Topology Ventures

Bairdon on Nebius $NBIS (Outperform, $250 PT) As AI workloads transition from training to inference, Nebius has positioned itself as a full-stack provider, with strong software-attach. Our bullish thesis is based upon: 1) strong full-stack offering well positioned for inference, 2) diversifying c

Launchpad space currently feels more like tons of teams fighting over scraps while the tokenized inference space feels like a wide open lane with compounding opportunities

correct, now imagine that setup but you can plug in additional sources of inference... really just a question of creating standardization for tokenization of different buckets of capacity/timeframes, but once you have that the meta-inference token becomes much more possible and can function much li

"In 2022, months before ChatGPT and years before Cursor, Claude Code, and Codex, Gavin Uberti, Robert Wachen, and Chris Zhu dropped out of Harvard to make a bet that looked, at the time, nearly irresponsible: they would build an entire inference system—new chips, boards, interconnects, racks—from sc

An agent that acts as a real economic participant needs several capabilities at once: ✦ Identity and permissions, scoped and durable ✦ Settlement at machine speed and micro scale ✦ Private inference ✦ Confidential execution ✦ Verifiability ✦ Universal liquidity ✦ Discovery and coordination T

Ultimately the game will be to have an agent that's so primed on your context that you can provide super minimal instruction and it will build exactly what you want in a single inference turn. Your agent governs a team of agents, each of which have specialisms and their own local context and will

i’ve been trying to build a codex-flavored pi by forking it. thanks to the vienna school of agentic coding, i no longer need to: pi’s extension system can now handle everything i wanted. pi-codex-compaction uses openai’s native compaction api. idk what happens server-side or at inference time, but

Intelligence is becoming a tradeable commodity ​ It's getting tokenized, it's becoming yield-bearing, the value is accruing onchain ​ We're about to witness the largest wave of Onchain AI — The Rise of Inference Capital Markets ​ Markets that are early enough for asymmetric upside yet structured eno

korea NAND flash exports hit $2.5B in June, +44.4% MoM and +300.5% YoY per customs. storage demand for AI inference is real and it's moving fast.

“If privacy is outlawed, only outlaws will have privacy” -Phil Zimmerman How many people in the AI world care about privacy-protecting inference or understand that the only way to get high-end confidential compute today is with open weight models? https://x.com/ramez/status/2079968862670561604