Preface

After the market close, China’s Ministry of Industry and Information Technology (MIIT) released a sweeping development plan for the information and communications sector that sent ripples through technology and telecom stocks. This article summarizes the key policy measures, market reactions, and the accompanying industry forecasts from global investment bank Goldman Sachs. It places these announcements in the broader context of accelerating AI compute demand — driven by large models and new generations of GPUs — and explains why optical modules, fiber infrastructure, and system-level compute deployments stand to benefit. The goal is to provide a clear, objective account of the policy content, market implications, and near-term drivers investors and industry participants should watch.

Lazy bag

MIIT’s new “15th Five-Year” communications plan calls for nationwide integrated compute-network infrastructure and fast optical transmission upgrades (400G+), while Goldman sharply raised global optical-module market forecasts through 2028. The market reacted with broad gains across optical supply chains, and developments such as OpenAI’s GPT-6 Astra and NVIDIA’s GPU expansion underscore ongoing demand for high-density, high-speed interconnects.

Main Body

The Ministry of Industry and Information Technology (MIIT) has issued an updated development plan for the information and communications industry that outlines a comprehensive strategy to upgrade national communications infrastructure and the compute ecosystem through 2030. The plan emphasizes a layered, integrated approach to compute facilities — described as a "hub–regional–edge" multi-tier architecture — and calls for large-scale, orderly deployment of intelligent-compute clusters in the ten-thousand- to hundred-thousand-card range. Specific transport-focused mandates include accelerating deployment of 400G and higher-speed transmission technologies, building fiber direct-access links, and promoting the wider use of 100G optical network equipment at deeper network layers.

These policy directions respond to a clear structural shift in compute and networking: the rapid expansion of AI model training and inference workloads that require significant on-premises and regional compute capacity as well as high-bandwidth, low-latency interconnects. The MIIT plan also foregrounds standards, monitoring and automation for compute resource management, adaptation to domestic chip solutions where applicable, and measures to improve power coordination and green, low-carbon compute facilities. In short, the plan seeks to create an industrial and operational foundation that supports both the performance and sustainability needs of next-generation AI deployments.

Market reaction was swift. Chinese A-share segments tied to the compute and optical value chain — including cloud and data-center equipment, optical modules, fiber and cable, and PCB manufacturers — recorded strong gains following the announcement, with many stocks hitting daily limits or rising sharply. The immediate equity response reflects investor expectations of accelerated capital expenditure across cloud operators, data-center builders, and telecommunications carriers to meet the upgraded capacity and transmission requirements set out in the MIIT plan.

Adding to the optimism, Goldman Sachs released a deep industry report revising its global optical-module market forecasts substantially upward for 2026–2028. The bank raised total-addressable-market (TAM) estimates by 33% for 2026, 81% for 2027, and 115% for 2028 relative to its prior baseline — a clear signal that investment banks see far larger demand for optical interconnects over the medium term. Goldman’s analysis points to several structural drivers: faster adoption of 800G and above modules, strong growth in rack-level and ASIC-based AI servers, and an increase in the number of optical modules used per NVIDIA GPU as systems scale up.

Goldman estimates explosive growth in the 800G+ segment, forecasting a compound annual growth rate of roughly 69% and projecting the 800G+ market to grow from about $45.2 billion in 2026 to some $129.5 billion by 2028. The bank also pushed up unit shipment forecasts across multiple speed tiers — notably 800G, 1.6T, and emerging 3.2T modules — and highlighted silicon photonics as a clear area of increasing penetration, especially at the highest data rates.

Concurrently, OpenAI’s public release of GPT-6 Astra and NVIDIA CEO Jensen Huang’s remarks about the arrival of an AGI-era have captured global attention. OpenAI reported training Astra on well over 100,000 GPUs, and Huang noted that another batch of some 400,000 GPUs is coming online — a comment that underscores the growing scale of compute infrastructure required for frontier AI. These developments serve as a practical validation of the MIIT plan’s urgency and of Goldman’s assumption of rising demand for high-performance optical modules and related networking gear.

From a technology-supply standpoint, several supply-chain dynamics deserve attention. First, higher-speed optics (800G and above) tend to favor silicon photonics and advanced integration techniques because of their power, size, and performance profiles; Goldman expects silicon-photonics penetration to increase across the highest-speed categories. Second, the growth of rack-level AI servers and the move toward ASIC-based AI acceleration (in addition to GPU-heavy architectures) will alter module counts per system and the mix of speed tiers deployed. Third, densification and the push for low-latency regional compute capacity imply growing demand for fiber-to-facility builds, local high-speed switching, and compute interconnect optimization.

Policy measures that stress standards, automated monitoring, and resource matching can materially improve utilization and reduce waste in large compute deployments if implemented effectively. Similarly, incentives for green power integration and coordinated grid–compute strategies will be essential for curbing the carbon footprint of rapid compute expansion. The MIIT plan’s emphasis on coordinated power-and-compute design (including green-power direct connections, source–grid–load–storage approaches, and efficient storage) is therefore an important signal that sustainability and resilience are core policy goals, not afterthoughts.

For investors and industry participants, the near-term watch list includes capital-spending plans from cloud providers and telecom carriers, procurement cycles for new optical modules and switches, announcements from major hyperscalers on regional facility builds, and the continued rollout of higher-speed optics from module vendors. Equally important are supply-side constraints: capacity to scale silicon-photonics production, high-volume manufacturing for new optical form factors, and availability of high-quality optical fiber and cable infrastructure will determine how quickly demand translates into revenue for vendors.

In conclusion, MIIT’s plan, Goldman’s upgraded forecasts, and the latest frontier-AI developments together form a coherent narrative: AI-driven compute demand is accelerating the need for high-capacity, low-latency networking and large-scale compute clusters. This will likely lift demand for optical modules, fiber infrastructure, and system-level networking equipment in the medium term, while also elevating the importance of standards, automation, and sustainability in large compute deployments.

Key Insights Table