New York is loud. The tech scene’s louder. But every so often, something slips through the noise that doesn’t just get heard, it changes the frequency. Today, that’s Mako, a company that took one look at the grind of GPU performance engineering and said, “What if we just made the machines do it better than any human ever could?” No late-night compiler errors, no senior engineer locked in a caffeine-fueled staring contest with CUDA code. Just AI that writes, tests, validates, and optimizes GPU kernels, across NVIDIA, AMD, and cloud, like it was born to do it.
Founded in March 2024 by Waleed Atallah, a UCLA alum with a track record at Intel and Untether AI, Mohamed Abdelfattah, a machine learning systems scientist fluent in the dark arts of compilers, and Łukasz Dudziak, a GPU runtime surgeon who knows how to squeeze performance until it begs for mercy, Mako didn’t just pop up out of nowhere. It was forged in the frustration of doing things the hard way, weeks of tuning reduced to under 60 seconds of machine-led brilliance.
That vision just got $8.5 million sharper, courtesy of M13 Ventures leading their seed round, joined by a quiet cadre of angels from AI infrastructure who prefer to let their influence speak in teraflops, not tweets. The funding is fuel for scaling MakoOptimize, the company’s autonomous tuning system for inference stacks like vLLM and SGLang, and for pushing deeper into partnerships with the big three clouds. Think faster launches, more integrations, and enterprise deployments that make CFOs smile because compute bills drop by as much as 70%.
The tech is not smoke and mirrors. MakoGenerate writes custom CUDA and Triton kernels, compiles them, benchmarks them, and keeps iterating until they’re the fastest thing your hardware can run. Then it keeps going, live-tuning workloads in production, hardware agnostic, without needing access to your source code. Deploy it in your VPC, your on-prem cluster, or straight through Azure Marketplace. It is performance engineering without the bottlenecks, and the patents pending suggest they are not just here for the current generation of GPUs, they are laying track for the ones still on the drawing board.
It’s easy to get lost in AI’s hype cycle, but Mako is aiming at the plumbing, not the pretty interface. And in AI, the plumbing decides whether you’re running a high-rise or watching the ceiling leak. With this raise, Waleed Atallah, Mohamed Abdelfattah, and Łukasz Dudziak are betting on a future where AI performance is automated, portable, and perpetually improving, because the only thing better than faster code is code that keeps getting faster without asking permission.
