Cerebras CS-4 doubles accelerator performance on the same chip
A new rack-scale AI accelerator, a nine-figure robotics raise, and fresh evidence that chatbots quietly steer sensitive-health decisions — all in the last few hours.
Cerebras has introduced the CS-4, a rack-scale AI accelerator its CEO Andrew Feldman calls the fastest system in the industry. The machine still runs on the same 5nm WSE-3 wafer as its predecessor, but Cerebras says it doubles the CS-3's performance by pushing clock speed higher through more power and better cooling. A single rack now packs three wafers instead of two and delivers up to 4,400 tokens per second per user — roughly 30 times faster than comparable setups running on Nvidia GPUs, by the company's accounting. Memory stays flat at 44 GB per wafer.
The bigger story is the design philosophy. The CS-4 uses a modular "Backpack" architecture meant for faster assembly, plus disaggregated inference through partners like AMD and AWS Trainium — a bet that inference, not training, is where the market is heading. Analysts at SemiAnalysis see the networking gains as fairly small, and more detail lands at the Hot Chips conference. It's the clearest sign yet that Cerebras, whose hardware already powers OpenAI's Codex Spark, is positioning itself as the speed-focused alternative in the Nvidia-dominated inference race — and claiming a 30x edge is a direct challenge to the incumbent's assumptions about what fast means.
XPeng says its robotics business raised over $900 million led by IDG and Gaorong Ventures at a $6.3 billion valuation, with Tencent and Alibaba participating. The round is the clearest sign yet that a Chinese automaker is serious about becoming a robotics player, not just an EV maker hedging on the side. Bringing in Tencent and Alibaba alongside the GPU-led deal signals that China's biggest platforms see embodied AI as the next growth lane — and that automakers, with their manufacturing muscle and supply chains, are increasingly the ones bankrolling the physical-AI buildout.
An AlgorithmWatch investigation across 270 responses found that ChatGPT, Gemini, Grok, and Claude regularly link pregnant users to anti-abortion websites without disclosing the sources' ideological stance. In at least one in four queries, the chatbots surfaced an anti-abortion link; the group Profemina, which AlgorithmWatch ties to US-based Heartbeat International, appeared in about 17 percent of all responses — often without any warning. Gemini referenced Profemina five times in a single conversation and then warned about the very same source later in the chat. The gulf between empathetic framing and what the underlying sources actually argue is what makes these responses feel trustworthy to people who don't know to look — a pattern that extends well beyond abortion to any sensitive topic where AI answers become gatekeepers.
Would you trust a chatbot's medical referral without knowing who's behind the source it recommends? Tell us in the comments.
Sources: The Decoder — Cerebras CS-4 · Cerebras · SemiAnalysis · Reuters via Techmeme — XPeng robotics · The Decoder — chatbots and pregnancy · AlgorithmWatch
