVIDIA.
IREN announcement
iren.gcs-web.com/static-fileโฆ
The two announcements, each emphasizing different aspects of the cooperation, carry extremely significant implications.
First, after careful verification, this is the first time NVIDIA has sought external compute leasing. There are three major turning points in industry development embedded in this move.
A reversal of roles: NVIDIA becomes a โmajor external compute customerโ for the first time
In the past, NVIDIAโs relationship with infrastructure companies was almost always centered around โselling hardwareโ or โborrowing hyperscaler data centers for DGX Cloud.โ But in this US$3.4 billion agreement with IREN, NVIDIA is, for the first time in its history, leasing third-party compute capacity at large scale and on a long-term basis as a customer, for use by its own AI research teams. This kind of โreverse leasingโ is unprecedented for NVIDIA in both scale and nature.
The selective external exposure of its most core secrets: this point carries the deepest implications
For a long time, NVIDIA has insisted on keeping its most critical R&D work โ chip design, driver optimization, and large-model training โ inside its self-built supercomputers such as Selene and Eos, creating a closed loop of โbuilding the shovels and mining with them itself.โ But this time, outsourcing a 60MW research workload to an external data center is highly significant. It signals that compute-chip R&D is beginning to transition toward external collaboration.
The first opening of stack management: introducing Mirantis to manage NVIDIAโs internal R&D clusters
Previously, NVIDIAโs internal cluster management was handled entirely by its own engineering teams. But under this agreement, NVIDIA is for the first time allowing third-party management, bringing in Mirantis to participate in cluster orchestration and operations. This also signals a transformation in NVIDIAโs latest compute architecture R&D approach โ beginning to โstrengthen external collaborationโ for lower-level operational work such as server cooling, restarts, and Kubernetes configuration.
As the ability of individual GPU chips to increase computing performance gradually approaches physical and engineering limits, the next phase of AI compute advancement is shifting from โsingle-chip performance competitionโ to โsystem-level scalability competition.โ This is NVIDIAโs direction of transformation.
The primary paths for the next stage of AI compute improvement include: GPU clustering, high-speed interconnects, rack-scale computing, and data-center-level coordination. This requires GPU manufacturers (NVIDIA), data center designers/builders/operators (IREN), and supercluster operating systems (Mirantis) to jointly collaborate on development.
What they are developing is precisely the NVIDIA DSX architecture referenced in the NVIDIA-IREN partnership announcement. And IRENโs hyperscale SW site in Texas is becoming the flagship deployment location for NVIDIAโs DSX architecture. This is absolutely not a simple narrative of NVIDIA investing in a company and becoming a shareholder.
For the worldโs leading company that holds the core secrets of AI compute chip R&D, this is not a trivial matter.
From NVIDIAโs perspective, there appear to be many potential partners, such as CoreWeave, Nebius, Oracle, Microsoft Azure, Amazon Web Services, and Crusoe, and NVIDIA has already invested in or partnered with these firms before. But why did it choose IREN for this most important transformation?
Because IREN possesses too many things that are uniquely its own:
Multiple GW-scale single sites with secured long-term power supply
Grid interaction capabilities
Vertical integration
Ultra-long-term site planning and abundant land supply
Green energy
Acting as its own design-and-build general contractor
Long-term accumulation of data center operational experience
Advanced design and technical capabilities
Compared with the companies above that NVIDIA has already partnered with, even if IREN temporarily lacked software capabilities, NVIDIA was still willing to wait until IREN acquired a software company before announcing this deep cooperation. Moreover, Mirantis has long been one of the three software companies that have collaborated with NVIDIA for many years. It is highly possible that NVIDIA itself played the role of connector behind IRENโs acquisition.
NVIDIA is transforming toward system-level compute scaling and building an AI factory template. In the future, the products it sells may no longer simply be GPU chips, but complete racks, clusters, or even entire AI factories.
