bobbysecrets (@bobbysecrets) header
bobbysecrets (@bobbysecrets) profile picture

There is a map on the back of the Declaration of Independence

Göbeklitepe
Joined July 2025
Photo by bobbysecrets (@bobbysecrets)Photo by bobbysecrets (@bobbysecrets)Photo by bobbysecrets (@bobbysecrets)Photo by bobbysecrets (@bobbysecrets)
bobbysecrets retweeted
$NBIS just dropped a third of its entire Q3 revenue from asset light. $325M non-recurring. And said Asset Light is the reason $NBIS can no longer give incremental ARR. That is not purely noise this has real reasoning behind it. The rest of the delays of the earnings call aside the $NBIS asset-light model working exactly as designed and exceeding all expectations. Assetlight means a partner (or LP/customer with capital, land, and power) funds and owns the GPUs and the building. Nebius does not put the hardware on its balance sheet. Instead it licenses different layers of its stack as a service and takes the capacity to market. Here is thorough reminder on the $NBIS asset light layers I have defined prior to this as well which will help you understand how the asset light model conflated with the traditional nature of ARR: L1/ The partner brings the capital, the site, and the power. They own the physical asset and $NBIS has nothing to do with this layer. L2/ Bare metal layer. Nebius designs the racks, networking, cooling, and hardware architecture as a service. Custom racks, supply chain, validation. Partner owns it, Nebius makes it production grade. This is them lending their expertise and hardware design IP as a service. L3/ Cluster and cloud layer. Slurm via Soperator, Kubernetes, multi-tenant cloud on their Aether platform. Full software stack so the capacity looks and behaves like native $NBIS capacity. This they will licence to the client as far as I can foresee. L4/ Managed inference. Token Factory on top. High margin, production inference without the partner having to build the serving or orchestration layer. L5/ Agentic and application layer plus go to market. Nebius brings the customers, the sales motion, and the higher level software so the partner gets better utilization and pricing than pure bare metal wholesale. The partner gets owned infrastructure that is instantly sellable into the AI market. Nebius gets capacity without the CapEx, plus high-margin revenue from whichever layer/ layers the partner licenses. That is where ARR starts to break as a metric. A big one time realization on the enablement work/resources (architecture, design, platform deployment) hits in a single quarter like an initial licensing deal. The recurring piece (share of usage once the capacity is live and sold) comes later and is harder to annualize cleanly from one month. Management is already moving away from ARR for exactly this reason. One time setup revenue plus ongoing cloud revenue from the same partnership does not fit neatly into a simple run rate number. What this also means is that every client they sign up to $NBIS asset light brings a massive amount of cash into the business with no capex cost which $NBIS can use to fund their own hardware expansion. This is what makes $NBIS different. Most neoclouds are pure owners. Hyperscalers have massive balance sheets. $NBIS can own strategic sites where control matters and license the full stack (or any subset of layers) where speed and capital efficiency matter more. $NBIS Asset light is not a side experiment. It is a scalable way to grow capacity and revenue without matching every dollar of GPU spend with equity or debt. If they keep signing these, the one time pieces become a regular feature of the model and the recurring piece compounds underneath. That combination is the long term edge that can turn into billions of bulk high margin revenue that can be realized early and then be deployed for tour own expansion.
10
11
95
8,230
Recent thought: AI capex may not lift earnings. It may just be the cost of not falling behind. Miss it and you lose relevance. Pay it and you mostly preserve the status quo. Real winners power chips connect and hyperscalers keep most of their moat
1
21
Fairly certain Arkady has a crystal ball @nebiusai
14
bobbysecrets retweeted
New matrix just dropped:
491
861
192
15,453
1,504,793
bobbysecrets retweeted
One my friends sent this beautiful post explaining Karma. I thought it’s worth sharing 😊
233
6,771
267
23,103
1,067,810
Has anyone noticed Nebius never locked a mega offtake with OpenAI or Anthropic? MSFT + Meta + Reflection/Cohere/Palantir, yes. The two labs that keep signing $50–100B+ compute commits… no. Did they see the open-weights wave coming the whole time and decide those two might not actually be able to pay the later years of the stack? Curious if that’s the read or just capacity timing.
80
$NBIS Nobody is pricing this in. Nebius can do what the hyperscalers did with OpenAI and Anthropic: give top AI startups cheap compute in exchange for equity, warrants, preferred access. That is a front row seat on the next generation of labs… The same playbook that minted the cloud giants. If NBIS starts writing compute-for-equity checks, look out!
2
115
Is it just me or do politics feel less and less relevant?
10
Still wearing my cap
11
bobbysecrets retweeted
Open models continue taking share. Not just tokens, more $ now spent on open models than OpenAI. Positive for the AI infra trade. Open models taking share shift $ margin from the model layer to the infra and app layers.
