1. Huang Renxun says older GPUs appreciate like "fine wine"
NVIDIA CEO Jensen Huang recently stated that driven by explosive AI computing demand, even GPUs released four to five years ago have not depreciated but instead continue to increase in value. This "fine wine effect" highlights the extreme shortage in the AI computing market, with H100 mainstream models seeing quarterly price increases of 15% to 20%, and computing resources selling out as soon as they become available.
2. U.S. approves H200 exports to China, but Chinese buyers show little enthusiasm
The U.S. Department of Commerce has formally approved the export of NVIDIA H200 AI chips to ten Chinese companies including Alibaba, Tencent, and ByteDance, with a maximum purchase limit of 75,000 chips per buyer. However, no substantial transactions have taken place so far. The U.S. requires a 25% revenue share as an附加 condition, and coupled with China's accelerated push for domestic computing alternatives, these chips face a situation of "licensed but no orders."
3. AMD beats Q1 earnings expectations, data center revenue surges 57%
AMD released its Q1 2026 earnings, with total revenue reaching 5.8 billion, a 57% surge and surpassing the sum of client and gaming businesses for the first time. The report shows that driven by the agentic AI wave, the CPU-to-GPU ratio in AI server clusters is shifting from 1:8 to 1:1. AMD is benefiting from a structural growth opportunity with its "dual-driver" strategy of EPYC server CPUs and Instinct GPUs.
4. ByteDance raises AI capital expenditure to RMB 200 billion, expands domestic chip procurement
ByteDance has increased its 2026 AI infrastructure capex budget by 25% to approximately RMB 200 billion, with about RMB 85 billion dedicated to AI chip procurement. The company is significantly adjusting its procurement strategy, increasing the share of domestic AI chips, and has pre-purchased over $5 billion worth of domestic computing products involving suppliers such as Cambricon and Huawei Ascend. This upward revision is seen as an important signal that domestic computing demand is being validated by major tech firms.
5. Tencent and Alibaba simultaneously boost AI infrastructure spending, expanding demand for domestic computing
The two internet giants have simultaneously signaled increased AI capital spending in their latest earnings reports. Tencent's Q1 capex reached RMB 31.9 billion, up about 16% year-over-year, and explicitly stated that more domestic chips will be put into use in the second half of the year. Alibaba announced that its future AI infrastructure investment will far exceed the previously committed RMB 380 billion, while its Pingtouge self-developed GPU chip has achieved mass production, with over 60% of its computing power serving external commercial clients, supporting AI tasks for more than 400 enterprise customers.
6. Domestic inference GPU unicorn Xiwang completes over RMB 1 billion funding round
Xiwang (Sunrise), a domestic full-stack self-developed AI inference GPU company, announced a new funding round of over RMB 1 billion, becoming China's first pure-play inference GPU unicorn valued at over RMB 10 billion. Xiwang was formerly the large-chip division of SenseTime. It has mass-produced two generations of inference GPU chips, with deliveries exceeding 10,000 units in 2025. The company aims to reduce token costs by 90% and build a trillion-scale inference infrastructure.
7. Cerebras goes public on Nasdaq, jumps 68% on debut as year's largest IPO
AI chip startup Cerebras officially listed on the Nasdaq, priced at 311.07, briefly reaching a market cap of over 5.55 billion, becoming the largest tech IPO in the U.S. so far in 2026. Cerebras focuses on the "wafer-scale chip" approach, turning an entire wafer into a single AI processor to achieve higher bandwidth and stronger computing power. It has already partnered with OpenAI, AWS, and others.
8. CME plans to launch world's first GPU computing lease futures
CME Group, in partnership with GPU market intelligence firm Silicon Data, announced plans to launch cash-settled futures linked to GPU computing lease rates, targeted for launch within 2026. These futures contracts would allow buyers such as AI model companies and cloud providers to lock in future computing supply and prices in advance. This is seen as a key milestone in the financialization and assetization of computing power, comparable to the launch of crude oil futures in 1983.
9. Huawei Ascend 950PR snapped up by three tech giants, price jumps 20%
Driven by factors such as the upcoming release of DeepSeek V4, Alibaba, ByteDance, and Tencent have placed bulk orders for Huawei's Ascend 950PR chips, totaling hundreds of thousands of units. The chip just began mass production this month, with Huawei targeting annual shipments of about 750,000 units. The concentrated procurement from the three giants has pushed the price of the 950PR up by 20% in recent weeks, indicating strong supply and demand in the domestic high-end AI chip market.
10. China's new "computing-electricity synergy" regulation raises bar for data center construction
Following the release of the "Action Plan for Bidirectional Empowerment of AI and Energy" by four national ministries, many regions have begun strictly enforcing "computing-electricity synergy" access standards. New large- and medium-sized computing centers must now submit green power supply and energy storage support plans simultaneously when applying for approval and registration. Unqualified projects will not be filed or granted grid access. This policy aims to end the chaotic, self-built status of the industry and push computing construction into a standardized operation phase based on carrying capacity planning.