That inevitably requires standardized data centers in order to guarantee performance, compatibility, scalability, and token efficiency.
What NVIDIA needs are facilities with massive long-term secured power supply, land, GW-scale campuses, HPC DNA, rapid construction capability, neutrality, automated scheduling capability, workload routing, GPU virtualization, fault recovery, and cluster operating systems capable of distributed training management.
At present, IREN is the only company that possesses all of these elements simultaneously.
What they are trying to build is the industrial standard for the next phase of the AI industry.
The greatest companies do not merely participate in industries โ they define the standards.
From this perspective, there is no larger strategic theme than this one.
Selling compute capacity to hyperscalers, partnering with Anthropic, or developing new sovereign AI businesses are all important, but none compare with this.
The deeper meaning of last weekโs announcement will require time for the market to fully interpret and understand. I believe I have already analyzed this trend relatively clearly.
This move by NVIDIA and IREN, once executed successfully, could once again widen the gap between the NVIDIA ecosystem and Google just as Google had begun catching up โ and it carries major implications for the entire AI industry.
Jarron Jackson retweeted
3 Imminent $IREN Catalysts
Not many companies have as much going for them right now as $IREN does.
While management has been relatively quiet since the last earnings call, I believe we're standing right before a wave of major, thesis defining announcements.
1) Australia Expansion
Given that Australia is where $IREN was incorporated, one might expect the company to already be operational there. Yet as of today, Australia remains merely the home of its HQ.
That will likely change very soon...
Just a couple of months ago, $IREN announced a sponsorship of the Sydney Swans, a prominent AFL team. As an isolated event, I wouldn't have thought much of it. The company's CEO is an Australian Football coach himself, so it could have simply been management paying homage to the company's roots.
However, this sponsorship was accompanied by a sweeping marketing campaign across Australia.
$IREN has seemingly gone all out on visual ad spend, plastering full trams with the company's logo and tagline across multiple Australian states, while also putting up new billboards outside Sydney's airport and other notables places.
Knowing how cost disciplined management is, I seriously doubt they're burning all this money on nothing. I strongly believe the company is close to unveiling a major expansion into Australia.
Currently, there are rumors that $IREN has at least two new data center sites lined up: one in South Australia, and one in New South Wales.
With how aggressive the regional ad spend has been, I'd expect any new site announcement to be accompanied by large-scale customer contracts.
If I had to speculate on who $IREN's first major customer in Australia might be, I'd wager on Anthropic, who recently announced plans to open an office in Sydney.
2) Sweetwater 1 Energization + Deal
$IREN is likely just days or weeks away from energizing its largest site to date; the massive 1.4 GW Sweetwater 1 campus.
With data center projects across the industry missing delivery timelines, largely due to an inability to secure reliable power, Sweetwater 1 stands out as a true unicorn.
Having this much grid connected power concentrated at a single site is virtually unheard of, and positions Sweetwater as one of the most valuable assets in the sector.
Successful energization will undoubtedly elevate $IREN's standing among operators industry wide, putting its execution capabilities on full display while competitors face severe delays and outright project cancellations.
Management is also aggressively hiring for the Sweetwater campus, including night shift positions, a strong signal that the company is gearing up to develop new data centers at rapid speed around the clock, 24/7.
This tells me we're likely nearing the signing of a new large-scale anchor client deal, possibly with another hyperscaler or frontier AI lab.
My expectation is that the first tranche of the Sweetwater build-out will be designed entirely for liquid-cooled Rubins, with commissioning likely sometime in H1 2027.
3) Childress Expansion
While Sweetwater is currently getting all the attention, we shouldn't overlook $IREN's first Texas campus; the 750 MW large Childress site.
So far, $IREN has contracted 40% of the site's total capacity to Microsoft, 300 MW gross across 4 tranches (Horizon 1 to 4).
That leaves 450 MW still up for grabs.
With management clearly signaling its intention to fully convert the remainder of Childress into an air-cooled AI cloud campus, the runway potential remains enormous.
