we got early access and hipkittens is now running on amd helios mi455! here we share an educational gemm ladder to demonstrate how to use the new helios features. overall, we found that the patterns identified in hipkittens for writing performant mi350/355 kernels translate well to helios making the forward port quite smooth!

TogetherLink, coming soon.

Today, Long Lake completed our $6.3B acquisition of Amex GBT. Long Lake acquires and transforms generational businesses with AI across the American services economy. Amex GBT is the travel partner for 17,500 businesses in 140 countries. Last year, Amex GBT booked 35 million trips for 10 million travelers. This marks Long Lake’s 40th acquisition, and our family of companies now employs nearly 30,000 people. We founded Long Lake three years ago with the thesis that AI is going to change every company, but there’s a large overhang between AI capabilities and how most businesses leverage AI tools. We partner with strong, profitable, and growing businesses to help them deploy AI and improve their customer service. Our first cohort of companies has doubled their EBITDA in less than two years through topline and productivity growth, all while increasing headcount. (We’re proud to say that we’ve never conducted a layoff.) We’re excited to share a bit more about what we’ve been up to.

Open collaboration is moving GPU kernel development forward. AMD and contributors from @Caltech, @Stanford, @togethercompute and the broader developer community teamed up to bring HipKittens support to AMD Instinct MI455X GPUs. The collaboration produced peak-performance GEMM kernels for the Helios GPU architecture using HipKittens. The walkthrough starts with a baseline kernel and progresses through optimizations including asynchronous memory transfers, Tensor Data Movers (TDM) and workgroup-cluster multicast. Developers can follow the code and see how each technique shapes kernel performance. Explore the technical breakdown and code: rocm.blogs.amd.com/software-…

our blog is inspired by the OG @si_boehm’s educational gemm ladder! check out our blog here: rocm.blogs.amd.com/software-… with @neoblizzz, ryan swann, @_williamhu, @seansiddens, @drewwad, julia zhang, alex underwood, alex dutu, @dylan__lim. thank you @amd @AIatamd for the support!

ThunderKittens is now on Vera Rubin! It's been a fun past week playing with new kernels and hardware. Check out the blog below for more details on how we pushed our GEMMs to cuBLAS levels of performance! 🐱⚡️🔭

thunderkittens is now running on nvidia vera rubin! our kernels team got early access to nvl72 and spent the past few days digging through the new isa and bringing nvfp4 + fp8 gemms to life after reworking the kernels for rubin, we pushed them past 22 and 12 pflops respectively — competitive with cublas + cute dsl

thunderkittens is now running on nvidia vera rubin! our kernels team got early access to nvl72 and spent the past few days digging through the new isa and bringing nvfp4 + fp8 gemms to life after reworking the kernels for rubin, we pushed them past 22 and 12 pflops respectively — competitive with cublas + cute dsl

Open source is becoming the default way enterprises build with AI. Not just which model runs, but where it runs. Today we're partnering with @Equinix and @nvidia on Equinix Inference Exchange: our open-model platform, live across Equinix's global data centers. Model choice and performance were never a trade-off. newsroom.equinix.com/2026-09…

We just signed one of the largest AI infrastructure deals for open source, period. 250MW data center. $5B+ in annualized revenue. Built with @HUMAIN in Saudi Arabia. tinyurl.com/4nzcjcu7

In-context continual learning requires models to accumulate experience and reuse it later in the same sequence. But an RNN compresses an ever-growing history into a fixed-size state, where each token gets a single write into memory. We study dynamic compression: letting the model revisit the past and reorganize its state as it discovers what needs to be reused.

We’ve seen an explosion of new ML chips with unique architectural features, but software support remains the critical bottleneck Achieving peak performance increasingly relies on hardware-specific optimizations in the kernels, but we observe that coding agents are particularly weak at this Introducing Hawkeye, a framework that brings hardware-awareness to coding agents by grounding them in a minimal and comprehensive taxonomy of optimization strategies For new GPU or ML accelerator architectures, you only need to write 10 unit tests and solution kernels (one per optimization strategy), and we show that coding agents can effectively scale test-time compute with this minimal supervision to write hardware-aware kernels Hawkeye can port kernels across architectures (Ampere, Hopper, Blackwell), vendors (NVIDIA, AMD), and precisions (FP8, NVFP4, MXFP4) while consistently leveraging hardware features and approaching expert kernel performance Work co-led with @keramakr and done in collaboration with Alexander Ingare @simonguozirui @18jeffreyma @ZishenW @simran_s_arora @Azaliamirh @profvjreddi