New satellite imagery offers a window into Huawei Technologies’ determined push for semiconductor independence, revealing what the Financial Times describes as an advanced chip production line undergoing rapid expansion in Shenzhen. The visual evidence shows the strategic effort taken by Huawei and China to develop domestic capabilities and lessen reliance on foreign technology, particularly as the United States continues to tighten export restrictions on essential AI hardware.
Rapid Build‑Out of the Guanlan Manufacturing Cluster
Satellite images taken between January 2024 and April 2025 show Huawei’s once‑empty hillside plot in Guanlan, northern Shenzhen, now dominated by a 50,000 m² clean‑room block, support annexes and a completed chilled‑water plant—a sequence of milestones that in standard industry schedules comes only weeks before large‑scale “tool move‑in.” The latest imagery even captures rows of rooftop scrubber stacks and bulk‑gas tanks being craned into place, visual proof that the fab is entering its final fit‑out phase. (See Maxar high-resolution images for a clearer view)
Roughly one kilometre east, sister company SiCarrier has finished cladding two lithography‑tool halls and erected a 70‑metre skybridge that connects directly to Huawei’s dock. Making its public debut at Semicon China on March 27, SiCarrier showcased a full suite of deep‑ultraviolet (DUV) scanners, etchers and metrology gear, claiming the line can reach 5 nm through self‑aligned quadruple patterning.
The third hub, memory‑packager SwaySure, topped out a fourth‑storey clean‑room in February 2025. High‑resolution Maxar tiles from March reveal copper‑column wafer frames and HBM3 test stacks staged outside, while Nikkei Asia sources say SwaySure will sample 24‑layer HBM to Huawei Cloud in the second half of the year.
Supporting infrastructure is racing to keep pace: Shenzhen’s municipal grid energised a new 220 kV substation on the cluster’s western edge in early April, and local permit filings show a semiconductor‑grade wastewater line connecting the site to the city’s “Project Starfish” utility corridor. Satellite shots from late April capture convoys of SMEE and Naura crate trucks queued at the northern gate—visual confirmation that the first lithography, etch and deposition tools are entering the building.
Huawei is also betting on a domestic extreme‑ultraviolet (EUV) path. A prototype laser‑plasma EUV source is undergoing shake‑down tests at the company’s optics campus in Dongguan; industry trackers expect trial exposures by Q3 2025, with limited pilot production in 2026.
Taken together, the imagery and on‑the‑ground disclosures suggest Huawei could begin initial 7‑nm wafer starts before year‑end 2025, with a road‑map to 5 nm once SiCarrier’s advanced DUV kit is proven. The pace—less than four years from earth‑moving to production—highlights how tightly Huawei is choreographing its in‑house fab and nominally “independent” suppliers to blunt the impact of widening U.S. export controls.
This physical build-out aligns with Huawei’s recent strategic maneuvers in the artificial intelligence chip market, suggesting these expanding facilities are geared towards producing the company’s Ascend series processors. The drive for self-sufficiency intensified following actions by Washington.
Effective April 15, the U.S. Department of Commerce imposed an “indefinite license” requirement for exporting Nvidia’s H20 and AMD’s MI308 AI accelerators to mainland China, Hong Kong, and Macau. Officials cited national security risks, with Commerce Department spokesman Benno Kass stating the department is committed “to acting on the president’s directive to safeguard our national and economic security.” This effectively sidelined the H20, Nvidia’s primary chip tailored for the Chinese market under previous export rules.
The financial impact on Nvidia was immediate, forcing the company to disclose in a regulatory filing a US$5.5 billion charge against revenue linked to the now-stranded H20 inventory. In response, Nvidia CEO Jensen Huang made a surprise visit to Beijing on April 17, meeting trade officials and reportedly stating Nvidia hoped “to continue to cooperate with China,” while also conferring with AI firm DeepSeek founder Liang Wenfeng, potentially about compliant hardware options.
Huawei’s Hardware Offensive Accelerates
Huawei appears positioned to capitalize on the market disruption. The company is reportedly starting mass shipments of its Ascend 910C artificial intelligence processor this month, May 2025, aimed at Chinese AI firms impacted by the H20 ban. Analyst Paul Triolo of Albright Stonebridge Group observed that the US restrictions “will mean that Huawei’s Ascend 910C GPU will now become the hardware of choice for (Chinese) AI model developers and for deploying inference capacity.”
The Ascend 910C employs chiplet integration – combining smaller, specialized silicon dies – linking two previous-generation 910B processors. This approach reportedly achieves performance comparable to Nvidia’s H100 (itself banned in China since 2022) and delivers around 780 TFLOPS using the BF16 numerical format (BFloat16), which balances computational speed and numerical precision for AI tasks.
Looking further ahead, Huawei announced its Ascend 920 chip on April 19, 2025. Built on a 6nm process, it targets over 900 TFLOPs and uses HBM3 (High Bandwidth Memory 3) – a type of stacked memory crucial for feeding data rapidly to powerful AI processors – aiming for mass production in the latter half of 2025. Concurrently, Huawei introduced the AI CloudMatrix 384 cluster. This system integrates 384 Ascend 910C chips, reaching an estimated 300 PFLOPS (BF16) and 49.2 TB of total HBM.
While these figures surpass Nvidia’s 72-GPU GB200 NVL72 system on paper (approx. 180 PFLOPS, 13.8 TB HBM), the CloudMatrix achieves this performance through sheer scale. This approach results in a much higher estimated power consumption of 559 kW versus the GB200’s 145 kW, highlighting a potential trade-off between compute density and energy efficiency, perhaps made more viable by lower energy costs reported in parts of China earlier this year. The CloudMatrix system utilizes an all-optical network, employing Linear Pluggable Optics (LPO), a potentially lower-power transceiver technology for shorter data center links.
Navigating the Manufacturing Maze
Successfully producing these chips domestically involves navigating existing US sanctions. While China’s top foundry, SMIC, reportedly uses its 7nm-class N+2 process for some components, reports have mentioned yield difficulties. Furthermore, Reuters and SemiAnalysis suggest many current 910C chips utilize silicon originally made by Taiwan’s TSMC for Sophgo, a Chinese design firm, allegedly obtained by Huawei through intermediary channels.
This purported link led to US regulators investigating TSMC’s work for Sophgo, as reported in early April 2025. Huawei denied using TSMC-made Sophgo chips to Reuters, while TSMC stated it complies with regulations and ceased direct supply to Huawei in September 2020. Similar complex supply chain maneuvers involving distributors like CoAsia Electronics and other partners are suggested for sourcing vital HBM components, potentially originating from Samsung.
China’s Strategic Shift
This intricate backdrop underscores the intensifying US-China tech competition. While Washington seeks to limit China’s access to advanced technology, these measures appear to spur China’s own efforts, supported by initiatives like the $47.5 billion “Big Fund,” aimed at semiconductor self-reliance.
Even before the H20 ban, Chinese firms were reportedly stockpiling Nvidia chips in anticipation of restrictions. With key foreign options removed, domestic players like Huawei, and potentially smaller Chinese GPU startups, are set to meet the demand. However, recent reports indicate that Nvidia prepares new China-focused AI chips designed to meet US export controls, just like the H20 chip did before.
