Boom! $100m Series A to bring ultrafast inference for open source. We believe the demand for ultrafast inference will be many times larger than slow inference. But right now it's never made it to market at scale. We are here to change it. First racks live in Q1 2027 🚀 Massive thank you to @jamespalmer13 at @BlackbirdVC and @dkwon98 at @abstractvc for co-leading. + participation from @QEDInvestors, @645ventures, @carbideventures, @endeavor_global and @ausgulfcapital, @icehouse_ventures and bold capital partners. Final hat tip to our entire cap table who doubled down. No better feeling as a founder than getting this level of conviction from your existing backers. @maxkilberg, @adwane and @bencasnocha from @villageglobal (pre-seed) @camborumand from @Fuse_vc (seed) @jthassel from evercrest (Seed) @morganschwanke from @manaventures (pre seed) @salil from Uncorrelated Ventures (seed) @themichaelljgo from @plugandplaytc (angel / pre seed) Spicy chips are going mainstream 🌶️ xitter.cf/general_compute/status…

call me bonkers but I think the Anthropic S1 could be the neocloud bottom

Reflections on the post from @chamath 1. Every single greenfield developer that I meet is delayed. Greenfield is an industry term for brand new developed buildings. Building fast is harder than people think. 2. This is why we at @general_compute convert operating brownfield sites from commercial or crypto. Main lead times are fibre + generators + liquid cooling upgrades. We can go live in 4.5 months. All sites are already powered. We skip interconnection timelines and many local government consents. 3. The sites that this applied too are typically sub 20MW although some bigger players like Raeden Data Centers raeden.com/data-centers/ are doing 100MW sites. (crack team) 4. I think the elephant in the room is the frontier model companies revenues slowdown. The counter argument is the speed up of revenue at folks like @FireworksAI_HQ and @baseten. But with revenues 5x lower per token the question is - can markets reallocate compute while supporting the returns needed to fund the breakout? 5. Finally - in true Chamath style i'll talk my book a little - we think ultrafast compute will be undersupplied for a decade at least. There just isn't enough decode silicon coming to market. Speed is a differentiator and can increase revenue per token for the open source inference clouds. 2027 will show what users are willing to pay for Kimi 3.1 @ 20x faster speeds.

Wild if true @Kimi_Moonshot 's new model jumping above astra and fable ? Its a big boys model, a little more expensive to serve but damn. Part of me is excited for open source Part of me asks what this means for @AnthropicAI and all the infra being built for them

opensource ftw dying to play with is @ +1000 tps in Q127

Big news incoming....

congrats @tenstorrent and @PreferredNetJP 🚀 love that chips are now becoming a sovereign play, and it's going to be more and more prevalent seeing that warfare is no longer just boots on the ground, agent swarms will form part of the future of world power struggles, if they aren't already look what japan picked to fund: 3d stacked memory bandwidth and photonics between chiplets, not more flops that is the ceiling our customers keep hitting, moving state rather than doing math my main thought is cyber sec and sovereign programs become a real buyer for decode silicon inside the next 12 months xitter.cf/tenstorrent/status/211…

Petition to rename all the sram/3d-dram super fast chip partners' silicon to 'Spicy Chips 🌶️' Can we make this an industry standard for everything slow i.e non GPU / hyperscaler chips @SambaNovaAI @dMatrix_AI @cerebras @mitesh711 @juliechoi Thank you for your attention on this matter

🍸 anyone?

Q1 we will bring you both. Ultrafast inference + open source models.

🚨 @general_compute is partnering with @positron_ai. Hella excited to announce a multi-year cluster scale deployment of @positron_ai silicon alongside @nvidia to serve even more fast inference in Q1 2027. Prefill on GPUs, decode on silicon built for it. Packaged up for inference clouds to get more fast tokens - serving open source - into developers' hands. Here are four reasons we're betting on Positron. 1/ The team @trsohmers was a Thiel Fellow who taped out his first chip as a teenager. He built REX Computing before co-founding Positron. (recommend you watch Thomas on 20VC podcast that went our last month!) @mitesh711 spent seven years at Lambda as COO and a founding team member, scaling its GPU cloud to incredible heights. He is a commercial machine ❤️. We've spent a lot of time with both. They will do whatever it takes to make Positron win, and they won't stop until it does. 2/ The technology We've gone deep on their roadmap. Atlas is what we are deploying today as disaggregated compute, Titan, built on their Asimov silicon, is where we will be 10x-ing our deployments together. We believe that all 'memory heavy silicon' with the right fundamentals will be sold out for at least a decade. Because it's the only architecture that can supply fast inference at scale. Positron has built a memory-first architecture that we think is a winner. Asimov is built around around the problem without depending on scarce HBM. We think it's the right formula for faster, cheaper tokens. 3/ The investors Look at the cap table. Positron just closed an $875M Series C at a $5B valuation.@gavinsbaker and Atreides co-led, and Gavin joined the board. He's been one of the sharpest voices on AI compute for years. Alongside them: NEA, Valor, @dylan522p 's SemiAnalysis Capital, Jim Clark, QIA, Arm, Jump Trading, Hudson River Trading and Cisco. When the people who know this market best write checks this size, we pay attention. 4/ Fast inference is still a luxury - only available on open ai's ultrafast tier We want to see fast tokens make it to the opensource market. We want to see builders and founders build with fast inference and open up entirely new product categories. Positron is a key unlock for this happening. xitter.cf/general_compute/status…

🚨Prediction: all decode silicon is going to be sold out for a decade at least.. follow the 🧵.... @nvidia reportedly plans to invest in @dMatrix_AI , an inference chip company (and our partner) that on paper is a competitor. A month ago, d-Matrix announced its Raptor XPUs will plug directly into Nvidia MGX racks over NVLink Fusion. Why this matters: 1/ Companies investors thing are competitors to Nvidia become co-decode partners. GPUs handle prefill and verification. Specialized silicon like Raptor handles decode and draft models. The whole workload stays inside an Nvidia AI factory instead of going to rival GPUs or hyperscaler chips. 2/ Jensen wants a bigger pie. We’ve heard from inside Nvidia that the only thing Jensen really cares about is growing the market. That’s why Nvidia backs open source, robotics and world models, and spends thousands of engineering hours expanding the AI TAM. d-Matrix is the same playbook applied to inference. 3/ Fast decode is the unlock. Decode is memory bound, and dedicated decode silicon (or a tonne more memory in GPU racks) is the only practical path to fast inference. Fast inference creates new workloads the way broadband did after dial-up. Nobody built Netflix on a 56k modem. 4/ The real bet is memory. D-matrix's new solution - Raptor - bonds custom 3D DRAM face to face with a 4nm logic die, with no PHY in between. d-Matrix measured 0.37 pJ/bit on early silicon and is targeting 100+ TB/s per card. Its HBM4 reference point: ~18 TB/s at 2 to 3 pJ/bit. Memory bandwidth is the bottleneck for decode. By backing d-Matrix, Jensen is helping make sure memory innovation reaches the market, and that it reaches it inside NVLink. Racks land Q4 2027. The heterogeneous AI factory is coming, and Nvidia intends to be the platform it runs on. At @general_compute , this is the future we’re building for. @JensenHuang is the GOAT.

Woah, @sidsheth at @dMatrix_AI making the big moves You know the future of compute is heterogeneous when @JensenHuang starts putting down some chips Now we know why they integrated into nvidia's nvlink a few weeks back P.s you could have mentioned this at dinner last night buddy 😄

I actually need to raise 1b for spicy compute

Over 1,000 person waitlist for our #sftechweek event today Top of @a16z recommended events this week