The AI hardware race isn't just about raw cores anymore. Semidynamics' latest funding round underscores a critical shift: tackling the "memory wall" with a new architectural paradigm.
Forget raw AI capability; the real bottleneck is now time. Fractile just raised $220 million on the audacious premise that the future of AI hinges on radically faster inference hardware.
As AI's energy appetite explodes, one startup is looking skyward, proposing data centers on satellites. It's a bold vision, aiming to harness solar power and escape terrestrial grid limitations.
Arm meets RISC-V in a bid to conquer the AI inference server market. SiPearl and Semidynamics are launching a new rack-scale platform, but the real question is who benefits and at what cost?
AI labs are desperate for faster, cheaper compute. Anthropic's reported interest in UK startup Fractile suggests a serious play for performance gains, potentially shaking up the chip landscape.
Tenstorrent isn't just entering the AI hardware race; they're declaring war. Their new Galaxy Blackhole servers, powered by RISC-V, claim performance metrics that could fundamentally shift the landscape, and with prices that might just make hyperscalers listen.
The AI gold rush is hitting a wall, and chipmakers are quietly scrambling. Intel and SambaNova's latest move isn't just another partnership; it's a pragmatic pivot away from the GPU-only playbook.
Nvidia shelled out $20 billion for Groq's team and tech. Now they're admitting it: GPUs alone won't cut it for low-latency AI inference anymore.
Deep in a blazing data center, VSORA's Jotunn8 chip devours inference workloads like a Norse giant feasting endlessly. No more data droughts—just pure, relentless AI power.
Jensen Huang's full-platform sermon just got empirical backup. Nvidia's software tweaks propel MLPerf inference benchmarks to absurd new heights, leaving rivals in the dust.
Forget the training hype—Nvidia's surprise Groq 3 LPU at GTC signals inference is now king. This lean beast prioritizes speed over brute force, reshaping data centers.
A fresh tape-out from France's VSORA targets the memory wall strangling AI inference. Sandra Rivera explains why this could flip the script on power-hungry GPUs.