thermodynamics enjoyer @CrucibleCap
gooch island
- Tweets 46,743
- Following 1,112
- Followers 313,895
- Likes 90,141
might let a man explain data center credit to me as a little treat for him 😘
will chop it up with @MTSlive today 5 pm ET
REGULATION EVERYWHERE | NEW MICROSOFT COPILOT | COSIGN nitter.net/i/broadcasts/1nGnRBZeD…
4
2
26
9,809
the taxi situation in New York is dire. drivers know nothing about streets, traffic patterns, just following Google Maps and when i tell them where to go / turn / shortcuts they don’t speak English.
cab drivers were incredible in hindsight 😭😭😭 miss them
58
5
324
34,545
Meltem Demirors retweeted
my biggest takeaways from ClusterMax 3.0 (there are many, always a banger from @JordanNanos and team). tldr demand for small managed clusters is rising exponentially and integration with the powered shell is more important as racks scale. DM if you are building/investing/tinkering:
- managing the cluster is a huge expense and complicated (172 ways an Nvidia GPU can fail). the market is decidedly moving in two directions: 1. labs buying bare metal at massive scale (100s of MWs) with talent and resources to manage own clusters vs. 2. start ups raising $10s to $100s of millions to spend on compute that need a neocloud to manage the clusters. paradox here! even CoreWeave has become focused on long-term bare metal contracts, which are easy to fund with an IG counterparty, rather than managed clusters, which have higher margins but which the market is less keen to fund (!)
- even as heavyweights take higher % of compute deployed (self managed), managed clusters for "smaller" actors is *still growing exponentially*. we experienced this first hand when we saw the demand for our small cluster at @CrucibleCap
- serving open source is more lucrative than many realize - open source could lead to managed clusters taking on more share against frontier lab bare metal
- access to debt is the name of the game, rise in insurance that protects neoclouds and lenders against contract termination
- the Blackwell -> Rubin transition is much smoother for operators than Hopper -> Blackwell - VR adoption is ahead of schedule as a result.
- "to identify failures, we recommend monitoring dashboards. A good dashboard shows the failed component, the affected jobs, the current scheduler state, and when each check last ran." -> we use @Aravolta_X25 monitor our cluster at @CrucibleCap
- **** there is lots more to do be done on cybersecurity, this is a structural risk for the industry *****
- agentic coding is putting margin pressure on managed GPU clusters because customers that use AI tools are more willing to take bare metal or lightly managed cluster, but AI amplifies skill gaps. trust the experts if you need to !
- all documentation is still human-first, not agent-first, many systems are worse because they are not built for agents. @nirvanalabsai is building bare metal infrastructure for agents
- all neoclouds need to have an endpoints business going forward as more and more customers are looking for the simplest way to purchase vanilla open source tokens for their workflows. Coreweave's new managed inference platform announced in March is now at $100M ARR
- there is a reason that CoreWeave is in a league of its own and why its separated from hundreds, literally, of startup neoclouds. startup clouds take notes! CoreWeave has automated its lifecycle controller for provisioning, testing and monitoring - keeps stats on all failures and uses intelligence to predict faults
- integration of colo + bare metal is going to be more important: as the physical demands of compute continue increasing, in CoreWeave’s view, “how the building operates is now part of how the computer operates-they are not two individual things anymore”
-another reminder that AMD's software stack is well behind Nvidia's, with datapoints, and GB300 stukk iffers best perf/dollar despite higher pricing
ClusterMAX 3.0 is here!
ClusterMAX 3.0 debuts with a comprehensive review of the neocloud industry, covering 77 providers.
We increase our market view to cover 323 providers, up from 209 in ClusterMAX 2.0, 169 in ClusterMAX 1.0, and 124 in the original AI Neocloud Playbook and Anatomy article.
We have now interviewed well over 200 end users of neoclouds as part of this research.
We update our itemized list of criteria across 10 categories, and update our direct descriptions of our expectations for Slurm, Kubernetes, Standalone Machines, Monitoring Dashboards, and Health Checks. All of this content is live on our website. We encourage providers to use these lists when developing their offerings. We still consider these lists as an amalgamation of our experience interviewing end users, making them representative of the features that end users expect from their cloud providers.
Nebius joins CoreWeave in the Platinum tier. While CoreWeave still sets the technical bar for others to follow, Nebius is now established as a provider that consistently commands a premium pricing over others. Strong business decisions by Nebius have put them in a position to serve an entire class of neolabs at seller’s prices.
Google Cloud joins Oracle in the Gold tier. Azure moves to Silver, Fluidstack moves to Unavailable, and Crusoe drops to Bronze. Lambda, Firmus and TensorWave remain in Silver, while GMI moves up to Silver from Bronze.
Many companies drop from Silver (or Gold) to Bronze or lower. We raise the bar this round as only 19 neoclouds globally achieve a Medallion rating.