Looks like today may be a record day for token volume % of open models on Vercel AI Gateway: 🟦 Open 78.4% 🟨 Closed 21.6% While spend 💲 usually tells a different story, #3 and #4 today are Moonshot AI & DeepSeek. Adding Z⁠.ai, their combined spend surpasses OpenAI (#2). (Do note that's the spend for inference of the model across providers (mostly in the US), not revenue going directly to the open weight labs.)
206
228
65
2,175
468,877
bobbysecrets retweeted
This is a topic that has been weighing heavily on my mind for over five years. I hope we can find a way for eveyone to move forward with “inquiry” as the true-north. For the best interest of the globe.
All-In Summit: Bill Gurley (0:00) Bill Gurley takes the stage at the All-In Summit (1:29) Bill's Presentation: Searching for Feynman Links to additional resources mentioned in talk and used in research: p3.institute/searching_for_f… @bgurley
159
215
57
1,430
236,740
All-In Summit: Bill Gurley (0:00) Bill Gurley takes the stage at the All-In Summit (1:29) Bill's Presentation: Searching for Feynman Links to additional resources mentioned in talk and used in research: p3.institute/searching_for_f… @bgurley
163
387
178
2,261
1,387,709
Peter Thiel has been pretty quiet…
18
$NBIS sketched this in July. Asset-light wasn’t a balance-sheet gimmick. It was a bet that inference would not live on 1 GW campuses — it would live on 20–40 MW sites you can plug into a software layer and sell as one cloud. Training still wants campus scale. Inference wants powered land, modular halls, and hardware you can move. That’s a distributed power market with a latency tax, not another Virginia mega-site. OpenAI/Anthropic shopping 20–30 MW is the tell. The market will reprice the names that already built the adapter for small sites instead of only pouring concrete. modular DCs distributed grid lower effective latency They were a step early. The demand is catching up.
OPENAI & ANTHROPIC HUNT FOR 20-30 MW AI CAPACITY IN UK, NORDICS OpenAI and Anthropic are looking at smaller compute deployments across the U.K. and Nordics, with similar discussions also taking place in the U.S., per CNBC. Anthropic has explored 20-30 MW deals in the U.K. and Nordic region, while OpenAI has been looking at opportunities in the Nordics. The move would complement their much larger multi-hundred-MW and gigawatt projects by giving both labs faster access to powered capacity that can come online sooner. These smaller clusters are particularly useful for inference, where AI workloads can be spread across multiple sites instead of relying on one massive tightly connected training facility. That shift is becoming more important as more compute moves from training models to serving them in production. JLL expects inference to overtake training as a share of data center capacity in 2027 and reach 37% of global workloads by 2030. Anthropic already has a roughly $45B deal with Nscale for about 460 MW in West Virginia, while OpenAI has committed to several multi-GW Stargate projects across the U.S. Source: CNBC
141
bobbysecrets retweeted
I’m sorry but America did not spend 250 years turning “what’s over there?” into the most powerful civilization on earth just to finally arrive at the most consequential technological frontier in generations and decide actually we’re scared of frontiers now. We ARE the fucking frontier. We cross oceans. We build railroads across continents. We split atoms. We put men on the moon. We turn sand into computers and computers into a global nervous system. Of COURSE AI is terrifying. The frontier is SUPPOSED to scare the shit out of you. That’s how you know you have, in fact, located the fucking frontier. The answer is not recklessness, but it sure as fuck isn’t paralysis either. America’s comparative advantage is an unusually high national tolerance for ambitious weirdos staring at impossible things and saying “Yeah... I think I can build it.” That trait is rarer than oil, harder to manufacture than semiconductors, and responsible for a frankly preposterous amount of the modern world. And you absolutely will not fucking regulate it out of us.
336
1,172
201
8,772
308,046
bobbysecrets retweeted
How did this country become a bunch of cowards. China is shipping stuff, optimistic about the future, producing most the worlds stuff. US is stuck in some limp ass culture where we scared of everything and focusing on fear. Be brave again, I’m tired of this.
199
159
14
1,795
39,664
Personally I stopped trusting @DarioAmodei the second he started Ozempic. He wants to be rich, sexy, and powerful. Don’t blame him. He just doesn’t care about safety. Don’t get it twisted
2
117
bobbysecrets retweeted
Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead. You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement. I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible. But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier. Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want. Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well. So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it. If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.
3,373
12,219
2,709
71,695
9,649,824
bobbysecrets retweeted
open source won’t pace
260
648
146
5,863
739,609
bobbysecrets retweeted
You better get to work and restock the Market Cap @RobinhoodCrypto
8
4
1
31
1,238