The most lucrative sector in AI is found in the underlying hardware.

Original content from Youdianshu · Digital Economy Studio

Author | Uncle You

In 2026, the AI industry has finally shed its image of being all talk and no substance in terms of profitability.

Over the past few years, concepts such as large-scale models, Artificial General Intelligence (AGI), and general intelligence have been heavily promoted, sending capital markets into a frenzy. Yet, when it comes to financial reports, the industry has been mired in a phase of 'burning money for future expectations' due to massive R&D investments, skyrocketing computing costs, and elusive commercialization.

However, this year's annual and first-quarter reports have completely reversed the trend: AI computing power has transitioned from conceptual hype to tangible financial results. The most profitable segment across the industry is not the glamorous large-scale model vendors or diverse application platforms, but the upstream hardware that forms the core of the entire AI ecosystem.

Today, let's delve into what exactly transpired in the AI business over the past year.

A Boom Across the Entire Supply Chain

From optical modules, AI servers, high-speed PCBs, to enterprise storage, chip testing, and liquid cooling, the entire computing hardware supply chain is experiencing a collective surge. Among these, optical modules lead the pack, emerging as the undisputed 'cash cow' of the AI computing era.

The rise in AI hardware is not a fortuitous breakthrough by individual companies but a synchronized high-growth phase across the entire industrial chain and all sub-sectors. Whether it's computing transmission, computing power calculation, or computing support and chip testing, nearly all segments have delivered results far exceeding market expectations, proving with solid profits that the real money in AI is ultimately being made by the underlying hardware.

This is even more evident when examining the performance data of listed companies. As the cornerstone of computing transmission, the optical module sector has taken the lead in this round of performance surge. Global optical module leader InnoLight Technology released its first-quarter performance forecast, projecting a net profit of RMB 850 million to RMB 900 million, a year-on-year surge of 245% to 260%. Its 800G high-speed optical modules maintain the top global market share, with 1.6T high-end products achieving mass delivery and orders scheduled through the third quarter of 2026. The gross margin of high-end products has exceeded 50%.

Yiwei Semiconductor closely follows, with a projected first-quarter net profit increase of 104% to 142%. The gross margin of its 1.6T optical modules reaches 47%, with nearly RMB 30 billion in existing orders covering periods up to the end of 2027. Overseas revenue accounts for 85%, deeply binding major global cloud service providers. Additionally, Dongshan Precision's first-quarter net profit is estimated at RMB 1 billion to RMB 1.15 billion, a year-on-year increase of 119.36% to 152.27%. Its optical module business has successfully penetrated global top-tier clients, becoming the core engine of the company's profit growth.

The storage and chip testing segments have witnessed remarkable growth. Shannong Xinchuang's first-quarter forecast projects a net profit of RMB 1.14 billion to RMB 1.48 billion, a year-on-year increase of 6714% to 8747%, directly surpassing its full-year profit in 2025. Enterprise storage has seen both volume and price increases driven by AI computing demand, becoming the biggest winner in computing support. Demingli's first-quarter profit is estimated at RMB 3.15 billion to RMB 3.65 billion, successfully turning from loss to profit, with storage chip price hikes and strategic reserves driving its profit elasticity. Probe card leader Qiangyi shares' first-quarter net profit is projected at RMB 106 million to RMB 121 million, a year-on-year increase of 654% to 761%. The explosion in AI chip testing demand has directly driven a surge in performance for testing consumables and equipment segments.

Core computing hardware is also thriving across the board. Hygon Information's first-quarter revenue reached RMB 4.034 billion, a year-on-year increase of 68.06%, hitting a single-quarter historical high. Demand for AI servers and computing chips continues to surge, accelerating the localization of domestic computing power. Sugon's first-quarter revenue reached RMB 3.072 billion, a year-on-year increase of 18.80%. Its high-end computing and intelligent computing center businesses are steadily advancing, becoming the core force in domestic computing infrastructure. In addition, leading companies in high-speed PCBs, optical devices, connectors, and other sub-sectors have generally achieved profit growth of over 50%, with the entire upstream hardware chain showing a high-growth state of full production, full sales, and supply shortages.

The core logic behind this collective surge is quite straightforward: The prerequisite for all AI applications to materialize is to first establish the computing infrastructure. Large-scale model training, multimodal interactions, industry reasoning, and intelligent agent operations all rely on massive, high-speed, and stable computing support.

Global cloud service providers are frantically expanding AI data centers, and domestic intelligent computing centers are densely deploying. The first step in computing infrastructure is to equip with underlying hardware such as optical modules, servers, storage, cooling, and testing equipment.

They are the 'water sellers' and 'shovel sellers' of the AI era. No matter which large-scale model prevails or which application becomes a hit in the future, these essential hardware components are indispensable. They are also the earliest segments in the entire industrial chain to deliver financial performance, with the most certainty and resistance to volatility.

