China’s AI Hardware Market Split on August 5: Optical-Module Risks vs. Domestic Chip Upside

The August 5 session exposed two different forces in China’s AI-hardware market. Optical-module shares sold off after reports that the U.S. was drafting restrictions on new Chinese data-center components, while PCB and domestic AI-chip names attracted renewed interest. The proposed measure was not yet a final blanket ban: Reuters reported that the Federal Communications Commission was working on a draft covering new Chinese optical transceivers, with important details still unresolved.

The market reaction: an early sell-off, then a partial recovery

Chinese optical-component stocks came under immediate pressure. Zhongji Innolight fell as much as 14% in Shenzhen before paring the decline, while Eoptolink also dropped sharply. The broader CSI300 Telecommunications Services Index fell as much as 9% in early trading.

The recovery suggested that investors were pricing the proposal as a scenario-dependent risk rather than an immediate disruption to the entire sector. Analysts also questioned whether a comprehensive ban could be implemented without damaging U.S. data-center construction, given the importance of Chinese suppliers to high-speed optical interconnects.

Victory Giant Technology rallied as the PCB and AI-infrastructure theme strengthened. The available market account confirms the move but does not establish a single, independently verified company-specific catalyst for that day, so the rally is best described as part of the broader PCB theme rather than attributed to a definite announcement.

Nomura’s framework: the scope of a ban matters more than the headline

Nomura’s analysis separated two possible outcomes.

A narrow ban on modules made in China

If U.S. restrictions covered only products physically manufactured or assembled in mainland China, suppliers with overseas high-end production would have a way to reduce their exposure. That would not eliminate the risk, but it could limit the immediate effect on companies able to serve customers from facilities outside China.

A wider ban covering Chinese suppliers or third-country output

A restriction aimed at all Chinese-owned suppliers—or at products made in overseas facilities controlled by Chinese companies—would be materially more disruptive. It could also create problems for U.S. cloud operators, which depend on a supply chain in which Chinese manufacturers account for a large share of high-speed optical interconnect production.

That is why the market debate focused on implementation details: whether the rule would cover only new models, where a product was manufactured, and whether ownership or control in a third country would trigger the restriction. Reuters’ reporting also noted that the proposal was still under development rather than an enacted rule.

Citi’s company-by-company exposure view

Citi’s central point was that Chinese optical-module suppliers would be difficult to replace quickly. The available commentary said seven of the world’s top 10 optical-module vendors were Chinese and that Chinese companies supplied more than half of U.S. cloud providers’ high-speed modules.

The practical implication was that geographic capacity mattered, but so did a company’s position in the supply chain. The commentary ranked TFC as the least exposed of the companies discussed, followed by DSBJ and then Eoptolink/New Easyst. Because the contemporaneous summary contains an internal wording inconsistency, that ranking should be treated as an analyst framework rather than a definitive measurement of revenue at risk.

TFC: mostly mainland capacity, but indirect exposure

TFC was described as having nearly all of its capacity in China. However, its role as a supplier of passive components to overseas module makers—including Thai operations—meant that a ban on finished Chinese-made modules would not necessarily translate into a one-for-one loss of its business. The company was also expanding in Thailand, creating a potential geographic buffer.

Eoptolink and New Easyst: overseas production as a buffer

The commentary attributed about 88% of Eoptolink/New Easyst revenue to its Thailand base. That would provide meaningful protection under a rule focused narrowly on mainland-made products. The main tail risk would be a U.S. measure that treated Chinese ownership or control of production in a third country as equivalent to mainland manufacture.

DSBJ: a mixed footprint with planned Thai expansion

DSBJ was described as having 69.7% of module capacity on the mainland and the remainder in Taiwan. Products supplied to U.S. cloud customers reportedly came from Thailand and Taiwan, reducing the relevance of mainland capacity for those particular shipments. Its stated 2027 plan called for 35 million units of 800G/1.6T capacity, with two-thirds in Thailand and one-third in Taiwan.

The broader lesson is that “China exposure” is not a single number. Investors need to distinguish between manufacturing location, customer destination, component sourcing, legal ownership and the ability to shift production.

Macquarie’s Biren thesis: domestic substitution and higher-value GPUs

The domestic AI-chip story moved in the opposite direction from the optical-module shock. Macquarie’s bullish case for Biren Technology focused on product adoption, the next-generation roadmap and the possibility of rising average selling prices as Chinese AI demand grows and access to advanced foreign GPUs remains constrained. 3

BR10X adoption and BR20X development

The BR10X GPU series was reportedly adopted by a major Chinese cloud-service provider. Biren’s next-generation BR20X was expected to complete tape-out through China’s supply chain within months, according to the investment commentary. The planned upgrades included higher compute performance, a chiplet architecture, greater die-to-die interconnect bandwidth and supernode-level system integration. 34

A separate China Merchants Bank International research note described BR20X as a second-generation product whose pre-silicon design had been completed in 2025, with commercial launch expected in 2026. 2

Price and financial assumptions

The Macquarie thesis assumed that domestic GPU and ASIC prices could rise as demand for Chinese AI computing increased. It said high-compute new chips could reach average selling prices above 100,000 yuan from 2027, supporting revenue and margin expansion. 3

Macquarie set a 140-yuan target for Biren against a cited share price of 31.44 yuan—roughly 4.5 times higher. That figure represents an analyst target under a particular set of assumptions, not a certain return or an objective forecast. 3

The report’s projections called for revenue to rise from 1.035 billion yuan in 2025 to 16.555 billion yuan in 2028, with profitability expected to inflect in 2027. It also estimated first-half 2026 revenue of about 1.2 billion yuan, a gross margin near 51% and a core net loss of approximately 524 million yuan as research and development and tape-out expenses remained elevated. 3

Those projections are ambitious and execution-sensitive. They depend on successful BR20X development, customer adoption, production scaling, pricing power and the ability to manage high development costs. The underlying product roadmap is supported by company and research materials, but the valuation case remains an analyst view rather than an established outcome. 24

What investors were really separating

The August 5 split was not simply a contest between bullish and bearish views on Chinese technology. It reflected two different risk profiles:

• Optical modules: near-term policy risk, with the impact determined by manufacturing geography, ownership and exposure to U.S. customers.
• PCBs: participation in the continuing AI-infrastructure build-out, although the specific reason for Victory Giant’s intraday rally was not independently established in the available material.
• Domestic AI chips: a longer-term substitution and product-cycle thesis, with Biren’s upside tied to BR10X commercialization, BR20X execution and higher-value Chinese accelerators.

The immediate conclusion was therefore conditional rather than definitive: a narrow U.S. restriction could be manageable for suppliers with overseas capacity, while a rule extending to Chinese-owned production worldwide would be more damaging. At the same time, the same geopolitical pressure that threatens some optical-module sales could strengthen the investment case for domestic AI-chip development—provided companies can turn roadmaps into reliable commercial shipments.