In the past, when discussing AI computing power, everyone's attention was focused on GPU and servers, but in 2026, the trend completely changed.

From top-level documents from the Ministry of Industry and Information Technology and the National Development and Reform Commission to a series of support plans from Guangdong Province and Shenzhen, a series of important policies have been implemented in quick succession. A small, inconspicuous device hidden deep in the cabinet— the optical module — once regarded as a "communication accessory," has been officially pushed to the strategic foundation of AI computing power and 6G networks.

Ⅰ.Three Key Figures That Reveal the True Message Behind the Policy

Many people only understand policies as "Positive Developments, Subsidies, and Industry Prosperity".

But beneath the surface, three irreversible red lines for the industry are hidden.

Figure 1: 800G/1.6T becomes the standard configuration for newly built 10,000-card intelligent computing centers

The policy clearly requires that newly built intelligent computing clusters with a capacity of 10,000 kilowatts in China must be equipped with 800G and 1.6T high-speed optical modules as standard, and 400G low-speed modules should be gradually phased out of the new procurement list.

In simple terms:

In the past, speed upgrades relied on gradual iterations driven by market demand; now, policies set timelines for upgrades.

Domestic large-scale computing infrastructure construction no longer leaves room for low-speed products. 800G is a current necessity, 1.6T has changed from a "future direction" to a task for large-scale deployment, and 3.2T has entered the track of technology reserve.

For manufacturers still clinging to 400G capacity, the ceiling of the domestic computing power market has gradually been lowered.

Figure 2: Localization rate for national-level intelligent computing clusters ≥ 70%

This is the rule that is most easily overlooked, yet has the most profound impact.

It's 70% domestically produced, not just a simple assembly of domestically made casings. The real bottleneck lies in the high-speed optical chip.

Currently, many domestic module manufacturers still rely on overseas procurement for laser chips. If upstream chips cannot be independently controlled, even if the finished product is assembled domestically, it will be difficult to overcome this hurdle of domestic production.

The profit structure of the industry will also undergo a major reshuffle.

In the past, profits were concentrated in midstream optical module manufacturing; in the future, the value center will shift upstream: indium phosphide substrates, high-speed EML optical chips, and thin-film lithium niobate materials will become the main battleground for the industrial chain over the next 3-5 years. The National Indium Phosphide Special Fund is precisely targeting this weakness to strengthen the supply chain.

Figure 3: The self-sufficiency target for high-end optical chips is 45% by 2028.

This is not just a slogan, but a task list with specific timelines.

High-speed optical chips have always been the weakest link in China's computing power supply chain. The policy's clear objectives mean that over the next two years, R&D subsidies, demonstration projects, and computing power procurement orders will be tilted towards companies that develop their own optical chips.

Ⅱ.With three major technical routes locked in, there are no more choices in the race.

A thorough reading of the complete set of policies for 2026 reveals a clear national technological direction: CPO, LPO, and NPO low-power optoelectronic interconnects, with silicon photonics modules being the long-term focus.

Why does the government strongly support this route?

There is a physical problem that cannot be avoided in the development of computing power.

As optical module speeds advance from 800G to 1.6T and 3.2T, the power consumption of traditional pluggable optical modules will increase dramatically. The interconnection of tens of thousands of GPU clusters and the massive number of optical modules bring enormous energy consumption pressure, significantly increasing the PUE management pressure in data centers.

Short term (1-2 years): 1.6T pluggable optical modules will remain the mainstay of market delivery ;

In the medium to long term (2-4 years): CPO optoelectronic co-packaging solutions will gradually be adopted in large-scale intelligent computing centers, reducing the power consumption of interconnected computing clusters.

In short: Pluggable is the present, co-package is the future.

Companies that are now investing in CPO, silicon photonics, and lithium niobate research and development will have a significantly higher chance of receiving local R&D subsidies and demonstration projects.

Ⅲ.Three-tiered policy division of labor ensures dividends are realized at every level.

Many people wonder what the relationship is between national, provincial, and municipal policies. In fact, the division of labor among the three is very clear, forming a complete system of "top-level target industrial chain support for projects that address shortcomings".

1. National Level: Set standards, establish thresholds, and overcome bottlenecks.

The government has introduced strict rules for computing power procurement, set domestic production limits and chip self-sufficiency targets, and established a special fund for the indium phosphide industry, with a focus on constraints and top-level research.

2. Guangdong Province Level: Laying out long-term development paths and addressing weaknesses in the industrial chain (2026-2035)

Optical communication, optical chips, and optical computing will be included in the province's new development track plan. The support program for optical chips will be continued, and R&D rewards will be issued for indium phosphide and optical chip tape-out projects to create demonstration scenarios for domestically produced high-speed optical modules.

3. Shenzhen City Level: Substantial Financial Support for Mass Production Implementation

An action plan for the AI server industry chain has been launched to encourage the mass production of 800G 1.6T servers, with CPO R&D projects receiving subsidies of up to 10 million yuan and a single project R&D funding cap of up to 30 million yuan. The goal is to create a leading local industry chain cluster for "server optical modules and optical chips" nationwide.

Ⅳ.Divergence in the Industry Landscape

Amidst a wave of policy changes, structural differentiation within the industry has begun.

Companies Entering a Period of Growth

A midstream module manufacturer that has completed 800G mass production, successfully ramped up production of 1.6T products, and is forward-lookingly planning the R&D of 3.2T;

An IDM company with the capability to independently develop high-speed EML optical chips, meeting the domestic procurement threshold;

The R&D team is proactively developing new technology routes such as CPO, silicon photonics, and thin-film lithium niobate.

Companies Facing Significant Pressure

Production capacity is concentrated on low-speed modules of 400G and below, while the iteration of high-speed products lags behind.

High-end chips are entirely sourced from overseas, with a lack of alternative domestic supply chains.

There is insufficient investment in the research and development of the next generation of low-power interconnect technology.

At the same time, the market has been divided into two tracks. In the domestic computing power market, domestic manufacturers are gaining an increasingly clear home-field advantage, while overseas markets remain highly competitive on a global scale.The growth path of leading companies, characterized by a "domestic + overseas dual circulation" model, is becoming increasingly clear.

Ⅴ.Finally, let’s discuss the risks behind the policy benefits.

Bonuses don't mean easy wins; we also need to see the hidden boundaries.

1.The 70% domestic content requirement applies only to national-level intelligent computing clusters. Commercially built data centers and overseas orders are not subject to this mandatory rule.

2.Government project incentives and demonstration awards do not equate to guaranteed orders. Whether a technology can be commercialized ultimately depends on its competitiveness in terms of performance, power consumption, and cost.

3.The product iteration speed is very fast, with continuous iterations from 800G to 1.6T to 3.2T. The R&D investment costs remain high, and if the iteration pace falls behind, the company will be left behind by the market.

Conclusion

The rise of optical modules has never been the result of speculation in a single market.

When policy elevates it from an inconspicuous transmission component to a strategic level as the foundation of AI computing power, we see a clear long-term trend: speed upgrades + domestic substitution + low-power optoelectronic interconnect transformation, these three driving forces propel the optical module industry into a new cycle.

In the computing power race, the final competition is not only about GPUs, but also about the speed of the "conveyor belt".

And this computing power conveyor belt is ushering in its own era.

Understanding policies is the key to understanding industry cycles ! Like and save this post to stay updated on the core logic of the optical module industry. Follow us to keep up with the latest trends in the AI optoelectronics industry.

Interactive Topic: When do you think the 1.6T engine will be widely commercially available? Feel free to leave your thoughts in the comments section.

References