Zhipu AI launches GLM-5.3-FlashX; Nvidia CEO predicts 2027 chip sales will double
On September 18, Zhipu AI launched GLM-5.3-FlashX, achieving 200 tokens/s inference speed, a 5x improvement, priced at 2.5x the original. The company raised its 2026 ARR guidance from $2.4B to $3.0B. Overseas, Nebius announced 17-21% GPU price hikes, Marvell and GlobalFoundries expanded AI chip capacity, and Nvidia CEO Jensen Huang predicted 2027 chip sales would roughly double this year's level. China's MIIT released a 15th Five-Year Plan targeting 9,800 EFLOPS by 2030.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Chinese AI Models Launch Rapidly; Compute Chain Remains High-Growth Core Sector
On September 18, Zhipu AI released GLM-5.3-FlashX with inference speeds up to 200 tokens/s, five times faster than the previous version, priced at 2.5 times the original. The same day, Alibaba's Qwen launched Qwen3.8-Omni-Flash, a native full-modal model supporting text, image, audio, and video input with 1M long context. Overseas, Nvidia CEO Jensen Huang stated at an AI summit in Scotland that the company expects chip sales to double next year due to accelerating AI adoption across industries. Cloud provider Nebius announced GPU price increases of 17-21% effective October 1. On the policy front, China's Ministry of Industry and Information Technology released the 15th Five-Year Plan for Information and Communication Industry Development, targeting 9,800 EFLOPS of intelligent computing power by 2030. Tianfeng Securities views the compute chain as a high-growth core sector amid complex macro conditions, with domestic AI investment from Alibaba and ByteDance supporting development. JPMorgan's Data Center Watch report for June shows strong growth in model call volumes, API spending, and GPU rental prices, indicating expanding AI infrastructure demand despite declining token prices. The article also promotes the Kechuang-Chuangye AI ETF (159142), which tracks the CSI Kechuang-Chuangye AI Index, noting its 71.97 PE-TTM at the 2.85th percentile over two years, suggesting valuation attractiveness.
Read sourceOverseas and domestic AI compute demand remains high, ChiNext ETF tracks infrastructure chain
As of 10:25, the ChiNext Compute Infrastructure Index rose 0.7%, with Sihui Fushi up 5.6% and Hongxin Electronics up 3.9%. China Securities Co. (CSC) noted that domestically, Zhipu recently launched GLM-5.3-FlashX, optimizing inference efficiency on domestic compute clusters and raising its 2026 year-end ARR guidance from $2.4 billion to $3.0 billion, validating accelerating commercialization of domestic large models. Overseas, Nebius raised GPU cloud rental prices by 17% to 21%; Marvell and Globalfoundries expanded AI high-speed optical interconnect chip capacity; Jensen Huang expects Nvidia's 2027 chip sales to be about double this year's. CSC concluded that AI training, inference, and infrastructure demand remains at high景气 (prosperity) both domestically and internationally, and continues to favor domestic large models and the compute supply chain including GPU/ASIC, optical modules, high-end PCB, servers, and liquid cooling. The ChiNext Compute Infrastructure Index selects 50 securities covering computing, networking, storage, and operations, and the ChiNext Compute ETF (158050) tracks it for investors seeking broad exposure.
Read sourceChiNext Computing ETF Tracks Index as AI Infrastructure Demand Stays High
As of 10:25, the ChiNext Computing Infrastructure Index rose 0.7%, with Sihui Fuji up 5.6% and Hongxin Electronics up 3.9%. China Securities Co. noted that Zhipu AI recently launched GLM-5.3-FlashX, which improved inference efficiency on domestic computing clusters and raised its end-2026 ARR guidance from $2.4 billion to $3.0 billion, validating accelerating commercialization of domestic large models. Overseas, Nebius raised GPU cloud rental prices by 17% to 21%; Marvell and Globalfoundries expanded AI high-speed optical interconnect chip capacity; and Jensen Huang expects Nvidia's 2027 chip sales to be about double this year's. The article concludes that AI training, inference, and infrastructure demand remains robust both domestically and internationally, and it is optimistic about the computing power supply chain including domestic large models, GPU/ASIC, optical modules, high-end PCBs, servers, and liquid cooling. The ChiNext Computing Infrastructure Index selects 50 listed securities covering computing, networking, storage, and operations in the ChiNext market, spanning data centers, communication equipment, PCBs, storage, and liquid cooling. The ChiNext Computing ETF managed by E Fund tracks this index. The article includes a risk disclaimer noting that funds carry risks, ETFs trade on exchanges with premium/discount and liquidity risks, and past performance does not guarantee future results.
Read sourceShow 2 older updatesHide older updates
CITIC Securities: Zhipu AI Commercialization Accelerates, Overseas Computing Demand Remains Strong
According to a September 21 research report from CITIC Securities cited by National Business Daily, Chinese AI company Zhipu AI launched GLM-5.3-FlashX on September 18, optimizing inference efficiency on domestic computing clusters. The company raised its 2026 annual recurring revenue (ARR) guidance from $2.4 billion to $3.0 billion, indicating accelerating commercialization of domestic large language models. Overseas, Nebius raised GPU cloud rental prices by 17%-21%, Marvell and GlobalFoundries expanded AI high-speed optical interconnect chip production, and Nvidia CEO Jensen Huang expects Nvidia's 2027 chip sales to be roughly double this year's level. CITIC Securities believes AI training, inference, and infrastructure demand remains robust both domestically and internationally, and maintains a positive outlook on domestic large models and the computing power supply chain including GPUs/ASICs, optical modules, high-end PCBs, servers, and liquid cooling.
Read sourceCSC Financial: Zhipu AI Commercialization Accelerates, Overseas Computing Demand Remains High
On September 18, Zhipu AI launched GLM-5.3-FlashX, achieving a maximum inference speed of 200 tokens/s, a 5x improvement over GLM-5.3-Flash, with a price increase of about 2.5x. Supported by a domestic computing cluster of 100,000 chips, the model's efficiency gains led CSC Financial to raise Zhipu's year-end 2026 ARR guidance from $2.4 billion to $3.0 billion, validating accelerating commercialization of domestic large models. Overseas, AI cloud provider Nebius announced a 17%-21% price increase for some Nvidia GPU rental services starting October, its second hike in three months, citing surging training and inference demand. Marvell and GlobalFoundries expanded their multi-year partnership to boost AI high-speed optical interconnect chip capacity for data centers. Nvidia CEO Jensen Huang predicted that Nvidia's chip sales in 2027 would be roughly double this year's level, a more aggressive outlook than the 70% official guidance given in late August. CSC Financial analysts conclude that global AI infrastructure demand remains in a high-growth phase, favoring domestic large models and the GPU/ASIC, optical module, high-end PCB, server, and liquid cooling supply chains.
Read source