Over the coming weeks, I'm expecting one of two things: either $IREN announces a new multi-hundred MW cloud contract for Childress, or management lays out a concrete plan to convert the remaining 450 MW into a large-scale cloud hub for multiple enterprise clients.
Either way, the conversion of Childress's remaining capacity is likely to begin very soon.
As with Sweetwater, the company is also actively hiring night shift HSE advisors for Childress construction, once again signaling an intent to scale development rapidly (night shift = 24/7 construction).
On a side note, I'm also expecting the successful delivery of Horizon 1 this quarter to act as a meaningful catalyst for the company's competitive standing in the market.
General Thoughts
While I've covered each of these topics in depth in previous Substack reports, I believe the time has now come for this wave of catalysts to materialize.
It's also worth pointing out that most Wall Street analysts fail to see around the corner when it comes to $IREN's cloud expansion. For the most part, they simply react to what's directly in front of them.
That means $IREN is one of the rare stocks where retail investors can front run institutional capital, getting positioned before the catalysts materialize and before Wall Street prices them in accordingly.
The irony is that over the past few weeks, retail has been doing the exact opposite: panic selling right before what I expect to be a major re-rate of the stock.
Earlier this month I also heard many investors claim that $IREN couldn't move up before new large-scale deals or other catalysts materializedโฆ
That's a very dangerous way to think.
Markets are inherently illogical. Trying to rationalize them is a mistake not only retail investors, but institutional ones too tend to make.
Last year, $IREN's share price increased by over 1,000% from its April lows, purely on the expectation of a deal being close. If you'd waited for the actual announcement, you would have entered around $70โฆ
In any case, with these 3 major catalysts in front of us, I'm very much looking forward to the weeks ahead and especially to the Q1 earnings call.
NFA, but I wouldn't be surprised if the stock cracks $100 in May.
Images S/O: @FransBakker9812, @tempocap2
Jarron Jackson retweeted
Horizon 1-4 for Microsoft coming together.
First 200MW - more underway. $IREN
Jarron Jackson retweeted
New $IREN Deep Dive: Childress Unlocked
Iโve spent the last couple of weeks writing the most important $IREN deep dive Iโve published to date.
Air-cooling at Childress is a MUCH bigger deal than the vast majority of investors and analysts realize.
Honestly, $IREN price targets across the board should be well above $100 at this point. But Wall Street missing the forest for the trees is nothing new.
Iโve extensively modelled out the companyโs near-term pipeline using conservative assumptions (below managementโs guidance), and itโs clear as day that the market isnโt properly pricing in $IREN's industry-leading earnings power.
$IREN is going to make BILLIONS of actual net income over the coming yearsโฆ not just meaningless EBITDA or top-line figures, but real profits flowing to the bottom line.
If anyone is the next hyperscaler, itโs $IREN.
Remember, real hyperscalers are actually profitableโฆ
At the same time, every investor should be aware of looming industry risks that affect all neo-clouds in the sector and evaluate how they could impact the investment thesis.
Thatโs exactly what Iโve done for all our readers.
These are the topics this new report covers:
โ Breaking down the new GPU orders + new guidance
โ Implications of air-cooling
โ Extensive pipeline modelling
โ Comprehensive analysis of the new $6b ATM
โ Risks to the investment thesis
โ + Plenty of bonus topics
This 40+ page mega deep dive covers everything $IREN investors should be aware of today.
Itโs written in a very reader-friendly way, with many graphics & embedded video clips throughout.
I chuckle when I read so-called โanalystsโ on X give their takes on $IREN after doing nothing more than surface-level analysis (at best).
Most investors have no idea where this is headingโฆ
If youโve read the new deep dive, Iโd love to hear your feedback in the comments.
