The chart tells the story faster than any analyst note. China’s optical fiber prices sat at ₹15 per fiber-kilometer at their trough in early 2024. They are at ₹50 per fiber-kilometer today — a 3.3x increase in 18 months — and the specialized bend-insensitive fiber used in AI data centers and drones has gone parabolic, hitting ₹65 per fiber-kilometer with no visible ceiling.
This is not a commodity cycle. It is a structural demand shock, and it has a specific cause: every GPU rack built for AI inference and training requires 16 to 36 times more optical fiber than a traditional CPU server setup. AI data center demand for fiber surged 76% year on year in 2025. The hyperscalers — Microsoft, Google, Amazon, Meta — are building at a pace that the global fiber supply chain was never designed to support.
The result is a shortage that is showing up in prices, in order books, and in the earnings of the few companies globally that can actually manufacture the product at scale. Three of those companies are Indian and rightly pointed by Abhay Jain on X, these 3 companies are sitting on gold mine.
Sterlite Technologies: The Only Indian Player With Full Preform-to-Cable Integration
Sterlite Technologies — STL — is a top-three optical fiber company globally outside China with roughly 8% market share. That positioning matters enormously right now because the fiber shortage is partly a preform shortage — preform being the glass rod from which optical fiber is drawn, and the most technically demanding and capacity-constrained step in the entire manufacturing chain.
Most fiber manufacturers buy preform and draw fiber. STL makes its own preform. That vertical integration means STL is not competing for constrained raw material in a tight market — it controls its own supply from the beginning of the process to the finished cable. In a shortage environment, that is the difference between being able to fulfill orders and turning customers away.
The AI data center angle is direct. GPU-dense AI racks require STL’s IBR and 160-micron fiber products — thinner, higher-density cables that pack more fiber into smaller spaces, exactly what hyperscale data centers demand. STL has already booked ₹500 crore in AI and data center orders and is targeting data center revenue at 30% of total within 12 to 18 months, up from 20% currently.
HFCL: Turning Away Orders Because It Cannot Make Enough
HFCL manufactures high fiber-count cables — up to 6,912 fibers per cable — for hyperscale AI data centers globally. It is one of a handful of companies worldwide with the machinery to produce cables at that fiber density, which is precisely what the largest AI infrastructure buildouts require.
The most telling data point about HFCL right now is not its revenue or its order book. It is the fact that the company is currently refusing orders due to capacity constraints. In manufacturing, turning away customers is either a crisis or a sign of extraordinary pricing power. In HFCL’s case it is the latter — demand is so far ahead of supply that the company cannot build capacity fast enough to keep up.
It is building faster. OFC manufacturing capacity is expanding from 30.5 million fiber-kilometers to 42.36 million fiber-kilometers by June 2026 — a 39% capacity addition in a matter of months. With that expansion, HFCL is targeting ₹3,500 crore in OFC revenue in FY27. The constraint is not demand. It has never been demand. The constraint is how fast the machines can be installed and commissioned.
Finolex Cables: The Margin Expansion Story
Finolex Cables is the third leg of this trade and arguably the most interesting from a pure margin expansion standpoint. OFC volumes at Finolex grew approximately 33% in Q3, riding the same wave of data center and telecom demand driving STL and HFCL. But the more significant number is what is happening to pricing.
Global fiber prices have hardened from $3 to $5 per fiber-kilometer — a 67% increase — driven entirely by data center demand. Finolex’s OFC EBIT margins currently sit at approximately 2.5%. The company is targeting 8 to 9% as preform integration and scale begin to flow through the income statement — a three to four times margin improvement from current levels.
The capacity story mirrors HFCL: Finolex is doubling its fiber draw capacity from 4 million to 8 million kilometers by Q1 FY27, targeting OFC revenue of ₹600 to ₹700 crore once the expanded capacity is fully operational. The combination of higher volumes, higher prices, and expanding margins compressing into the same 12-month window creates an earnings profile that current valuations may not yet fully reflect.
Why This Is Different From Previous Fiber Cycles
India’s optical fiber companies have been through boom-and-bust cycles before — typically driven by government broadband rollout programs that start, pause, restart and disappoint. BharatNet created optimism and then frustration. 5G rollout demand came and then plateaued. Investors in this sector have been burned by demand that evaporated.
The AI data center demand cycle is structurally different in three ways. First, it is driven by private capital — hyperscalers spending hundreds of billions of dollars on infrastructure with multi-year commitments — rather than government programs subject to budget cycles and policy changes. Second, the physics of AI infrastructure means fiber demand scales with compute, and compute is growing exponentially. Third, the preform constraint means supply cannot respond quickly even when prices spike, which is why prices have tripled over 18 months rather than normalizing after six.
The chart at the top of this story shows a parabolic spike that looks like it might correct. It might. But the underlying demand that caused it — every new GPU rack requiring 36 times more fiber than what it replaced — does not go away when prices pull back. It just means the next wave of orders comes at a higher base.
STL, HFCL and Finolex are three different ways to own the same thesis: the world needs far more optical fiber than it is currently producing, prices reflect that imbalance, and Indian manufacturers with the technical capability to serve hyperscale AI data centers are among the few global players positioned to fill the gap.
This article is for informational purposes only.