Run on a single jetson per laser. Incredible. Everyone will own their own compute.
Norway now has more than 2,000 underwater AI laser robots shooting lice off its salmon
they're made by a company called Stingray, and they already guard more than 50 million salmon and trout 24/7
the reason is sea lice, tiny parasites that latch onto salmon and feed on them
as few as 11 can kill a small fish, and one pen can hold 150,000 salmon
the usual fix is pumping the fish out and blasting them with warm water or water jets to knock the lice off
but the handling is so rough that Norway's fish health report ranks it the top cause of death in full-grown farmed salmon
so instead, Stingray puts a robot inside the pen that hunts lice one at a time, all day:
> two cameras watch every salmon that swims past
> AI checks two frames to make sure it's really a louse
> then a green laser fires from up to 2 meters away
> the salmon's scales reflect the beam, so the fish swims off fine and the louse is dead
we now live in a world where AI lasers do pest control on your dinner lmao
1
3
397
Everyone will own their own compute
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
Consensus view of ai is as a centralizing force. Might actually do the exact opposite, and give visibility to local businesses drowned out in the digital marketplace that now acts as almost the singular source of discovery for consumers today.
The pest control industry is worth $29B. Largest 2 firms (Rollins, Rentokil Terminix) account for 50% of revenue among top 100 pest control firms in US.
Market cap for both are down between 20-40% this year. One reason: people using AI chatbots for pest control reccomendations, and are being offered local options (while these global brands get hammered with bad reviews on Reddit).
Conor Moore retweeted
USDAI led a $106M senior secured financing for @CloudNexgen’s 1,728 NVIDIA B300 GPU deployment in Finland.
Silver Point Capital, a $48B credit platform, and Victory Park Capital, a Janus Henderson-affiliated alternative asset manager, are lending alongside USDAI through GLX.
Spot on. As we have more conversations with rating agencies, this analogy becomes incredibly important for guiding their underwriting heuristic - airline abs issuances often have 10-15 loans in them, which makes sense for gpu financings too given how chunky the deals are (impt to remind them of this data point because otherwise they’ll say we need 50!! obligors to go to market)
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…
1
19
1,084
If you are a non investment grade neocloud looking for debt, USDAI is open for business
Replying to @chamath
Pretty much.
As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.
SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.
That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
The dispersion of revenue potential for ai workloads makes the compute supply:demand analysis so tricky. DRW can turn a $3 gpu hour into $100 of profit with a new trading strategy. Meta can with ads, Eli Lilly can with drugs.
Compute bears point to underutilized/hoarded gpus as evidence of a bubble, but if the potential profit margin is so high, why wouldn’t you over consume gpu hours compared to your visible demand by a factor of 10?
Replying to @drwconvexity
Don, there is a missing term in your equation.
The crunch spread is not: Token Price - GPU rental price.
Rather it is: (Token Price * Token Volume) - GPU rental price.
More simply: Revenue per GPU - GPU rental price.
Continuous throughput improvements for the same exact token mean that volume growth can more than offset token price declines for the exact same model/token
i.e. Kimi K3 token throughput per GPU has recently been improving at a 37% annualized rate driven by inference engine optimizations.
GPU prices are vertical because:
1) Mix is shifting from higher margin frontier tokens to lower margin open-weight tokens. $10 in Claude revenue requires $1 of inference compute. $10 of of open-weight revenue requires $3 to $7 of inference compute given competition between inference providers leads to lower margins. Open-weight tokens taking share is positive for AI infrastructure demand given lower margins at the model layer. The reverse is also true in that frontier tokens taking share is negative for AI infrastructure demand. Open-weight models have been taking share since early summer.
2) Throughput is steadily improving for each individual token type such that GPU compute becomes more valuable even as true like for like token prices decline.
1
13
959
Conor Moore retweeted
spiritually, it feels inevitable that AGI is born in California as a kind of useless and largely uneconomic toy, and then dragged to New York, caged and put to work by men in grey tall towers that poke and prod it until it spills trillions out filtering through PDFs
Conor Moore retweeted
We’re co-hosting a conversation on the future of compute with @USDai_Official × @Alliance in NYC on 10/28.