Appreciate all of you, cheers! โ๏ธ
agrippa.investments/p/iren-cโฆ
Jarron Jackson retweeted
Update on our Alt Inflation Indicators:
Real time shelter continues to run quite below official stats. We show more real time shelter running just +0.2% vs 3.0% in BLS stats.
Subbing this alt shelter into CPI, we get headline inflation < 1.6% vs 2.4% official.
Alt Core CPI at 1.43% vs 2.5%.
A commodity supply disruption should be 'transitory' - and the trend is our friend when it comes to shelter disinflation.
Jarron Jackson retweeted
$IREN GB300 or VR200 for $MSFT?
"The Microsoft agreement includes a hardware substitution clause that allows Microsoft to request replacing the originally planned NVIDIA GB300 GPUs with NVIDIA VR200 GPUs, provided the request is made at least about four months before the relevant delivery date. The substitution is not automatic or unilateral; both parties must agree on revised delivery timelines, pricing adjustments, and any necessary contract changes, and the request is also subject to GPU availability and IRENโs ability to procure or exchange orders. Pricing for VR200 GPUs would be renegotiated based on cost differences, including bill-of-materials changes, with a structured good-faith negotiation process; if the parties cannot reach agreement, the contract continues under the original GB300 configuration"
Let's consider what this means:
- @IREN_Ltd designed and optimized the Horizon data center design for Vera Rubin.
- All of the capex for the buildings, substations, cabling, and the site โ is already compatible with VR200, and budgeted into the current deal economics.
- $IREN "only" runs 76K GB300 GPUs in 200MW IT load, this is very modest, or should I say undersubscribed. If we look at the thermal power of the GB300 (max TDP of 1,400W), then $IREN could have easily ran north of 130K GB300s in Horizon 1-4 in terms of 200MW IT load.
- The $MSFT contract is not a $/MW contract, but a $/GPU hour contract. If $IREN substitutes 76K GB300s for 70-72K VR200, this means the contract pricing would also be renegotiated as such.
Let's stake a look at the VR200 latest specs:
teapawt.net/mingchikuo/status/2008โฆ
"My supply chain checks indicate VR200 NVL72 will be offered in two power profiles:
Max Q and Max P.
โข Jensen referenced NVL72, rather than the previously used NVL144 naming.
โข Max Q and Max P share the same hardware design. โข Max Q GPU/rack power (TGP/TDP): ~1.8/190 (kW). โข Max P GPU/rack power (TGP/TDP): ~2.3/230 (kW). โข Both are meaningfully higher than GB300 NVL72 at 1.4/140 (kW).
โข Offering two power profiles improves deployment flexibility under data-center power constraints. It also suggests NVIDIA is increasingly factoring real-world physical infrastructure limits into product specifications."
We heard rumors about "white space" in the racks, which means that there are empty servers in the rack under the GB300 composition. This makes sense given that the "flexible rack density" just means that they add more or less servers per rack.
Did we solve the complete puzzle here?
-> $IREN designed Horizon for a fixed quantity of racks, not a fixed rack density.
-> Horizon is over-specced for liquid-cooling, power draw at the rack level, and all the components that would support a VR200 composition.
-> If Optimized for VR200, then 200MW of IT load at Max Q 190kW = ~1,053 racks.
-> 1,053 racks x 72 GPUs (NVL72) = 75.8k GPUs.
-> IREN disclosed 76K GPUs for the MSFT contract.
-> The real intention is to run VR200 Max Q at 190kW inside Horizon, and this new disclosure of the hardware substitution, shows that Microsoft has this plan as well.
What does @satyanadella have to say about the VR200?
โWe are building the worldโs most powerful AI superfactories to serve any workload, anywhere, with maximum performance and efficiency. With the addition of NVIDIA Vera Rubin GPUs, we will empower developers and organizations to create, reason and scale in entirely new ways.โ
Well, how about scaling into your own data center from a mere 140kW with GB300 servers, to ~190kW with VR200 servers?
But as investors, we care about the economics.