We establish a tier between Bronze and Underperforming: the Participation Ribbon tier. 15 providers join this rating, which more accurately describes our opinion that they do the bare minimum to get by.
5
2
27
7,200
hosting an event tonight. only 2 beverage options are champagne and Diet Coke.
you can just do things.
55
3
263
30,662
thx, felt this was so obvious? but clear i need to write a long form research paper so everyone can construct sequencing markets from first principles
Satoshi called Bitcoin “timechain” for a reason but most so-called crypto experts have never read a Bitcoin talk forum 🤷🏽♀️ SAD!
I completely understand what @Melt_Dem is talking about. Other people in the industry completely understand what she is talking about. It's a live interview in a noisy venue so obviously this isn't as polished. Folks need to calm down.
Agents may not necessarily perceive the passage of time because they only update when they receive an impulse of tokens from somewhere. Those tokens may contain quite literally contain "The time is now [the time]." But this is not a requirement.
The interesting thing here is the MEV stuff. What if there is some use case where you can influence an agent's next token(s) by being the highest bidder for the next token? What is the use case, I am not sure. There is something interesting there.
12
4
87
19,536
is the muse zero day true or false?
cannot tell what is and isn’t real with the AI narrative wars raging
7
1
22
18,508
continue to believe <$100M early stage funds can deliver >5x gross MOIC at 48 months -> monetize via secondaries to long term buyers —> create outlier cash on cash returns
delivering venture alpha is harder because the market is so efficient post series A sometimes even seed
Go back to March 2023 and pick an allocation:
1) OpenAI, Stripe and Databricks
2) Google, NVIDIA, Palantir
One represents three titans of private markets. The AI leader (at the time), a payments giant that indexes software and AI, and an infrastructure play.
The other holds three extremely popular public companies, with relatively low risk / higher standards for governance and financial reporting.
You'd have gotten ~11x from the public basket, versus ~6x from the private basket.
Of course, this assumes you're able to get liquidity at the most recent secondary marks. You're also facing the potential that some catastrophe wipes them out; much less likely for a public company.
Alternatively you could have waited until March 2026 to get allocation via RVI, and gotten a return of about +14% in half a year (versus +21% for the public basket).
Beware anyone selling the idea that the public should want exposure to late-stage private companies. It's an compelling narrative that will cost them their savings.
If these companies want retail investment, there's a simple solution: go public. It seems to be where the superior returns are, anyway.
16
7
124
43,175
did not realize how big Nvidia’s residual value book was - interested to see the numbers on the back half of the year as Nvidia starts scaling up its “bank of compute” approach to backstop the long tail of the buildout
As their AI spending plans begin to outrun their cash flows, tech companies have become more focused on managing their balance sheets in order to keep their investment-grade credit ratings and maintain access to the deepest pools of low-cost corporate debt.
Overall, Morgan Stanley analysts have tallied more than $3.1tn in off-balance-sheet commitments and credit support by seven hyperscalers and chipmakers.
(FT)
8
8
58
19,888
nuance - PE firms and specialist lenders active, IG borrowers will continue to get funding but more scrutiny, the long tail is drying up very quickly. extreme divide between haves and haves nots.
compute is an underwriting game, and most credit quality is bad and getting worse.
33
12
140
60,858
RT @deanwball: Financial engineering is the secret weapon of American industry. No one comes close to us in it. By and large our own elites…
65
2,050
1/ the data center buildout is driving a wave of investment in new power systems technology leveraging novel materials and software defined orchestration
excited to back Merge4 as they build solid state circuit breakers and PDUs with an AI intelligence and actuation layer
11
8
99
16,696
2/ more about Merge4: globenewswire.com/news-relea…
we’re actively working to deploy Merge4’s power stack in our own Crucible Compute operations
more on 800V DC, solid-state power and why power systems are changing in Part II of our data center series 👇
cruciblecapital.substack.com…
2
17
5,416
3/ there is no one more qualified to build this - Merge4’s CEO Anna Demeo is an industry veteran, having built, scaled, and sold multiple energy companies.
if you’d like to learn more and meet the team, join us next week in NYC - luma.com/nf9lsmgy
6
1,684
🔥 CAPITAL MARKETS MAXIMALISM 🔥
America has always been incredible at financial engineering. this is how we win.
cruciblecapital.substack.com…
its still so underrated the degree to which our intelligence supercycle was as much reliant on innovations in capital formation as it was innovations in the technology itself.
europe is losing not only because they can't built it, but primarily because they can't finance it
5
5
48
13,470
LA friends - one of our founders in the Hawthorne area wants to see a data center and get a look at the racks.
let me know if anyone has some racks he could come ogle in the next few days! can sign NDA, will not take photos
14
3
61
20,269
am i the only person who isn’t finding instinct useful or particularly helpful? would love to hear one thing you did with instinct that was mind blowing / an unlock, looking for inspiration and want to be blown away!
70
4
150
45,658