Orders are booked through 2027 with only 730 employees, signaling that computing infrastructure is moving away from labor-intensive models toward capital intensity and process barriers.
While the entire capital market is frenzied over the new generation of computing-power GPU architectures, and even footwear, apparel, and real estate companies are trying to pour money into intelligent computing centers, the underlying consumables propping up this grand industrial movement toward AI-driven manufacturing are quietly operating under a brutally capital-intensive logic.
The offhand remark by a senior executive at Jiangsu Hengtong Optical Fiber Technology Co., Ltd., that "orders are booked through 2027," could easily be simmered by the public discourse into a cheap chicken-soup-for-the-soul story about getting rich on the AI wave. Outsiders habitually assume that since large-model training and industrial-brain deployments demand massive data throughput, fiber optic factories should, like the assembly workshops of yesteryear, be hiring frantically, working around the clock, and reaping enviable excess profits from a demographic dividend.
However, merely piercing the surface of this grand narrative and looking at the underlying labor logic at this company reveals just how suffocating the real barriers to entry are. Tracing back through the industrial and commercial and social security records in the Tianyancha system, this manufacturing giant, which has taken on massive orders from computing-power infrastructure projects, reported a social security contribution headcount of just 730 people in 2025. Facing a backlog of orders stretching three years out, its headcount expansion over the entire past year was held extraordinarily restrained at just 78 additional employees.
In the large-model and embodied-intelligence track, where tens of thousands are needed to clean data and thousands to debug robotic arms, a mere 730 flesh-and-blood workers are gripping the data lifeline of the entire computing-power supply chain.
This extremely abnormal productivity ratio lays bare the ruthless truth of the specialty fiber optic manufacturing sector. The fiber optics required in computing-power data centers are no longer the cheap, high-volume commodity cables used to roll out home broadband years ago. To achieve near-zero-latency, aligned transmission at trillion-parameter scale between densely packed liquid-cooled racks requires ultra-low-loss single-mode, multimode, and specialty optoelectronic coupling components. This is not a capacity issue that can be solved by stacking piece-rate workers; it is a heavy-industry game that depends heavily on precision drawing towers, high-purity preforms, and fully automated clean-room facilities.
Those 78 new hires are most likely not cogs on an assembly line, but rather process engineers and system inspectors tasked with maintaining those multi-million-dollar heavy automation systems.
If we continue to unpack the underlying structure of this established company in the Tianyancha records, the interplay of interests behind it becomes even more sophisticated. Beyond Hengtong Optic-Electric, the massive listed parent entity, its joint-venture roster prominently lists Fujikura Ltd. of Japan. In the deep-water territory of fiber optic preforms and specialty optoelectronic components, long monopolized by overseas giants, the legal framework of a Sino-foreign joint venture serves not only to absorb the core process know-how of a long-established Japanese titan, but also to ensure, in the coming global battle for computing infrastructure, that its most critical supply chain is not easily severed by geopolitical frictions.
The war of computing centers, on the surface, is about chip inventories; in reality, it is about the efficiency of moving data through the physical world. Without this layer of specialty glass-fiber mesh buried underground, even the smartest industrial brain is just a lonely, overheating metal box.
The business world operates on no faith in empty frenzy. In this cycle where true bandwidth, extreme asset depreciation, and process yield ultimately determine survival, shell companies that merely trade on hype and resell cloud interfaces will eventually face liquidation under the weight of physical laws. The starkly sparse social security roster that Hengtong Optical Fiber leaves in the Tianyancha system reads as a sober verdict for the AI era: in the coming years, the most core wealth will not flow to arbitrageurs spinning fantasy narratives, but will be extracted and consumed—in the heaviest yet most stable way—by these hardcore manufacturing oligarchs standing guard over high-temperature furnaces in their workshops, drawing glass into fiber.
