IO Intelligence at @ionet has crossed 1 trillion requests processed. Getting here also meant closing two launch ready products - BC8 and Quantum Agents, and shelving a mobile app. We had to choose where to put our focus. Last year, we were running Cyber Monday campaigns to make B2C subscriptions work. We ran hackathons across Asia and Eastern Europe and tried products from agent creation to training as a service. Several attempts did not find enough traction. We gradually focused on reaching developers inside the tools they already used. That meant partnerships with vibe coding platforms and becoming an inference provider in tools such as @goose_oss. On @OpenRouter, we went from roughly 500 million daily tokens in March to 20 billion. We also became a day zero launch partner for leading open model labs. Two engineers have carried much of the work on integrations, tool calling, serving costs and reliability while the workloads kept changing. Many others helped us get here. Some are still at io.net. Others have moved on. Their contributions are still part of what we build on, and this milestone belongs to them too. Congratulations to the team, past and present. Thank you to the partners and developers who supported us when progress was harder to see. Before io.net, I thought product market fit was the point where things would settle down. Now I understand how much doubt comes before it, and how a new model or price change can reopen questions you thought you had answered. Growing usage does not make those questions go away.

I have had a front row seat to this market at @ionet since 2024. One thing a GPU price index cannot tell you is whether a funded startup, cash in hand, will get told no. Or asked for 30-40% upfront and a guarantor (preferably the lead VC or top client). Compute increasingly behaves like a landlord business. Parts of AI landscape still behave like a VC growth bet. They do not want the same things. The compute supplier has financed hardware and needs a tenant who keeps paying. A fresh funding round of the customer helps, but it does not guarantee that the customer will survive the contract period. The AI players want adoption measured typically by subs and LTV on retained users. A new model or agent workflow brings traffic. Their customers experiment, change direction and switch providers. A feature that attracts users this month can become a basic expectation next month. The capacity commitment is much harder to change. That leaves the AI solution provider in the middle committing cash upfront while selling to their customers the freedom to leave. Token volume can easily grow with more compute while more cash gets locked into supplier advances. Engineering improves efficiency, competition pushes prices down, and the customer gets much of the saving. The AI solution provider still carries the risk of empty GPU hours. Two teams can run the same hardware and serving stack yet have very different economics because one gets better payment terms and has customers who commit. That is where I expect the market to move: committed enterprise demand alongside flexible API traffic, more minimum spend agreements and a premium for guaranteed capacity. Turning recurring usage into contracts that suppliers can finance could become a meaningful cost advantage. It also adds another dimension to the @silicondata charts. The hourly rate needs to be read alongside the advance, guarantees and buyer approval requirements. A higher rate paid monthly can be easier to fund than a lower rate that ties up the cash needed to grow. Getting the GPUs, paying for them and keeping them busy are three different problems. An inference business has to solve all three.

34M+ compute hours delivered. Up from 33M in July. No new data center required. Just GPUs that already existed, put to work. @ionet lets you use what's built. And pay a fraction of hyperscaler prices while doing it.

Two GPU compute DePINs Same product, Opposite shapes 🔽 ⚪ Revenue, 365 days → $IO : $8.73M → $AKT : $2.01M ⚪ Revenue per $1M mcap → IO : $123K → AKT : $9.0K ⚪ Volume as % of mcap → IO : 31.1% → AKT : 3.39% ⚪ Supply issued → AKT : 100% → IO : 49.9% ⚪ Price vs all-time high → AKT : 9.38% → IO : 2.71% ⚪ CertiK Skynet → IO : 86.62 → AKT : 84.21 Two live businesses in a sector full of ideas. Both earning, both building.

$31.6 trillion That's what PwC estimates will be spent on AI infrastructure through 2050. And you still can't get a GPU. The AI compute crisis isn't just a chip shortage. It's a coordination problem. Capacity is reserved, idle, and siloed behind procurement cycles. More chips alone won't fix that. That's why @ionet is building a different solution.

500M tokens a day in March 2026. 15B tokens a day yesterday. That is daily volume across @ionet hosted models on @OpenRouter, served on an almost same mix of H100, H200 and B300 GPUs since Day 1. Along the way, we became day zero launch partners for open source reasoning models including @Zai_org, @Kimi_Moonshot, @Alibaba_Qwen , Nemotron, @MiniMax_AI and GPT OSS. Nvidia has built something useful with the recent Nemotron 3.5 Lightning. A fast, capable and efficient open model, with weights, datasets and recipes that give teams like ours room to optimise. Yesterday, the OpenRouter provider token share showed @ionet at 64% token share for Nemotron 3.5 Lightning, the largest provider share. Effective prices per million tokens: Input: $0.0335 Output: $0.179 Our volumes and price reflect both usage and affordability. We are serving this workload profitably on an inference stack built and operated end to end in house. Our small team still runs two to three experiments every week. We track tool calling changes, frontier lab updates and what communities on vLLM and Sglang are discussing, then test what matters on our own workloads. Demand changes with time zones and days of the week. Launch excitement fades. New providers arrive and price wars start. You get used to revisiting your assumptions. With GPU rental costs rising, compute efficiency and reliable tool calling matter more. A low token price only helps if the task gets done correctly. Open models are becoming capable enough to build around and economical enough to serve at scale. There is a lot of engineering behind making both true at once. Thank you @NVIDIA and the Nemotron team, @OpenRouter and @alexatallah for allowing providers like ourselves to work seamlessly, and our engineers, who know how much work sits behind a neat row in a pricing table.