data center revenue · X 热门 · 2026-10-08 07:32 UTC

# data center revenue — X 热门讨论 (2026-10-08 07:32 UTC)

## @athuinvests (Athu) · 10-08 03:03 · ♥60 ↻3 💬22 $IREN isn't a good long-term hold.

A lot of people I respect in this community hold this company, so I want to be honest and fair about this.

Power is one of the hardest things to get in AI right now, and despite $IREN having it, I have genuine concerns about what it takes to turn that into a long-term profitable, cash-flowing business.

This comes right after SemiAnalysis' latest review and Goldman's note warning that rising opex and a capex plan that needs more financing could pressure the stock.

My concern isn't demand. I think IREN will get capacity online and find plenty of customers. It's what it costs to get there, whether the revenue and the business behind it are strong enough to turn into real profits and cash flow, and whether shareholders keep any of it at the end.

The money:

IREN guided $25-30B of capex for FY27. They have about $14B lined up in cash, committed GPU financing and customer prepayments, and they're targeting another $8B. There's still a gap of several billion, and that's only one year.

Jensen Huang puts an AI factory at $50-60B per GW, and IREN's pipeline is 5GW+. Where is that kind of cash going to come from? Definitely not traditional places.

How they're funding it (so far):

IREN has raised about $19B over the past year, with only around $3B coming from equity. Here's where the rest came from.

- Customer prepayments. The cheapest money there is, but only top-tier customers like Microsoft write those checks. Microsoft's alone was $1.9B.

- GPU-backed debt. $3.6B against the Microsoft contract, plus $2.4B from Blue Owl for Mackenzie.

- Convertible notes. $3B in May at a 1% coupon. Cheap in cash, paid for in future shares.

- Equity. Straight share sales through the ATM.

- Stock as currency. Acquisitions like Mirantis, paid in shares.

- Next up, data center financing against sites they've kept unencumbered.

The cheap sources are limited by how many Microsofts you have. Everything beyond that costs more, either in interest or in shares.

The cost of money:

Line the deals up and one thing is clear. In this industry, the interest rate follows the customer.

When Microsoft is behind it, IREN borrows like the leaders. Its Microsoft financing came in around 5.9%, due 2031, essentially the same pricing CoreWeave got on its Meta-backed loan in March, the first GPU loan ever rated investment grade. Nebius got SOFR+2.5% in July against an investment-grade customer, with debt and customer cash flows covering over 100% of the capex. Credit where it's due, the Microsoft deal is well structured.

The difference shows up everywhere else. Mackenzie, without Microsoft behind it, cost 9% from Blue Owl on a 30-month term. IREN has to pay that back in two and a half years, while its new contracts average about four. That's debt running off faster than the revenue that pays for it.

Now, someone will say the same applies to $NBIS and $CRWV. And they're right, the rule is the same for everyone. Nebius raised $5.75B in convertibles in August, and CoreWeave pays up too when the customer isn't investment grade. The difference is how many Microsofts each one has, and how long their money lasts.

- CoreWeave has over $100B of backlog, and even its riskier loan this August, priced about the same as Mackenzie, runs to 2031. It went from ~15% loans in 2023 to investment grade this year, and has cut GPU-backed loans from 73% of its debt in 2024 to 38% this June.

- Nebius has a Meta deal worth up to ~$27B and a multi-billion Microsoft contract, plus $40B+ of further customer commitments it can borrow against.

- IREN has Microsoft ($9.7B) and NVIDIA ($3.4B), roughly $2.6B of its contracted ARR. Beyond that, it's Mackenzie-style money, and that's what most of the 5GW pipeline will need.

Same rules, very different hands. SemiAnalysis also says Nebius can charge premium prices because of its software, which means more margin to service the same debt.

IREN's cheapest money today is really Microsoft's credit rating, and debt has already gone from under $1B to roughly $7.6B in one fiscal year.

What it means for shareholders:

Share count went from 258M to 394M in about 14 months, up roughly 53%, with ~$3.5B still left on the ATM.

The $3B convertible from May could add around 41M more shares if it converts at ~$73, partly offset by a capped call.

The $625M Mirantis deal is paid in shares too. And NVIDIA's "$2.1B investment" is a right to buy shares at $70.

With the stock around $40, that money only comes in if it rallies more than 70%.

Revenue growth, but at what cost?

I do think revenue grows massively. It could go from $707M in FY26 to roughly $7.3B in FY28 and $14B in FY29, and EBITDA looks incredible. But look at what's left after paying for the hardware and the debt.

On consensus estimates, depreciation and interest swallow almost all of it. FY28 pre-tax is still slightly negative, and FY29 is negative again at about -$478M, with ~$1.7B going to interest alone.

Kinder estimates show a modest profit in FY29 and real EPS only around FY30, with free cash flow still deeply negative.

