1. ByteDance in Talks to Purchase 50,000 AI Inference Chips from Iluvatar CoreX
ByteDance is negotiating with domestic GPU maker Iluvatar CoreX to purchase at least 50,000 AI chips, primarily for inference workloads. If the deal goes through, Iluvatar will become ByteDance’s third GPU supplier after Huawei and Cambricon. In 2026, ByteDance is expected to invest approximately RMB 150 billion in global computing power procurement, of which about RMB 60–65 billion will be spent domestically, with domestic chips accounting for over RMB 40 billion.
2. Jiangyuan Technology Launches Flagship AI Chip “Jiangyuan T1”
On June 24, Jiangyuan Technology unveiled its self‑developed flagship high‑end computing chip, the Jiangyuan T1, at the APEC SME Business Forum. The chip achieves full domestic localization from architecture design to manufacturing and packaging, with comprehensive performance rivaling top international high‑end computing products, placing it among the first tier of domestic mass‑produced chips. Jiangyuan Technology has built an integrated product system covering “chips + cloud services + large models + industry applications.”
3. Enflame Technology Passes IPO Review on STAR Market
Enflame Technology, one of the “Big Four” domestic GPU players, successfully passed its STAR Market IPO review. The company reported revenue of RMB 287 million in the first quarter of 2026, a year‑on‑year increase of 1,474.85%, and expects first‑half revenue to reach RMB 1.06–1.15 billion.
4. Computing Supply Chain Players Gather at MWC Shanghai
MWC26 Shanghai opened on June 24, with many computing supply chain vendors exhibiting. Currently, core components such as CPUs and optical fibres are in tight supply, and server order backlogs are piling up.
5. China’s Large Model Weekly Inference Volume Continues to Lead the World
According to industry platform statistics, from June 15 to 19, China’s large model weekly inference volume approached 13 trillion Tokens, maintaining a global lead. Domestic models and domestic chips are accelerating their synergy; DeepSeek V4, GLM 5.2, and others have achieved day‑zero adaptation with domestic chipmakers including Huawei Ascend, Cambricon, and Hygon.
6. NVIDIA Announces Vera Rubin NVL4 Supercomputing System to Ship from Q4
On June 22, NVIDIA unveiled the Vera Rubin supercomputing platform, combining the Rubin GPU and Vera CPU. A single system delivers over 7 exaflops of AI computing power and approximately 5 PF FP64 scientific computing capability, supporting up to 144 GPUs per rack. Dell, HPE, Supermicro, and others will also offer high‑density supercomputing systems based on this architecture.
7. OpenAI and Broadcom Jointly Launch First In‑House AI Inference Chip “Jalapeño”
On June 24, OpenAI and Broadcom released their first custom AI inference chip, Jalapeño, an ASIC designed specifically for large language model inference, with a design‑to‑tapeout cycle of only nine months. OpenAI handled the underlying architecture design, while Broadcom handled silicon implementation and networking hardware. Engineering samples have successfully run workloads such as GPT‑5.3, Codex, and Spark in the lab. OpenAI stated this is the first step in its multi‑generation computing platform roadmap.
8. AMD’s Stock Price Surges Over 130% This Year, with Multiple Wall Street Firms Raising Price Targets
As of June, AMD’s stock price has soared more than 130% year‑to‑date, and at least six Wall Street institutions have raised their price targets this month. Citigroup upgraded AMD to Buy, calling its MI450 GPU a “legitimate alternative” to NVIDIA. Citigroup estimates AMD’s AI GPU revenue could reach USD 33 billion in the near term. AMD posted Q1 revenue of USD 10.25 billion, with data centre business growing 57% year‑over‑year.
9. Qualcomm Unveils Data Centre Strategy and Dragonfly Product Portfolio
On June 25, Qualcomm formally announced its overall data centre strategy and introduced the Dragonfly data centre product portfolio, already securing customers such as Meta and Microsoft. The simultaneous moves by the three giants – NVIDIA, OpenAI, and Qualcomm – indicate that AI chip competition is evolving from a single GPU race into a new phase involving GPUs, ASICs, CPUs, memory, networking, and software ecosystems.
10. Global GPU Rental Prices Continue to Rise, Reinforcing Scarcity of Computing Assets
According to rental market monitoring data, as of June 15, the average on‑demand rental price for NVIDIA B200 rose about 7% from June last year, while B300 increased over 50% from November last year. Older models H200 and H100 also saw modest increases of about 3%. One cloud service provider plans to raise B200 renewal prices by approximately 94% in October. Delivery lead times for thousand‑GPU procurement orders have extended into 2027.