Market speculation that AMD is working with Google on technology for its v10-generation tensor processing unit (TPU) has prompted questions over the potential impact on Google's existing application-specific integrated circuit (ASIC) partners, including Broadcom and MediaTek.
Ethernet chipmaker IC Plus said on August 17 that its operations in 2026 have improved significantly after Airoha Technology took a stake and joined its management, while resources from MediaTek also provided substantial support. The company said it turned profitable in the first half of the year.
Samsung Electronics is reportedly on track to complete development of 10-nanometer-class seventh-generation DRAM, also known as 1d DRAM, before SK hynix. According to The Bell, citing industry sources, Samsung is targeting September 2026 for completion, while SK hynix is aiming for December 2026, giving Samsung a lead of about one quarter.
China AIoT chip designer Rockchip posted record first-half 2026 results, with revenue rising 40.6% year on year to CNY2.88 billion (US$427 million) and net profit surging 61.73% to CNY859 million, supported by accelerating edge AI adoption.
VeriSilicon posted record first-half 2026 revenue of CNY1.86 billion (US$276 million), up 91.37% year-on-year, as AI computing demand lifted custom silicon and mass-production services.
Chinese AI chipmaker Biren expects first-half 2026 revenue to reach CNY1.15-1.3 billion (US$171-193 million), up about 1,852% to 2,107% from a year earlier, as demand for GPGPU computing expands across AI coding, long-horizon AI agents and generative AI workloads, according to Cninfo data cited in a report by the Hong Kong Economic Times (HKET).
Nvidia has reached a deal with OpenAI to provide up to US$105 billion in credit guarantees for the AI startup's upcoming 8GW data center campus in Ohio. The support will help OpenAI to secure a lease from SB Energy, while Nvidia will be the facility's exclusive chip supplier.
On February 20, 2026, Canadian AI chip startup Taalas unveiled its HC1 inference chip. Taalas said the chip ran the Llama 3.1 8B model at a single-user inference speed of 16,960 tokens per second, roughly 48 times the inference speed of an Nvidia B200 in the company's testing.
Arm Holdings' expansion from asset-light IP licensing into finished data center processors is putting the company into direct competition for wafer and memory capacity just as demand for its new AGI CPU exceeds US$2 billion.
Humanoid robots are moving from demonstrations toward mass production, changing the priorities of chip design. Instead of simply chasing higher AI compute and TOPS, or trillions of operations per second, the industry is increasingly focused on how efficiently perception, computing, decision-making, and control can work together within tight power, cost, and latency constraints.
MediaTek has reportedly landed major orders for Google TPU products in recent years, even appearing to challenge Broadcom's position. At a recent earnings call, MediaTek also raised its 2027 ASIC market share target to 15% to 20%, and industry watchers largely credit that goal to the breadth of MediaTek's ASIC services.
Less than a week after LG Group and Nvidia signed a strategic memorandum of understanding (MOU), Madison Huang, Nvidia's senior director of product marketing for Omniverse and robotics, is reportedly preparing to visit South Korea and head directly to one of LG's key robotics R&D sites.
High-bandwidth memory (HBM), a core component in AI semiconductors, has drawn fresh scrutiny after former Intel CEO Pat Gelsinger called it "lousy" at a recent AI summit in Paris. SK Hynix, the market leader in this type of memory, also said it is not the final answer to AI memory bottlenecks, underscoring the industry's push to find a next-generation memory architecture beyond HBM.
