Qualcomm has been in talks to acquire Tenstorrent, an AI chip startup, The Information reported, citing a person with direct knowledge of the deal. The two companies have discussed a price of US$8 billion to US$10 billion, the person said, a significant premium to Tenstorrent's last known valuation.
Shanghai Enflame Technology is nearing a STAR Market listing, bringing another Chinese AI chipmaker closer to public markets while losses, Tencent concentration, and a small share in Nvidia-led accelerators remain unresolved.
Nvidia's dominance in AI is extending beyond model training and deeper into inference—the fast-growing segment of the AI market responsible for running deployed models and generating revenue.
As cloud service providers ramp up investment in AI infrastructure, Edom chairman Wayne Tseng said rising costs for materials, production equipment and labour will keep overall market supply tight in the second half of 2026, with price increases likely to continue. He also said AI is moving from the cloud into enterprise applications, with power, cybersecurity, optics and the medical sector set to become four key growth areas.
Nvidia has begun telling Chinese clients that its new Vera central processing unit (CPU) could be available as soon as August and that they can start placing orders, Reuters reported, citing three people familiar with the matter.
Taiwanese power management IC supplier Global Mixed-mode Technology (GMT) will follow peers in seeking price increases from customers as tight chip supply continues to reshape capacity allocation across the industry.
Silicon Labs is deepening its presence in India through expanded research operations and a greater commercial focus on smart infrastructure applications, even as the US-based wireless chipmaker prepares for an acquisition by Texas Instruments.
In power electronics engineering, Silicon Carbide (SiC) companies are competing to achieve the absolute lowest thermal resistance (Rth), with the mindset that lower heat signature equates a superior system. However, at PCIM Europe 2026, a collaborative project between Rohm Semiconductor, Schweizer Electronic, and eMoveUs GmbH exposed a revolutionary counter-intuitive shift in design philosophy: willingly accepting a localized thermal performance deficit to ultimately achieve dominant system-level advantages.
Breaking the inference barrier requires a rethink of the whole system architecture, not just faster compute. This was the key takeaway from a recent panel discussion at SuperAI Singapore, which brought chip makers and an AI model accelerator together to address how to overcome inference bottlenecks at a time when compute workloads are hitting up against physical limits.
Concerns have emerged in recent weeks that the commercialization of co-packaged optics (CPO), a technology widely viewed as the future of AI networking, could face further delays as manufacturing yields remain inconsistent.
Astera Labs expects its Scorpio switch-chip business to become its largest revenue contributor by the end of 2026, marking a shift for the cloud AI connectivity chip supplier as demand grows for scale-up infrastructure in AI clusters.