The True Cash Cow in AI Computing

Among the booming AI hardware tracks, optical modules stand out the most. Against the backdrop of overall market volatility, the sector remains robust, with leading stocks repeatedly hitting new highs and capital pouring in, becoming the most solid main theme in the entire market. Many investors wonder: How can a communication component become the core of the AI computing market? How can it continue to 'print money'?

The logic is actually quite simple and becomes clear at a glance:

On one hand, large-scale model vendors are still competing for users, traffic, and commercialization, with massive R&D investments, high computing costs, and unclear profit paths. On the other hand, optical module companies are overwhelmed with orders, experiencing both volume and price increases, doubling their performance, and enjoying exceptionally solid cash flows. One is still burning money to expand, while the other has already achieved stable profits. The gap between the two is stark.

The core position of optical modules comes from the underlying necessity of AI computing. To achieve high-speed transmission of massive data within AI servers and between data centers, high-speed optical modules are essential.

Simply put, without optical modules, even the most powerful AI chips and advanced large-scale models would remain isolated information islands, unable to complete data interactions and model operations.

As AI model parameters continue to expand and multimodal interactions become fully popularized, data transmission volumes grow exponentially, driving up the requirements for the speed, quantity, and performance of optical modules. From 800G to 1.6T, technological iterations continue to accelerate, with high-end products maintaining firm prices and high gross margins, directly igniting industry dividends.

The business model of optical modules perfectly aligns with capital market preferences.

First, demand is rigid. Every expansion in AI computing power increases the demand for optical modules, making them essential infrastructure with no alternatives. Second, technological barriers are high. High-speed optical modules require multiple core accumulations in optical chips, packaging processes, algorithm design, and reliability verification, making it difficult for small and medium-sized manufacturers to enter and solidifying the advantages of leading companies. Third, customer barriers are deep. Once entering the supply chain of global cloud service providers, orders become scalable and sustained, with clients unlikely to switch easily. Fourth, both volume and price increase. Iterations in high-end products drive up unit prices, while economies of scale reduce production costs, continuously improving gross margins.

Capital's attitude is always the most honest. Large amounts of capital have withdrawn from other tech sectors and continue to flow into the optical module track, with institutions and northbound funds simultaneously increasing their positions in leading stocks, forming a strong market consensus. The underlying logic is simple: Among the entire AI industrial chain, optical modules are one of the earliest segments to deliver financial performance, with the highest certainty, best cash flow, and deepest barriers. Compared to the high uncertainty of large-scale model commercialization, the returns on optical modules are more predictable, stable, and sustained, making them the 'core assets' of the AI market.

Many investors always want to bet on the next super large-scale model but overlook the simplest truth in the gold rush: The most stable profits never come from the gold diggers relying on luck but from the shovel and water sellers. Optical modules are the most solid 'water sellers' in the AI computing era. No matter how the industry landscape changes in the future, they will continue to share in the growth dividends of AI.

Has the High Growth of the AI Track Peaked?

With the collective surge in AI hardware performance and related company stock prices repeatedly hitting new highs, many investors wonder: Has the high growth of the AI track peaked?

Judging from research reports by mainstream securities firms, AI high growth is far from peaking and is still in its golden stage of long-term growth.

Securities firms generally believe that the demand for AI computing power is not a short-term pulse but a long-term necessity. Currently, global cloud service providers are still frantically expanding AI data centers, and domestic intelligent computing centers are densely deploying, with a significant gap in computing infrastructure.

Taking the core optical module sector as an example, Huatai Securities predicts that the global optical module market growth rate can be maintained at around 60% this year. Moreover, technological iterations are accelerating, with 1.6T optical modules just entering mass production and 3.2T products already in preparation, expected to begin validation in 2027-2028. The trend of both volume and price increases will not change.

Soochow Securities points out that the AI computing power industrial chain will encounter multiple opportunities in 2026. In terms of cloud computing power, domestic GPUs are entering a period of financial performance delivery, and the localization of domestic computing power will continue to accelerate. Edge-side AI is also taking over, with scenarios like glasses, cars, and robots accelerating their deployment, driving new demand explosions.

Besides optical modules, securities firms are also bullish on multiple sub-sectors. For example, AI storage will see a surge in high-bandwidth memory this year as edge-side AI volumes increase, becoming a new growth highlight. Rings like PCBs, optical-copper interconnects, and server power supplies will also continue to benefit from computing demand.

Of course, securities firms also caution that short-term volatility may occur, such as valuation corrections and technological route debates. However, in the long run, the growth space for the AI industry remains vast.

Simply put, AI has only completed the first step of transitioning from 'hype' to 'real money.' In the future, as computing demand continues to release and technology keeps iterating, the high growth of the entire track will persist, with clear opportunities in core hardware.

THE END

This article is an original piece from Youdianshu View and cannot be reproduced in any form without permission.