SigroResearch

Tracking the $680B AI capex cycle: compute, power, and physical AI. Daily market synthesis from my own portfolio system. Not financial advice.

[SigroResearch @SigroResearch]

$NBIS reports its second quarter on Wednesday 12 August 2026, before the open. The revenue line will almost certainly be fine. Everything underneath it is set up to be worse than the first quarter, and the reason is not demand — it is that the company guides on five measures and every one of them is calculated before the cost of the assets the business consists of.

Contracted power, connected power, annualised run-rate, revenue, adjusted EBITDA margin. None of them carry depreciation. None carry interest. None carry the accretion on the convertible notes that paid for the fleet. There is no guided GAAP measure of any kind. Q2 is the quarter where the excluded costs start arriving faster than the metrics that exclude them.

https://t.co/aJQK8XJDtE

[SigroResearch @SigroResearch]

@InvestNorthwise Appreciate the model. Still watching the gap between contracted power and actual active capacity. Last disclosed active was only ~170 MW.

[SigroResearch @SigroResearch]

@capybaraReborn The cleaner way to frame it:
Nebius guides on contracted power, ARR, and adjusted EBITDA. None of those numbers include the cost of the machines.Depreciation and convertible accretion are the real P&L. Those costs start showing up before a lot of the capacity is even active.

[SigroResearch @SigroResearch]

@deepvaluedude Agree. The capital intensity is the real issue.
They’re funding a massive fleet with convertibles that accrete and customer prepayments. The guided metrics just don’t show the cost of the machines yet.

[SigroResearch @SigroResearch]

@mvcinvesting Serious investors also notice when active power is still only ~170 MW while the guidance assumes a clean path to $7–9B ARR.

[SigroResearch @SigroResearch]

I get the pushback on the Vineland framing. One site delay by itself probably isn’t enough to break the company-level guidance.
The bigger issue for me is still the gap between contracted power and actual revenue-generating power. Last disclosed active capacity was only ~170 MW. Everything else is still mostly contracted or connected on paper.
That’s the sequence risk. The costs of the machines start showing up in the P&L before a lot of that capacity is actually live and producing revenue.

[SigroResearch @SigroResearch]

@leopoldasch We’re so fucking back

[SigroResearch @SigroResearch]

The case against $SPCX SpaceX's AI segment writes itself: revenue tripled on one cloud contract, gigawatts are counted before the hardware is in service, and a company with $100 billion in cash is buying GPUs faster than it can plug them in. Most of that survives contact with the filings. One central piece of it does not, and the pieces that do turn out to be sharper than the version in circulation.

SpaceX listed in June 2026, so this is the first quarter where any of it can be checked. The 10-Q, the earnings release and the IPO prospectus between them settle nearly every question — including two the commentary has been getting backwards.

https://t.co/NFxwywo3vn

[SigroResearch @SigroResearch]

Nfp -23k vs 80k exp

[SigroResearch @SigroResearch]

The case against $SPCX SpaceX's AI segment writes itself: revenue tripled on one cloud contract, gigawatts are counted before the hardware is in service, and a company with $100 billion in cash is buying GPUs faster than it can plug them in. Most of that survives contact with the filings. One central piece of it does not, and the pieces that do turn out to be sharper than the version in circulation.

SpaceX listed in June 2026, so this is the first quarter where any of it can be checked. The 10-Q, the earnings release and the IPO prospectus between them settle nearly every question — including two the commentary has been getting backwards.

https://t.co/NFxwywo3vn

[SigroResearch @SigroResearch]

@Hedgeye Thats USDJPY not JPYUSD

[SigroResearch @SigroResearch]

@WatcherGuru Great analysis

[SigroResearch @SigroResearch]

@jaymesrosenthal @novogratz Fair. Stories still matter, but the last one got ahead of the filings. That’s what forced the de-risk.

[SigroResearch @SigroResearch]

@jonathan_mg27 Concrete partnerships like this are what actually move the long-term case. Unfiled timelines are what create the air pockets.

[SigroResearch @SigroResearch]

@jdorman81 The storytelling problem showed up clearly in June.The run-up was driven by comments that never made it into the filings.

[SigroResearch @SigroResearch]

@KyleReidhead Buying more land doesn’t prove the current 830MW is close to being leased.
The June timeline was never in the filings.

[SigroResearch @SigroResearch]

@madridraptor The June run-up was never in the filings.

[SigroResearch @SigroResearch]

$GLXY Galaxy Digital reported its second quarter on 5 August and the stock closed down 13.9% at $19.065, 44.3% below its 18 June high of $34.20. The usual reading is that management lost credibility on the call. The filings say something narrower and more useful: the disclosure that moved the stock in June was never in the filed record at all, and the disclosure that was in the filed record never changed.

That distinction matters, because it points at a different remedy.

https://t.co/gB2hktO5Bg

[Jaymes R. @jaymesrosenthal]

All you need is a 100k clip and a fat pitch.