Here are the implications:
- No incremental infrastructure capex for VR200.
- A similar GPU count per MW IT load.
- A much higher token output per MW IT load.
- A re-negotiated pricing per GPU hour, reflecting the increased output of the VR200 (VR200 NVL72 delivers roughly 3.5ร/5ร the training/inference AI computing power of GB300 NVL72)
- Higher unlevered IRR
- Higher operating margins
Sure, the VR200 will be more expensive than the GB300, $IREN will need to source new funding for the increased capex for the chips, but since the data center won't see any meaningful additional capex, the new negotiated price of the GPU hours for the VR200 should warrant a premium based on the token output for both AI training and inference.
Not to mention, if $IREN has delivered GB300 for Horizon 1 and 2's supercluster of 100MW with 4 or 5 9's in uptime, could it not be supportive of a premium in terms of revised contracted revenue for the VR200 GPU hours for Horizon 3 and 4?
The substitute hardware agreement could really just be a price negotiation between MSFT and IREN, and everything that has been delivered at that point in time, will assist in the price negotiation, in IREN's benefit (if they delivered).
I believe IREN will have gained pricing power on their VR200 GPU hours at that point, and that's even without considering the much higher performance the chips itself can deliver, and the relative higher GPU hour pricing that should reflect, relative to the cost of the chips.
So will we see $MSFT and $IREN opting for VR200, 4 months before shipment commences?
We know that in March 2026 the first batch of GB300 will be shipped to IREN. If we consider this a quarterly happening, then MSFT will need to inform IREN this month for the second tranche, by May for the third tranche, and by August for the last tranche.
In other words, MSFT will have to decide if they opt for VR200 โ if any โ by the latest in 6 months from now.
Going back to the delivery schedule based on the post from @mingchikuo:
VR200 NVL72 rack assembly is expected to enter mass production by the end of 3Q26; factoring in yield ramp, 2H26 rack shipments are estimated at ~5,000โ7,000 units.
IREN would only need 10~15% of the global shipments of the VR200 NVL72s in 2H26. Could their preferred status, $NVDA backstop talks, and $MSFT partnership help out here?
The question then is, will $IREN and $MSFT know by May or August if they can get Horizon 3 and 4 as a 100MW VR200 supercluster?
Could $IREN return the GB300s for Horizon 1 and 2 at some point to replace them with VR200s?
I personally believe that "the $MSFT deal" as we know it, was never the real deal that it will be.
This new disclosure of the hardware substitution is making it very obvious that there is a lot more at stake here.
I think the people that have sold IREN because of the poor MSFT deal economics, will scratch their head when they see the new unlevered IRR calculations, and a positive NPV from revised GPU hour economics.
Know what you own.
Horizon was never meant to run GB300s.
Keep an eye out for $NVDA updates around VR200 deliveries, that's is where our alpha comes from.
H/T @jmonroeinvest
Key Updates on NVIDIA Vera Rubin/VR200 NVL72
(Integrating my latest supply chain checks and NVIDIA CEO Jensen Huangโs CES 2026 keynote)
1. NVIDIA has renamed the AI server VR200 NVL144 (die-based) to VR200 NVL72 (package-based). My supply chain checks indicate VR200 NVL72 will be offered in two power profiles: Max Q and Max P.
โข Jensen referenced NVL72, rather than the previously used NVL144 naming.
โข Max Q and Max P share the same hardware design.
โข Max Q GPU/rack power (TGP/TDP): ~1.8/190 (kW).
โข Max P GPU/rack power (TGP/TDP): ~2.3/230 (kW).
โข Both are meaningfully higher than GB300 NVL72 at 1.4/140 (kW).
โข Offering two power profiles improves deployment flexibility under data-center power constraints. It also suggests NVIDIA is increasingly factoring real-world physical infrastructure limits into product specifications.
2. VR200 NVL72 upgrades the GPU thermal solution for both MaxQ and MaxP.
โข Both use a micro-channel cold plate (MCCP) paired with a gold-plated lid.