We’re bringing together founders, traders, GPU buyers and sellers, investors, and operators to discuss the most pressing questions in compute, energy, and frontier tech.
Have a hot take? Building in the space? Join us.
luma.com/theopenfloor
With all of the focus on capex heavy businesses, VCs need to establish credit sleeves to stay relevant in competitive equity rounds
Natural evolution of things. Owning servers in rented Colo space is not differentiated. The end state is that every company will own their own gpus just like every company has a website. Data privacy is the only defense legacy businesses have against hungry ai labs
it's going to get harder to stay a neocloud. if you can't raise enough $ to keep up with buying chips and power, you'll get bought
probably not by other neoclouds though. we're seeing more neocloud acquisitions, but never by other neoclouds because there's no point
a neocloud gets almost nothing from buying another neocloud:
- for gpus, big neoclouds can buy new ones directly from nvidia. (though in a shortage, buying a struggling small cloud just for its gpus could maybe make sense depending on rising prices)
- no power. most neoclouds rent their data centers
- acquiring a neocloud means taking on the debt that they have from borrowing $ to buy gpus
that's why we see neocloud acquisitions by non-neocloud companies:
- @VoltagePark merged with @LightningAI
- @MistralAI bought @gokoyeb
- @Brookfield bought ori
- @rumblevideo bought @NorthernDataGrp
buying a neocloud makes more sense for a video site or an AI lab
when neoclouds buy, they buy software and power
The outcome of this constraint is many more small clusters. To turn non-IG credit into IG, you need diversification. That’s the alchemy that unlocks large swaths of capital. Not 1bn to one borrower, 1bn to 100 borrowers. Then you can forget the Nvidia backstops and ig wrappers, and turn GPUs into a standalone financable asset class. Standardization is what makes ai legible to Wall Street.
The parallel constraint in the power market is already pushing us towards 2-5mw deployments instead of 250mw. Both of these forces will be positive for the sustainability of the industry, as small deployments with a couple of month lead time can better match supply to actual, current demand, reducing the risk of overbuilding.
Ironically, over time we might even see these smaller deals pricing similarly to larger ig deployments, despite worse credit, as the project complexity becomes a major tradeoff if there is an abs market to sell into.
The AI buildout needs more capital than backstops can cover.
“There’s bank lenders, insurance lenders, they don’t want to take a lot of risk. You’ve got private equity, they want to take a lot of risk. You’ve got private credit, they’re in the middle.”
“What you do is you tranche risk. If there are losses, the equity will take the loss first, then the B piece, and then part of the A2. The loss has to be so severe to affect the senior tranche, which is how you can create different risk profiles.”
“We think NVIDIA has the capacity to support up to 46 gigawatts. That’s still not enough. There’s 240 gigawatts that’s gonna come in.”
“The lending market has to evolve to take risks outside of backstops. It’s really the biggest question of our time.”
As someone who is definitely on “the edge” of ai infra financing, I can tell you the days of cowboying back-to-back contracts and calling yourself a neocloud are over. If you aren’t actually differentiated, you are toast.
25
1,055
Crucible Capital founder @Melt_Dem reveals how she financed her own GPU cluster with stablecoin debt and personal credit risk because she refused to sit out the AI boom:
"We're a no-name operator. They're like, who are these two women building compute? Their perception's like, who's this crypto lady?"
"We got debt from USDAI, which is a protocol that crowdsources capital through stablecoins, and the GPUs back the loans. We also took on some personal credit risk. I think being levered to the tits is a great way to live."
"In the greatest build-out of infrastructure in human history, are you really just gonna sit on the sidelines and tweet? I'm gonna ship some bangers, and I'm gonna take some risks, and if I die, I die."
"I'm not gonna sit there and tell my children, during the AI boom I sat back and I tweeted."
@CrucibleCap
Construction is hard and speculative. Operating an installed cluster is by comparison easy and predictable.
Waiting for installation before funding is an under appreciated risk mitigant to financing GPUs
Sources: ~$18B of debt tied to an Oracle New Mexico data center slid into stressed territory, as investors grow wary of construction delays amid local pushback (Financial Times)
(Visit Techmeme dot com for the link and full context!)
8
717




















