An ASIC (application-specific integrated circuit) is a chip built to do one job — such as AI training or inference — instead of general-purpose computing. Because they’re optimized for a narrow task, ASICs can be faster and more power-efficient than GPUs for that workload. This hub tracks the custom AI chips built by hyperscalers and chipmakers, and what they mean for the AI hardware race.
Etched builds a chip that only runs transformer models, betting narrow focus beats Nvidia's flexibility — and investors are reportedly valuing the three-year-old startup at up to $20 billion.
Read more →
ASICs are chips hardwired for one job. Here's what makes application-specific AI chips different from general-purpose GPUs, and why some AI companies build their own.
Read more →Chipmaker Etched is negotiating a funding round that would value it at $20 billion, quadrupling its December price, as investors chase alternatives to Nvidia for AI inference, the Wall Street Journal reported.
Read more →Shanghai startup Dongfang Suanxin unveiled its DF1000 AI chip on July 13, bypassing US export curbs with 3D memory stacking instead of banned high-bandwidth memory.
Read more →Meta plans to begin manufacturing its in-house "Iris" AI chip in September, aiming to double its data center computing capacity to 14 gigawatts by 2027, an internal memo reviewed by Reuters shows.
Read more →SambaNova closed the first tranche of a $1 billion Series F round led by General Atlantic, and said JPMorganChase will deploy its chips for on-premises AI inference.
Read more →Intel- and Microsoft-backed chipmaker Syntiant filed to list on Nasdaq under ticker SYTN, betting on demand for its low-power, on-device 'Physical AI' processors.
Read more →Qualcomm on June 24 unveiled the Dragonfly data center chip family — a 250-core Arm-based CPU and a new inference accelerator — with Meta signed on to deploy the hardware in its server fleet from 2028.
Read more →An AI inference chip is specialized hardware built to run trained AI models quickly and cheaply at scale. Here is why every major tech company is now racing to design its own instead of buying from Nvidia.
Read more →OpenAI and Broadcom revealed Jalapeño on June 24 — a purpose-built inference chip designed in nine months with AI assistance, targeting deployment by end of 2026.
Read more →