That's my whole point. If tens of billions of capex turns into single-digit net margins at best, before more dilution, what is the shareholder actually getting?

And that's the base case. It assumes everything goes their way. Every gigawatt delivered on schedule, utilization staying high, financing markets staying open on similar terms, and opex kept under control.

Any slip in execution or the cost of capital lands directly on margins that are already thin, or still negative.

Management and communication:

This is the part I find hardest to get comfortable with, and I don't think anyone should be. Communication patterns tend to repeat, so they matter as much as the numbers.

- In May, the NVIDIA partnership led the same release as a quarter where losses widened to $248M and revenue fell again. The stock jumped ~9% after hours. Five days later, IREN priced a $3B convertible. Raising on strength is normal, but it shows where the focus is.

- On the Q4 call, management said they deliberately don't tie the $25-$30B capex to a specific number of megawatts, because data center spend runs ahead of delivery. That's a lot of capital without a clear measure.

- The co-CEO RSUs vest on time, not performance. The chair's explanation was that earlier performance hurdles were cleared too quickly. That's a reason to set harder targets, not drop them.

- The 10-K says IREN owns and operates every layer of the stack, while its CCO says no managed services are live yet. Explain that, please?!

The cloud layer (software matters, and IREN is behind):

Owning GPUs in a building is infrastructure. What customers pay a premium for is the software on top, the scheduling, health checks, monitoring and automatic fault recovery.

$CRWV and $NBIS built that and are rated Platinum by SemiAnalysis, while $IREN was placed in "Not Recommended - Underperforming."

Mirantis is a smart step toward closing that gap, but today the company mostly sells bare metal, the most commoditized layer of the stack.

Reliability:

The pushback on SemiAnalysis has fair points. Outages happen everywhere, including at Nebius this August, and Mackenzie is still being commissioned, so this is mainly about Prince George. My concern is narrower.

The multi-day outages reportedly came from grid brownouts the site couldn't ride through without enough backup generation, which is the design issue SemiAnalysis raised. And customers renewing during a GPU shortage shows strong demand more than it shows quality.

What's going right:

Childress and Sweetwater look far better than the Canadian sites, and even SemiAnalysis says so. The Microsoft financing is genuinely well structured.

Microsoft, NVIDIA, Perplexity, Figure and a frontier lab are real customers. Most of the $684M Q4 loss was non-cash mining write-downs. And nobody builds gigawatts cheaply. Nebius' own CEO put 5GW at around $250B of financing.

Where I land:

For me, $IREN isn't a good long-term hold. Not because AI demand is slowing, and not because the team hasn't executed on power, but because the financing structure makes it hard for shareholders to keep the profits from all that growth, and that burden looks set to weigh on the stock for years.

I can get exposure to this theme higher up the stack with less risk to underwrite. If you hold it, I'm not here to tell you you're wrong. I just think the risks deserve as much airtime as the upside, and I'd love to hear how you're thinking about them.

If you want this theme with a lot more peace of mind, $NBIS is where I'm at 🤝 https://x.com/athuinvests/status/2108030548186660941

## @definikos (nikos) · 10-07 12:03 · ♥42 ↻7 💬3 Fluid Mechanics: Every dollar works three times https://x.com/definikos/status/2107803940158947458

## @OpulentVenture (Opulent Ventures) · 10-07 20:31 · ♥42 ↻5 💬2 APPLIED DIGITAL $APLD Q1’27 EARNINGS:

Revenue: $341.9M (Est. $132.16M) 🟢; +322% YoY Adj. EPS: -$0.01 (Est. -$0.30) 🟢 Adjusted Revenue: $300.4M Adjusted EBITDA: $64.4M Net Operating Income: $58.8M GAAP Net Loss: -$221.0M GAAP EPS: -$0.76

HPC Hosting: Revenue: $262.6M Base Rent Revenue: $65.8M Tenant Fit-Out Revenue: $183.5M Tenant Recoveries: $13.3M Segment Operating Profit: $33.4M

Data Center Hosting: Revenue: $37.8M vs. $37.9M YoY Segment Operating Profit: $13.3M Jamestown + Ellendale facilities operating at full capacity

AI / HPC Pipeline: Critical IT Load Under Lease: 1.41 GW across 5 campuses Contracted Revenue: ~$36B over initial lease terms Contracted Revenue Including Renewals: ~$86B Polaris Forge 1: 250 MW live following Oct. 1 expansion North Dakota Critical IT Load Target: 300 MW by year-end 2026 Delta Forge 2: 210 MW, 15-year lease; ~$5.2B contracted revenue Finland Power Capacity: Up to ~1 GW

Balance Sheet:

