Chinese AI Euphoria Outpaces Chip Technology

China’s artificial intelligence chip sector is enjoying a surge of investor enthusiasm, highlighted by the blockbuster market debut of Moore Threads Technology Co earlier this month. The listing marked one of the country’s most successful technology IPOs in years and reinforced optimism that China is rapidly closing the gap with US semiconductor leaders. However, beneath the market euphoria lies a far more challenging technological reality.

Rising Stars Fuel AI Chip Optimism

A growing list of domestic chipmakers — including Moore Threads, Cambricon Technologies, and MetaX — are being hailed as potential challengers to Nvidia within China’s borders. Internet giants such as Alibaba Group and Baidu are also stepping up investments in in-house semiconductors, aligning with Beijing’s push to secure AI self-sufficiency.

Confidence was further boosted after researchers at TechInsights dismantled a Huawei smartphone and discovered a processor built using more advanced methods than previously believed possible for Chinese manufacturers. Reports that Beijing is considering up to US$70 billion in funding support for the semiconductor sector have only added to the excitement.

Manufacturing Bottlenecks Limit Progress

Despite these developments, China’s AI chip ambitions face significant constraints. Advanced chip production remains heavily dependent on Semiconductor Manufacturing International Corp (SMIC), a firm restricted by US sanctions and unable to access the most advanced fabrication equipment.

The limitations are stark. Cambricon’s most advanced chips reportedly achieve yields of only around 20%, meaning the vast majority of manufactured chips are unusable. TechInsights has cautioned that SMIC’s technological progress comes with substantial cost and efficiency trade-offs.

Even Moore Threads has warned against excessive optimism, with its stock falling sharply shortly after its initial surge — a sign that markets may be running ahead of fundamentals.

Investors Bet on State Support

Market enthusiasm reflects a belief that Beijing will spare no expense to nurture domestic alternatives to Nvidia, viewing AI hardware as a strategic pillar of national competitiveness. That long-term policy commitment remains firm, even as short-term options emerge.

US President Donald Trump’s recent offer to allow Nvidia to sell its H200 chips to China could temporarily ease performance bottlenecks. Still, analysts say such measures are unlikely to derail China’s broader push for semiconductor independence.

“Access to foreign chips may help in the near term, but the strategic focus on domestic innovation remains unchanged,” analysts note.

Software Innovation Offsets Hardware Gaps

China’s AI ecosystem has shown resilience by innovating around hardware constraints. The recent success of DeepSeek, which released a capable AI model without full access to top-tier Nvidia accelerators, highlights how Chinese firms are optimising software and open-source models to compensate for silicon limitations.

Industry experts argue that these constraints may even foster creativity, pushing Chinese developers to extract maximum efficiency from less powerful chips.

Critical Supply Chain Challenges Remain

However, fundamental obstacles persist. Chinese AI chip designers cannot use Taiwan Semiconductor Manufacturing Co’s most advanced processes due to export controls, placing them at a disadvantage against Nvidia and AMD.

Memory supply is another weak point. High-bandwidth memory chips remain largely unavailable domestically, forcing reliance on imported or stockpiled components. Huawei’s latest AI chips reportedly still use older-generation memory from South Korean suppliers.

Meanwhile, restrictions from the US, Japan, and the Netherlands prevent access to key manufacturing tools such as ASML’s extreme ultraviolet lithography systems — essential for producing cutting-edge AI processors.

A Long Road Ahead

Huawei plans to produce roughly 600,000 Ascend 910C chips next year, while Cambricon targets 500,000 AI accelerators by 2026 — impressive figures by local standards. Yet Nvidia alone is estimated to have sold about one million China-compliant H20 chips in 2024.

The contrast underscores a sobering truth: while China’s AI chip sector is advancing, investor enthusiasm may be outpacing the industry’s current technological and manufacturing realities.