โข The market has been expecting a move to a micro-channel lid (MCL) once GPU TGP reaches 2.3 kW; however, MCL mass production is unlikely before 2H27.
3. Benefiting from upgrades across the GPU, HBM, and NVLink Switch, VR200 NVL72 delivers roughly 3.5ร/5ร the training/inference AI computing power of GB300 NVL72, driving a sharp increase in rack-level power demand.
โข VR200 NVL72 upgrades power shelves to 3*3U 110 kW (6*18.3 kW PSUs), vs. the most common GB300 NVL72 configuration of 8*1U 33 kW shelves (6*5.5 kW PSUs).
โข VR200 NVL72 power shelves adopt an 3+1 redundancy design.
โข VR200 NVL72 raises the power whip rating to 100A (vs. 60A for GB300 NVL72), tightening data-center power infrastructure requirements (e.g., busway and tap-off boxes).
4. VR200 NVL72 relies even more heavily on liquid cooling.
โข Both the compute and NVSwitch trays adopt a fanless design.
โข Rack technology cooling system flow (TCS flow) nearly doubles (~+100%) vs. GB300 NVL72, benefiting spec and/or count upgrades for CDUs, manifolds, cold plates, and quick disconnects (QDs).
โข Rack airflow requirements fall by roughly 80% (in CFM) vs. GB300 NVL72.
5. VR200 NVL72 compute tray adopts a midplane for the first time, enabling a truly cableless design.
โข Key midplane specs include 44 layers (22+22), M9 CCL (EM896K3), and an approximate size of 420ร60 mm.
โข Jensen noted that this design can reduce compute-tray assembly time from roughly two hours to about five minutes (vs. GB300 NVL72).
6. Rubin CoWoS are estimated at 300โ350k wafers in 2026, with pilot production expected in early 1Q26 and mass production by late 2Q26. VR200 NVL72 rack assembly is expected to enter mass production by the end of 3Q26; factoring in yield ramp, 2H26 rack shipments are estimated at ~5,000โ7,000 units.
7. Once rack power approaches ~200 kW, 54V distribution begins to face material constraints in space (busbars/cabling) and conversion efficiency. As a result, VR200 NVL72 can be viewed as the final generation of the Oberon rack architecture. To address the rising rack-power requirements driven by continued AI compute scaling, NVIDIA is expected to transition to the next-generation Kyber rack design, which supports 800V HVDC.
Jarron Jackson retweeted
Another quarterly update completed for $IREN.
We published our Q2 update earlier today, and interest was strong enough to briefly overwhelm the site at release. If you missed the deck, itโs available here:
๐ iren.gcs-web.com/static-fileโฆ
The past few months have seen continued progress across capacity, customers, and capital. Demand remains the strongest weโve seen and, importantly, weโre building the infrastructure and capital structure required to deliver against it.
Weโre still at an early stage of our AI Cloud build-out, but have already scaled ARR under contract to more than $2.3bn. Our $3.4bn ARR target utilizes only a portion of our now 4.5GW secured power portfolio.
More detail below ๐
For $CORZ this marks a clear inflection from the October trip up. Denton is now live, scaled, and delivering capacity for $CRWV. From an infrastructure and execution standpoint, this looks far more like a temporary trip up where I'm sure they learned alot, rather than a structural problem within the company.
Looking forward to hearing more come earnings. GJ @_Adam_Sullivan1 @Core_Scientific
This sounds like a lot > CoreWeave deploys 16,000 GPUs at delayed #DataCenter in Denton, Texas - DCD buff.ly/rTYL2Wr
@PelotasN2Boca @RepzdaCanes
teapawt.net/i/status/2011549848445โฆ
$CORZ cashing out some BTC to put up capital towards CAPEX for new customer build out ๐
2
8
1,323