Cash + Cash Equivalents + Restricted Cash: $3.7B Total Debt: $6.4B Senior Secured Notes Issued: $1.59B at 7.0% https://x.com/OpulentVenture/status/2107931805093675434

## @allday_stocks (alldaystocks) · 10-07 20:29 · ♥31 ↻5 💬3 $APLD 𝐀𝐩𝐩𝐥𝐢𝐞𝐝 𝐃𝐢𝐠𝐢𝐭𝐚𝐥: 𝐑𝐞𝐯𝐞𝐧𝐮𝐞 𝐒𝐮𝐫𝐠𝐞𝐬 𝟑𝟐𝟐% 𝐚𝐬 𝐀𝐈 𝐃𝐚𝐭𝐚 𝐂𝐞𝐧𝐭𝐞𝐫 𝐂𝐨𝐧𝐭𝐫𝐚𝐜𝐭𝐬 𝐑𝐞𝐚𝐜𝐡 $𝟑𝟔𝐁 ⠀ 📊 𝐑𝐞𝐬𝐮𝐥𝐭𝐬 • Adj. EPS: -$0.01 (Est. -$0.30) ✅ • Revenue: $341.9M (Est. $124.57M) ✅, +322% YoY. • Adj. Revenue: $300.4M, vs. $64.2M YoY. • Adj. EBITDA: $64.4M, vs. $0.5M YoY. • Net Operating Income: $58.8M. • GAAP EPS: -$0.76, vs. -$0.07 YoY. • Net Loss: $221.0M, vs. $18.5M YoY. ⠀ 🎯 𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐎𝐮𝐭𝐥𝐨𝐨𝐤 • Contracted Capacity: ~1.41 GW across five AI data center campuses. • Base-Term Contracted Revenue: ~$36B. • Potential Contracted Revenue: ~$86B if all lease renewal options are exercised. • Polaris Forge 1: 250 MW operational as of October 1. • North Dakota Delivered Capacity: 300 MW expected by year-end 2026. • Finland Expansion: Up to ~1 GW of potential power capacity secured. ⠀ 📌 𝐊𝐞𝐲 𝐓𝐚𝐤𝐞𝐚𝐰𝐚𝐲𝐬 • Revenue surged 322% YoY, significantly exceeding consensus, while Adj. EBITDA jumped to $64.4M. • Applied Digital now holds ~$36B in contracted base-term revenue across 1.41 GW of leased capacity, potentially reaching ~$86B with all renewal options exercised. • HPC Hosting generated $262.6M in quarterly revenue, including $183.5M from tenant fit-out services and $65.8M in base rent. • The company delivered 250 MW of operational capacity at Polaris Forge 1 and targets 300 MW across its North Dakota campuses by year-end. • ChronoScale announced a planned 50 MW AI compute deployment with Microsoft using NVIDIA GB300 NVL72 systems. • Cash, including restricted cash, totaled $3.7B against $6.4B in debt, while quarterly interest expense rose to $77.4M. • Non-GAAP results exclude majority-owned ChronoScale, limiting direct comparability with consolidated results and potentially with consensus estimates. ⠀ 💬 𝐌𝐚𝐧𝐚𝐠𝐞𝐦𝐞𝐧𝐭 𝐂𝐨𝐦𝐦𝐞𝐧𝐭𝐚𝐫𝐲 “Our goal is to establish Applied Digital as the category leader in the design, construction, deployment, and operation of purpose-built AI factories.” https://x.com/allday_stocks/status/2107931438939304035

## @HunterAllen4 (THE GAP FATHER) · 10-07 12:57 · ♥30 ↻1 💬8 I’m buying this one heavy.

I have posted 3/4 times already.

Also I’m working on an affiliate link to my channel, so jump in here and let’s keep finding bangers!

There’s a part of the AI trade I think investors are still seriously underestimating.

Everyone talks about GPUs and electricity generation, but there’s a massive layer between the grid and the chip power conversion, UPS, switchgear, distribution, thermal management and liquid cooling for increasingly dense racks.

As rack power climbs, 800V architectures + liquid cooling are becoming increasingly important. More capacity doesn’t just require more electricity it requires more equipment to move, manage and remove the heat from every MW being deployed.

I’m digging into a company sitting directly in the middle of that grid-to-chip bottleneck.

What caught my attention isn’t simply the AI exposure.

It’s the combination of power + thermal + liquid cooling + services that can all benefit as data-center density increases.

And the numbers are getting hard to ignore: massive backlog, explosive order growth, accelerating revenue and expanding margins.

I’ve already called out $NRG and $BWXT before. This could be another interesting piece of the same power theme.

🔥 Guess the Rip Salad.

Drop your guess below. 👇

I’ll reveal the full setup + levels in the Gap Father swing channel.

Link below.

https://t.co/0dNvIlw1Ey https://x.com/HunterAllen4/status/2107817469574295875