CONNECT WITH US
Sep 10
China's Nvidia rivals raise prices as HBM shortage bites
Chinese chipmakers challenging Nvidia are sharply raising prices for current and next-generation accelerators as high-bandwidth memory (HBM) becomes scarcer and more expensive, adding another cost constraint to China's push for domestic computing hardware.
The clearest read yet on whether the AI build-out is still accelerating arrived this week in a batch of unaudited Taipei filings, and it accelerated. The 49 Taiwan-listed companies spanning AI server assembly, cooling, power, substrates, materials, networking, test, and optics reported combined August revenue of NT$2.84 trillion (US$89.74 billion), up 76.5% from a year earlier and 9.9% from July, according to their filings to the Market Observation Post System. Both the total and the growth rate are the highest of the cycle, and the year-on-year rate is a step up from July's 61.3%.

MediaTek reported August 2026 revenue of NT$64.183 billion (approx. US$2.03 billion), up 32.41% from July and 44.08% year-over-year. Industry sources said the chipmaker's August performance came in well above expectations and puts its third-quarter revenue guidance within reach, with a chance to beat it.

Rayzher Industrial, a high-tech factory gas supply system solutions provider, reported consolidated revenue of NT$369 million (approx. US$11.7 million) in August 2026, up 83.26% year-over-year and a record monthly high. Cumulative consolidated revenue for the first eight months of 2026 reached NT$1.993 billion, up 27.91% from the same period in 2025 and also a record for the period.

Breaking from the industry's mainstream reliance on III-V compound semiconductors, Taiwanese startup Taiwan Nano & Micro-Photonics (N&M) is using complementary metal-oxide-semiconductor (CMOS) processes to develop a single chip capable of emitting more than four wavelengths of mid- to far-infrared light.

Chinese electronics manufacturer Shenzhen Kaifa Technology, also known as Shenzhen Kaifa or Kaifa, plans to invest CNY1.85 billion (about US$260 million) to expand high-end memory-chip packaging and testing capacity through its wholly owned subsidiary Payton Technology (Shenzhen), according to a company announcement.
Falling token prices are putting pressure on the economics of GPU-based AI inference, potentially prompting data center operators to rethink how workloads are divided across compute and memory resources, according to AI chip startup d-Matrix.
Synopsys has reported a 22% reduction in die area for a neural processing unit (NPU) implemented with its Fusion Compiler design tool on Samsung Foundry's SF2P process, providing a concrete design-level result as Samsung seeks to turn its 2nm technology into more commercial foundry business.
Rising demand for generative AI, high-performance computing (HPC), and data center buildouts is pushing the global IC substrate industry into a new phase marked by upgrades in high-end product specifications, tight supplies of critical materials, and intensifying competition in advanced packaging technologies.

SK Hynix is targeting 2028 to apply High-NA extreme ultraviolet lithography to DRAM mass production and is considering joining an industry initiative to develop 12-inch photomasks, according to South Korea's fnnews. The company has separately confirmed both plans.

Academia Sinica's Quantum Chip Fabrication Space (QC-Fab) in Tainan will open to external users at the end of September 2026, marking a new stage in Taiwan's in-house quantum chip development and manufacturing efforts. After the cleanroom was completed, the institute finished installing process equipment, connecting factory utility lines, testing equipment functions, and training operators.

Nuvoton Technology sees robots and other physical AI applications becoming an important source of demand for microcontrollers (MCUs) over the next decade as computing moves from the cloud into edge devices, smart factories, drones, and care applications.