K = (log₁₀P − 6)/10. Humanity ≈ 0.73. Working on the remaining 0.27 in this lifetime.
San Francisco, CA
Joined July 2012
- Tweets14.7K
- Following3.6K
- Followers11.2K
- Likes78.7K
Ayush Ranjan retweeted
Data center timelines aren't just a construction problem, they're a revenue problem.
Every month between groundbreaking and go-live is a month of deferred compute capacity. For hyperscalers and GPU cloud operators, that gap has real consequences.
We've spent a lot of time studying where data center timelines actually break down and what the best teams are doing to compress them without cutting corners on quality or reliability.
We broke it all down in our latest post: where delays originate, what drives them, and how to structure your next build to stay ahead of them: hubs.la/Q04CVxl70
Ayush Ranjan retweeted
🚨🚨CRUSOE IS NOW BUILDING ITS OWN MEDIUM VOLTAGE SWITCHGEAR🚨🚨⠀
Crusoe says building its own medium-voltage (MV) switchgear cuts lead times by 40%. We think owning the gear on the critical path is becoming one of the ways neoclouds get to power faster. (1/7)🧵
xitter.gq/electron_cowboy/status…
Attacking the critical path - medium voltage switchgear is one of the longest lead times in the AI infrastructure supply chain, so @CrusoeAI has vertically integrated the capability to manufacture this gear in-house through Crusoe Industries. This brings down our lead time by 40%. Crusoe is now one of only three approved manufacturers in the U.S. for this kind of ABB gas insulated switchgear.
Ayush Ranjan retweeted
“There is no moat.”
But moats are everywhere:
Ambition.
Taste.
Agency.
Curiosity.
Judgment.
Courage.
Obsession.
Persistence.
Integrity.
Patience.
This era belongs to those who refuse to be mediocre.
Ayush Ranjan retweeted
Comparing GPU financing deals to conventional ABS sectors. Part 2 of blog is out.
Ayush Ranjan retweeted
Full write up here: xitter.gq/alexsongis/status/2116…
Full blog post here: alexyangsong.com/blog/alumin…
Ayush Ranjan retweeted
Part 2 of GPU financing blog post is up!
TLDR:
• Instead of clarifying things, I muddle them up even more by introducing equine residual value curves (and why I'm cooked)
• Cost of maintenance vs. cost of the hard asset
• Pooling and diversification as a way to hedge tenant risk (e.g. SASB CMBS vs. credit card ABS)
• Moving away from GPU modularity worsens residual value risk -> but we might be going back to modular!
• Literally no one is doing the one thing that every other ABS deal requires you to do.
And thanks @chamath for the nod on Part 1 :)
A lot of ink has been spilled on GPU financing lately. I spent over a decade trading ABS on Wall St, and structured over $2B+ of debt financings as an operator (including at @tryramp). Figured I'd try to make sense of this market, so I went back to a framework I knew well: aircraft ABS and aircraft engine ABS.
My take: a GPU neocloud runs on airline economics, and the debt behind it looks a lot like aircraft engine ABS.
The parallels:
• Perishable inventory. An empty seat after the door closes is gone forever. So is an idle hour on an energized H100 rack.
• 85%+ of cost is fixed (servers, 10-year rack leases, 10-14% debt service). One more hour of compute costs pennies of power.
• Utilization is everything. Airlines break even around 80% load factor, a levered GPU cluster around 65-75%. The returns live in those last 25 percentage points.
• NVIDIA is Boeing and Airbus rolled into one. Powered shells and grid interconnects are the new airport slots.
• Wall St copied the aircraft financing playbook: take-or-pay contract plus servers in a bankruptcy-remote SPV, 70-80% LTV.
Part 1 of my blog post goes into a lot more detail:
alexyangsong.com/blog/alumin…
Ayush Ranjan retweeted
A lot of ink has been spilled on GPU financing lately. I spent over a decade trading ABS on Wall St, and structured over $2B+ of debt financings as an operator (including at @tryramp). Figured I'd try to make sense of this market, so I went back to a framework I knew well: aircraft ABS and aircraft engine ABS.
My take: a GPU neocloud runs on airline economics, and the debt behind it looks a lot like aircraft engine ABS.
