young americans protesting datacenters forget we offshored -uranium enrichment in the 90s and spent 30 years buying fuel from Russia -same story in rare earths and China controls 90% of magnet production, making it impossible to build robotics without Chinese parts -same with pharma precursors, all your drugs are from China if you don’t want compute built here i hope you like renting it from countries that want to win

nvidia's revenue breakdown by customer hyperscalers can't keep up with token demand and are struggling to secure powered shells outside of the US bullish AI native inference and neoclouds

customer concentration of neoclouds over time could be your indicator that the capex cycle is still in its early chapters. the caterpillar deal is an under appreciated note from coreweave’s earnings call. neoclouds have been working on the industrial enterprise in autos, healthcare and robotics for a while, but physical AI was still 1% or so of CRWV’s Q1 revenue backlog. this figure wasn’t shared this quarter, is surely still small but clearly going to grow w the largest global construction co signed “demand is broadening beyond frontier labs into software, industry (e.g., Caterpillar on Vera Rubin for physical AI), life sciences (Isomorphic Labs), financial services (Flow Traders, IMC), public sector (Leidos), and more” - Mike 1 year ago Microsoft was 71% of revenue (others were Nvidia, OAI, labs). this quarter the top 3 customers were 36%, 26%, and 10% of revenue that’s on 0.47GW of capacity then to 1.5GW today a long way to go to continue diversifying customers and feeding a deep backlog we are in very early days of industrial enterprises adopting and managing physical AI systems, CAT is a strong signal

we financed, colo’d, contracted and the first racks are humming at Crucible Compute DM if you: - want to operate with me and @Melt_Dem - or need future compute - or have other ideas (financing etc.) check out these racks

Nirvana is storage and egress first cloud infrastructure, fit for agents, RL and storage hungry workloads Nirvana is storage and egress first cloud infrastructure, fit for agents, RL and storage hungry workloads Nirvana is storage and egress first cloud infrastructure, fit fo

lumilens is on a sick run, congrats to early investors. scale up networking is the place to be semiconductors are now the FCF daddies after hyperscalers spent $ hundreds of B on M&A over the last decade. pressure to max tok/watt will only accelerate hardware acquisitions, networking is a top 2 bottleneck

reflecting on @CrucibleCap turning 2 next month. emerging trends in venture platforms that will accelerate as infrastructure economics change the economics of venture. what am i missing? compute — firms building or funding their own compute businesses. (Our compute biz has 8-figure run rate; Cambium built an opco/propco with capital partners to deploy portco Sambanova’s chips; nearly every large fund is doing something here to route immense capex and opex margins back to their bottom line) asset optimization — firms buying businesses or asset portfolios and optimizing them with their portcos. (GC buying a hospital; we’re looking at distressed energy assets to optimize w a portco and harvest levered depreciation.) restructuring — firms doing bankruptcy buyouts, restructurings, and recaps of legacy companies. (we did @standardnuclear with @decisivepointvc; working on a second energy buyout/restructure w another partner now) trading — firms moving into trading: prop desks, market making, and/or managing their own VaR/exposure. (we run a prop book; many HFT/MM VC arms have strategic capital offerings; hearing one large SV fund is trying to hire traders.) strategic finance — firms moving beyond talent, biz dev, media platform into strategic finance as companies need more debt: ABS, EBL, trade finance, PO finance, serviced by an in-house lead. (we’re building a syndication network and a suite of template docs/tools.) securitization / tokenization — as financing complexity grows, more firms will help companies securitize or tokenize fixed assets/liabilities to tap larger capital pools without an IPO, either partnering with a platform or building their own since it’s largely a commodified service now

The AI model pricing wars have started, according to our Token Index. With the latest OpenAI price reduction, they have joined the open-source models, while Anthropic remains the "sole rider" in the premium price category. The question is for how long?