Imagine spending billions buying GPUs to capture the AI boom, only to find out the Bitcoin miners sitting on cheap power might have the better business 🙂 The neocloud meltdown is making that possibility look a lot less ridiculous. NVIDIA just reported $89B in quarterly data center revenue, up 117% YoY. Meanwhile, Elon warned that 200+ neoclouds could face consolidation, and Gartner expects over 80% of providers to lose their independence by 2030. But if a neocloud goes bust, the GPUs don't just vanish. Someone eventually buys the hardware, restructures the debt, or puts that capacity back to work. Maybe the next big opportunity is figuring out who can actually bring paying customers to that compute. Think Airbnb for GPUs, except the supply is fragmented compute capacity and the customers are developers who don't wanna pay hyperscaler prices. That's where I5've started getting bullish on #DePIN again. > @akashnet / $AKT already offers H100s around $2.73/hr, compared with roughly $6.16/hr for CoreWeave in its pricing survey. > @AethirCloud / $ATH has ~433K GPU containers, 2.4B+ compute hours, ~$45.5M ARR and $191M in cumulative network revenue. > @ionet / solana:BZLbGTNCSFfoth2GYDtwr7e4imWzpR5jqcUuGEwr646K showed ~1,305 GPUs in a sampled October epoch, including 640 H100s. > @nosana_ai / solana:nosXBVoaCTtYdLvKY6Csb4AC8JCdQKKAaWYtx2ZMoo7 has a more direct payment mechanism where compute purchases get converted into NOS before providers are paid. > @rendernetwork / solana:rndrizKT3MK1iimdxRdWabcF7Zg7AR5T4nud4EkHBof has ~82.6M lifetime frames rendered and ~1.59M tokens burned, although its existing business is more rendering-focused than generic AI cloud. There's another second-order effect here that got me thinking about AI agents again. Agents are basically continuous consumers of inference. Every reasoning step, API call, browser action and retry costs money. If compute gets cheaper or easier to access, suddenly a lot of agent workflows that were too expensive to run might start making economic sense. > @virtuals_io / $VIRTUAL is building its own agent commerce layer where agents can hire and pay each other for work, so it's always on my watchlist if paid agent activity really starts accelerating. But maybe the most unexpected winners of this whole mess are actually Bitcoin miners. > @Hut8Corp has 949MW of contracted AI capacity worth ~$26.6B in base-term contracts. > @TeraWulfInc has a ~$19B Anthropic lease. > @CipherInc has contracted around 600MW of HPC capacity. > @IREN_Ltd, @Core_Scientific and others are also moving toward AI hosting, but their exposure to GPU financing and customer credit can be very different. GPUs get outdated every few years. Power connections, energized sites and long-term customers are a lot harder to replace. Which makes me think this whole meltdown could create 3 different rotations. TradFi starts repricing scarce power, DePIN gets a shot at aggregating fragmented compute, and AI agents benefit if inference becomes cheap enough to actually make money. Every time capital gets destroyed somewhere, value tends to migrate somewhere else. Got a gut feeling we're gonna see some of that play out here.