The parallels:
• Perishable inventory. An empty seat after the door closes is gone forever. So is an idle hour on an energized H100 rack.
• 85%+ of cost is fixed (servers, 10-year rack leases, 10-14% debt service). One more hour of compute costs pennies of power.
• Utilization is everything. Airlines break even around 80% load factor, a levered GPU cluster around 65-75%. The returns live in those last 25 percentage points.
• NVIDIA is Boeing and Airbus rolled into one. Powered shells and grid interconnects are the new airport slots.
• Wall St copied the aircraft financing playbook: take-or-pay contract plus servers in a bankruptcy-remote SPV, 70-80% LTV.
Part 1 of my blog post goes into a lot more detail:
alexyangsong.com/blog/alumin…
Ayush Ranjan retweeted
Crusoe CEO Chase Lochmiller says the neocloud debt angst is overblown because Crusoe is seeing an "overwhelming tidal wave of demand" that is deep and broad
“I think people are very focused on the public companies, the public neoclouds like CoreWeave and Nebius. And, I don’t have the exact intimate details of all the different contracts that they’ve signed. But generally speaking, I think we are very focused on the demand side of this equation."
"And we do have a lot of insight in terms of what we’re seeing from a demand perspective, both with the very big buyers of compute, as well as sort of the long tail of demand. And what we are seeing today is an overwhelming tidal wave of demand, and it is deep and it is broad."
"And so from our perspective, not having all of the exact details of all the different contracts and all the different loan agreements et cetera I do pretty firmly believe that a lot of this angst is overblown and is a bit of an overreaction."
"I know CoreWeave has made some very good deals. The deals they have in place, they are very good deals. And the interest rates on their unsecured bonds are going up. The rate they’re paying is still whatever it is, nine, nine and a half, nine seventy-five, something like that."
"So it may cause them to grow slower because their cost of new debt may go up. But from our perspective, the existing deals that have been made and sort of the demand picture of what’s unfolding, I think they’re in a pretty good position.”
Bill Ackman says Elon Musk might have to borrow a trillion dollars to open gigawatts of data centers, keeping rates persistently high, and that neoclouds are like levered real estate developers sensitive to the cost of capital
“So, I think it is sustainable because, you know, intelligence is again an enormously valuable asset. And I think the demand therefore for the compute that enables intelligence, I think, is going to be sustained."
"And, you know, the companies that are, you know, the hyperscalers that are building out a lot of this infrastructure have the financial resources, the creditworthiness in order to access the capital."
"I worry more about the pure play neocloud companies that are a bit like highly levered real estate developers playing a theme. I think they’re much more at risk. They’re much more sensitive to the cost of capital."
"But you have, I mean, SpaceX, Elon on the earnings call gave some projections for how many gigawatts of data centers he was going to open. If he meets his objective, he might have to borrow a trillion dollars of capital."
"So, you have companies competing at sovereign levels for debt. And the result is, you know, this massive, you know, more and more supply of debt. So, I don’t see a world in which, you know, there’s a dramatic, you know, there’s a lot of demand for capital and a finite supply. So, I think that makes rates persistently high.”
Ayush Ranjan retweeted
Power, land, capital
He who shapes power, data centers, chips and models into an economically feasible endpoint, disproportionately shapes AI/SI right now.
Why?
Power is THE binding constraint in AI/SI. Few really have it.
Data centers don’t work without power. This is not about promises of power tomorrow. Or forecasted power in 2028. Power TODAY.
Chips can’t operate profitably unless installed in a functional data center. If you can install chips, it’s because you’ve solved the power problem.
Models aren’t economically viable unless run on cost effective infrastructure which requires a functional data center with low input costs and reliable management. This can’t happen without power.
AI/SI ROI doesn’t pencil unless models generate profitable outputs which necessarily needs functional data centers as the critical input. Again, power.
Ergo, he who can shape data centers, chips and models together into a tightly coordinated stack is someone who has solved power.
Solving power allows you to disproportionately shape AI/SI right now.
Ayush Ranjan retweeted
The longer you postpone the life you want, the more responsibilities accumulate, optionality shrinks, and what once felt temporary becomes your permanent life.
