⚒️open sourcing my own TOKEN Framework (+) pre-lim $ANT numbers for everyone to do their own forecasting ⚒️ TOKEN FRAMEWORK (below) - consolidate all my thoughts from last 12m & tried to organize into a more coherent framework for how I am thinking about it (+) linking all my sources.. 3 important things imo 1. "Count Dollars not Tokens" 2. Pricing power is real but narrow *YOU have to have a view on Token Growth and Open Weights SHARE OF DOLLARS to have conviction ... On Vercel it went from about 5% to 23% in three months, led by pricier open models such as Kimi K3, not the cheapest ones. If that continues, we're in Bear: open-weight takes about 57% of 2030 API spend and frontier-lab spend only doubles. If it stalls, Base holds: about 24% open, about 10x for closed labs. That's a $700B difference in frontier-lab revenue by 2030 **I am sure the hive will find some errors -> good... send them and will update $ANT : 5 things you need to believe .. the CORE bet is #3# and #5# for the true believers - adjusted "RRR" waterfall for S-1 disclosures (this is run rate revenue and not ARR tbc) *"NRR adds" a black eye -> I would really focus on this with mgmt in IPO meetings ... really understanding the headwinds and tailwinds - my true est. SHARE COUNT (wow) - profitability bridge to help me understand moving pieces - Funding Traces - **BULL v BEAR** what you all want TOKEN MODEL: ANT_ARR_TOKEN_MODEL: Both of these are meant to encourage debate + for you to stress test your own assumptions and add to what I have here - needless to say nothing here is investment advice and you agree to that stipulation before opening any of the models - I will be tweaking and updating my own daily so can change without notice

We're open sourcing TorchNeuron. Your existing @PyTorch code runs on Trainium with one config change, and we're contributing upstream to make PyTorch more hardware-agnostic for everyone.

We are open-sourcing V1.0 of our agentic robotics framework, ASPIRE, with full real robot skill discovery and self improvement capabilities! On the day Astra arrived, our lab switched the backend model and saturated all the benchmarks to ~99% over night. And the frontier is already on the move: - Opus 5.5 slightly edges out Astra on our robot simulation bench. - Sonnet 5.5 hits the Pareto frontier of cost vs. robot capability, $12/task. We first introduced ASPIRE and /robot skill library in June, way before the big models could reliably do robotics. A rising frontier tide lifts all boats, but Astra was a 0->1 tsunami on many tasks. Now you can also feel the Physical Bitter Lesson at home like we did on that fateful day ;) links in thread

DeepSeek: "...open-sourcing the infrastructure components for Huawei's Ascend computing platform, including the TileLang high-level language compiler, computing libraries, and distributed communication libraries. All components correspond one-to-one with the previously open-source components for the NVIDIA platform. TileLang has already been validated on NVIDIA platforms and currently powers the majority of the operators used in training DeepSeek’s state-of-the-art V4 series models." Qwen says: Currently, DeepSeek uses a hybrid training strategy: NVIDIA for the massive 33T token pre-training phase, and Huawei Ascend for post-training (RLHF/SFT) and inference. However, with DeepSeek recently ordering 160,000 next-generation Ascend 950DT/960 chips, the company is building the infrastructure to move pre-training to Ascend by 2027, once the new chips support the necessary low-precision (FP4) training formats.

I am urgently sourcing pretrain data; must be deliverable immediately or near-immediately, not captured in ordinary webscrape e.g. Common Crawl, more niche/technical topics are preferred. As many tokens as possible, as high quality as possible. DM please!

The challenge isn't sourcing individual technologies, it's making them work together. We’re approaching AI infrastructure as an integrated system, bringing together power distribution, liquid cooling, compute infrastructure, and controls into a unified architecture optimized for performance, efficiency, and operational resilience. 📌 Visit us at OCP Global Summit booth #C69# to learn more. 🔗 Event details: See you at #OCPSummit2026#.👋

Everyone is open sourcing everything right now. Why? AI made software a commodity. So you don't sell the software anymore, you sell everything around it: the hosting, the support, the enterprise tier, the trust, and the community. Open source is the default now!

I'm open sourcing one of my greatest cultural contributions to Cursor's slack, in honor of it going away soon. Not sure if I should even be giving competitors access to this tech. This is :goat-huuuge: one of our most popular emojis. It conveys that someone is a huge GOAT way more than the non brolic and utterly lame 🐐. Think it's time for it to spread across the Slackiverse