Conor Moore

building @usdai_official

[Conor Moore @_ConorMoore]

If you are a non investment grade neocloud looking for debt, USDAI is open for business

[Elon Musk @elonmusk]

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.

[Conor Moore @_ConorMoore]

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?

[Gavin Baker @GavinSBaker]

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.

[Will Manidis @WillManidis]

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 @_ConorMoore]

Truly believe this is an education gap btwn credit markets and ai. Working to close it. Spoke with about 25 asset managers this week collectively representing couple hundred bn. They want to deploy

[First Squawk @FirstSquawk]

NVIDIA CEO HUANG: NEOCLOUD DEMAND IS ‘OFF THE CHARTS’ BUT CAPITAL IS SCARCE - CNBC

[Imran @lmrankhan]

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.

https://t.co/5Reeb8hcdp

[Conor Moore @_ConorMoore]

@flappyairplanes Congrats guys!

[Conor Moore @_ConorMoore]

With all of the focus on capex heavy businesses, VCs need to establish credit sleeves to stay relevant in competitive equity rounds

[Conor Moore @_ConorMoore]

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

[Jessie Dong @jessiedong_]

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

[Conor Moore @_ConorMoore]

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.

[SemiAnalysis @SemiAnalysis_]

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.”

[Conor Moore @_ConorMoore]

@katieruthmishra https://t.co/2rTS0AIuaP is this, I have already purchased couple hundred mil of RVI from them. Dm if you want an intro

[Conor Moore @_ConorMoore]

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.

[Conor Moore @_ConorMoore]

Some responses below:

$CHIP the newest game in town, underwriting the hottest trend: AI CAPEX. Digging into this one, I found several issues. first, there is no shortage of capital to underwrite good companies that are looking to build/operate AI data centers.

A: There is a tremendous shortage of capital for anything non investment grade (which is basically everything except for the hyperscalers). This is why the opportunity exists in the first place - banks are slow to adopt / understand new technologies, and there is no ABS market for GPU financing (https://t.co/a79nvzTR22 is the closest thing to ABS for the space, w/ sUSDai trading >20BN in volume to-date).

Their customers are small entities that struggle to raise money from other lenders, so you would expect some adverse selection in their loan book.

A: Maybe more true early on, but https://t.co/a79nvzTR22 customers today are typically well capitalized and often publicly traded. Its also important to note that the majority of the credit support comes from the contracted offtake, rather than the operator.

Since they're still in the early innings of deploying capital and collecting fees, I would be cautious about their ability to avoid liquidations. Another interesting finding: Crucible Capital pays the lowest rate in their loan book (so you could view it as the safest borrower). To me, that fact raises questions about the rest of their loan book.
Moreover, their current loan book is highly concentrated, so you should be pretty sure that those large borrowers will be able to keep operating despite some frothiness in web2 data center names (Oracle). Their revenue is mostly transaction-based (one-off, as borrowers pay a hefty upfront fee to set up a credit line), while recurring credit revenue is smaller. So, to keep grinding revenue higher, they need to keep finding new attractive customers on a regular basis (a pressure the other companies discussed above don't face).

A: Don't need to find new customers, need to find new capital sources (ie, sell on loans to perm capital to recycle).

The lack of a track record through a default cycle is another risk, and only time will tell whether they're doing it right. They don't have an edge in origination speed (60 days for the second cohort despite the 30 days claimed on their site), and their own report shows they struggle to close big checks (avg size: approved ~$16M, expired ~$61M, probably found better terms elsewhere, pending ~$31M).

A: 60 days to originate and close is at least 2x if not 3x faster than bank origination process. Closed 129M loan last week. 98m, 55m as two prior.

The last nuance: they state that collateral should not be retail devices, yet you can find small checks backed by RTX 5090s.

A: Was early loan, already repaid

Not a big deal in terms of loan book size, but it says something about how they operate. I'm not an AI bear in any sense, but I would be cautious about allocating money to the hottest game in town, where the room for adverse selection is huge. On the bright side, their product is pioneering a new source of yield with a competitive APY and a big TAM, and if managed properly, it could be an interesting opportunity

[MTS @MTSlive]

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

[Conor Moore @_ConorMoore]

USDAI fights for the compute middle class

[Conor Moore @_ConorMoore]

@richardchen39 100M ARR -> 1M gross profit

many such cases

[Conor Moore @_ConorMoore]

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

[Techmeme @Techmeme]

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!)

[Conor Moore @_ConorMoore]

compute twitter has finally come to the realization that "everything is neocloud" as the industry converges on owning gpus as the optimal use of funds

the next coreweave is no coreweave.

the new american dream is to own your own compute.

[Conor Moore @_ConorMoore]

i spend one half of my day listening to lenders say GPUs will be worthless in 24 months

and the other half listening to neo labs say GPUs will cost 10x more in 24 months

a truly incredible opportunity to be in the middle

[Conor Moore @_ConorMoore]

@pauljlei @USDai_Official It is a deposit into sUSDai