Ayush Ranjan retweeted
Launching RNAFoundry.com – we manufacture personalized cancer vaccines.
Ayush Ranjan retweeted
Life becomes 100x more enjoyable when you start playing it like a video game. You have free will. You can literally do anything you want. Go anywhere. Build anything. Become anyone. The only rules that are stopping you are the ones you inherited from people who also felt stuck. Drop them. Step into the game with wild self belief and borderline greedy ambition. Dream so big it feels irresponsible. The life you always wanted is not out of reach. You just finally have to decide to go collect it.
Ayush Ranjan retweeted
You can reasonably believe this will be delayed from Elon’s 2-3 timeline. But it’s increasingly unreasonable to believe this won’t happen within 5 years and even more so within 10. Google just wrote a haiku with their mini TPU datacenter in space. And are stringing two together to create coherent clusters in their next mission. Blue Origin is hiring for project sunrise. Nvidia is dedicating resources to StarMind racks and space dedicated chips. 6 months ago there was an argument to make that it was just lip service. But there are actually investments being made (outside SpaceX) and results being seen.
You may not intuitively get it because it’s space but it’s happening. And it’s much sooner than you think.
Ayush Ranjan retweeted
SpaceX is able to get compute online faster than ANYONE else. SpaceX compute per building:
• Macrohard: ~340K GPUs
• Macroharder: ~220K GPUs
• Minihard: ~220K GPUs
• Gun Range: ~220K GPUs (coming online 11/26)
• Spam: ~200k GPUs (coming online 12/16)
That’s roughly 1.2M Blackwell GPUs (~2.2 GW+) 🤯
The crazy part is Elon said he’s highly confident SpaceX can add ~10 GW of compute in 2027. Massive growth is coming next year as well.
Macroharder and Minihard both have 220k GB300s with 800 gig nics configured for training. Gun Range will have the same. Spam (used to house Hormel products) will have ~200k GB300s, but configured for inference, so much simpler networking and less than half the fiber.
I walked the whole Colossus 2 site around midnight and again this morning.
Good progress by the SpaceX team and contractors. Gun Range is tracking to be operational by end of November and Spam in December.
Ayush Ranjan retweeted
$IREN vs $CRWV: The Market Is Getting This Wrong
$CRWV's ascent has been nothing short of impressive, with exceptional revenue growth of 1,262% in a little over two years.
However, that growth came at a cost, one that has the potential to take the entire company down.
In order to achieve near-hyperscaler status, the company took on extraordinary amounts of debt. More than $35B, to be exact.
With that debt came the requirement to pay interest, which now sits at ~25% of revenue. During its growth spurt, interest obligations actually outgrew revenues. And with that heavy interest burden, the company’s net income remains deep in the red.
Bulls argue that as compute prices increase and the debt gradually gets paid down, the company can eventually turn a profit. Yet that ignores the bigger issue, which is that $CRWV's business model is inherently unsustainable.
In our new deep dive, available to X and Substack subscribers, I broke down exactly where the company’s fragility points sit and why I believe even an improvement in credit conditions is unlikely to be enough to save $CRWV from eventual bankruptcy.
I also think the market selling off neoclouds and AI data center stocks indiscriminately as a result is creating an amazing opportunity to double down on select names that are built to weather this storm.
Having thoroughly contrasted $IREN's financial situation and business strategy with that of $CRWV, all laid out in the new report, I firmly believe $IREN is one of those diamonds in the rough that the market is getting completely wrong.
Likewise, $NBIS looks to be in much better shape than $CRWV, although the extent to which that is true, and how exposed it still is to this credit environment, is something I’ll cover in a follow-up deep dive going live to subscribers next week.
As I’ve now said repeatedly throughout this selloff, know what you own!
That means building conviction by doing the necessary due diligence.
Cheers!
Ayush Ranjan retweeted
Big goals for SpaceDCs moved from 27-28 to 28-29 now. Goals remains just as big to pump stock because $SPCX will need huge funding in both debt and equity for their 2027 terrestrial buildout.
Elon is the smartest entrepreneur of our time. The secret to deriving signal from his statements is to distinguish when he is talking literally versus figuratively.
Working at Tesla has taught me how to distinguish deadlines made up to make you work hard as possible versus real deadlines.
