Pulkit

Happy Diwali to everyone 🪔 🎇 may the divine bless each of us with good health, joyful hearts and endless prosperity. Let the light of our resolve guide Bharat to rise stronger, completely self reliant, and undefeated. Jai Hind 🇮🇳

The AI infrastructure race is becoming less about how many GPUs companies can buy and more about how efficiently available compute can reach the people who need it. Neoclouds are taking on billions in debt to expand GPU capacity: - SpaceX closed $40B for GPUs - CoreWeave is carrying $35B while power shortages, construction delays and rising energy demand are making that expansion harder. Data centers are projected to reach 945 TWh by 2030, even access to funding is becoming more selective. Meanwhile, AI workloads are changing. Open weight models now handle 70%+ of tokens on OpenRouter and developers need flexible compute across different providers rather than being tied to sold out clusters or long term reservations. Two approaches are playing out: • Debt funded new capacity: Borrow → buy GPUs → build campuses Constraint: debt cost, power queues, concentration Result: scarce on demand access • Utilize existing GPUs : Aggregate hardware already powered and connected Constraint: needs durable demand Result: faster access, less new power required. Neither removes the need for new infrastructure but making better use of existing resources can help meet demand without waiting for every new data center to come online. That's the part of @ionet's approach i always find worth following. The important signals aren't just how many GPUs join the network, but how consistently they're being used (35M+ compute hours and counting), whether customers return and how much demand translates into actual revenue ($29M total network earnings, with surplus being burned). More compute supply matters, making that supply useful and consistently accessible matters just as much.

This chart is useful the 249x gap in the same sector is extreme but the more practical signal for people who are actually in AI infrastructure is which networks have revenue that looks durable from AI workloads versus early or subsidized activity. @ionet, with 6x multiple reflects a network designed from the start for AI training and inference, aggregating GPUs from independent providers globally so teams can spin up clusters in under two minutes at 50-70% below hyperscaler rates, with native multi GPUs support rather than forcing full nodes or long reservations that’s different from general purpose marketplaces or networks that still lean heavily on rendering and other non AI jobs. When the revenue is coming from AI teams that need the capacity running and keep coming back, the multiple starts to look less like speculation and more like a function of utilization and honestly that’s the part which makes ionet different.

It's over for crypto ig 👀

Akash just voted 98% to move from Cosmos to Solana. i've been in DePIN space for a while now and maybe this is the story i've been waiting to see play out the thing that always stood out to me about @ionet, is the simplicity of the choice - no own chain, just build on @solana, that's it. Be where the DePIN users already are. And look at who's on Solana now. Helium, Render, Hivemapper, XNET, Geodnet, ionet… and Akash is walking in the door. That's not a coincidence bcuz the whole DePIN ecosystem actually lives on solana only. Actual users, real revenue and low fees. Akash took the longer road, ionet took the direct one. Either way, they're neighbors now. As someone who creates content in this space, i think the "DePIN chain" question has a pretty clear answer at this point and it's not the one most people expected 2 years ago. thedefiant.io/news/blockchai…