Fiber Optic Demand Surges Due to AI Data Center Boom
The construction of AI data centers has led to a fiber optic shortage: orders are booked through 2027. Their share of consumption is expected to grow from 5% to 30% by 2027.
Fiber Optics — The New Oil: Why the AI Boom Has Caused a Glass Shortage That No One Talks About
When the market discusses multi-billion dollar GPU deals, I look at the infrastructure one level below. What is happening now with fiber optics is not a temporary supply-demand imbalance, but a fundamental shift in the physical architecture of AI. The glass thread is becoming a strategic resource, and this will change the game faster than anyone expects.
The Core: What Is Really Happening
Headlines scream about chip shortages, but the real structural crisis is unfolding in the fiber optic segment. Data from the last 48 hours paints a picture I would call "quiet panic": in 2025, demand for optical fiber from data centers grew 76% year-over-year, and by 2027 this category will account for 30% of global consumption — up from less than 5% in 2024. This is not gradual growth, but a sixfold increase in market share in three years.
The price of standard fiber has jumped from $3.70 per kilometer in 2021 to $6.30 now — a 70% increase. In China, which remains a key supplier for many markets, prices have risen 2.5 to 4 times in just the last six months. Lead times for large buyers have stretched to 20 weeks, while small customers wait a year. Corning's factories are running at full capacity, and new preform production lines — a key component — take 18–24 months to ramp up.
Nvidia and Corning announced a partnership on May 6, 2026: Corning will increase its US optical connectivity capacity tenfold, build three new plants in North Carolina and Texas, and create over 3,000 jobs. Jensen Huang called it "the largest infrastructure build of our time." But even these paces are insufficient — demand is growing faster than production can physically be expanded.
Timeline and Context
The events leading to the current crisis have unfolded at a cascading speed:
- 2021: Fiber optic prices at an all-time low of $3.70/km. Market oversaturated, producers cut investments.
- 2023–2024: Launch of ChatGPT and explosive growth of AI training clusters. One cluster with 1,000 GPUs requires 36 times more fiber than a traditional server rack.
- January 2025: Meta signs a $6 billion contract with Corning — just for fiber optics for AI data centers. One Meta facility in Louisiana with 5 GW capacity will require 13 million km of fiber.
- 2025: Corning's optical segment revenue reaches $16.4 billion, up 13%. The company raises its incremental revenue forecast for 2028 from $8 billion to $11 billion.
- Q1 2026: Corning reports 36% revenue growth in Optical Communications to $1.8 billion, segment operating profit surges 93% to $387 million. Margins expand by 220 basis points.
- May 6, 2026: Nvidia–Corning partnership with a $500 million investment, aiming for a tenfold expansion of US optical factories.
- May 18–19, 2026: CRU data shows that by 2027, AI data centers will consume 30% of global fiber. Suppliers at industry conferences confirm: the shortage will last at least until mid-2027.
Who Wins and Who Loses
Winner: Corning. The company with a 10.4% market share becomes an indispensable player in AI infrastructure. Its contract portfolio includes Meta at $6 billion and at least two unnamed hyperscalers with similar terms — guaranteed minimum commitments and prepayment for capacity. The Springboard target revenue is raised to $40 billion by 2030, implying a 19% CAGR. The market has recognized this: GLW shares have risen for eight consecutive quarters.
Winner: Nvidia — strategically. The $500 million investment in Corning's plants is not charity. Nvidia gains priority access to a critical resource that limits the deployment of its own GPUs. Huang understands: without fiber optics, clusters of 100,000 GPUs are inefficient. Control over the optical supply chain becomes as important as control over HBM memory supply.
Losers: Telecom operators. Manufacturers are shifting capacity from telecom fiber to higher-margin data center fiber. Traditional customers — telecom operators — face shortages and rising prices for standard G.652D fiber.
Losers: Small AI companies. Small buyers wait up to a year for deliveries. This means startups planning to build their own training clusters physically cannot obtain infrastructure. The concentration of AI power in the hands of hyperscalers intensifies not only due to GPU prices but also due to access to fiber optics.
What the Media Isn't Saying
Non-obvious insight: The fiber optic shortage will accelerate the adoption of CPO and silicon photonics, but not in the way everyone thinks.
Industry analysts at Bernstein point out: the problem for AI data centers is increasingly shifting from GPU/HBM/CoWoS to "interconnect systems." The bottleneck is not computing power but the speed of data exchange between GPUs. With tensor parallelism and expert parallelism, the volume of inter-node traffic is astronomical, and latency on optical connections eats away the gains from additional accelerators.
CPO (co-packaged optics) technology promises 70% energy savings compared to traditional optical modules. Nvidia has already introduced Spectrum-X Photonics — silicon photonic switches with 3.5 times better energy efficiency and 63 times better signal integrity.
But here's what is missed: the shortage of traditional fiber paradoxically slows down CPO adoption. The industry first needs to build factories for existing technology, and then for the next generation. Corning is investing billions in expanding standard preform production because that is what is needed right now. R&D investments in hollow-core fiber, demonstrated at the Wuhan Optical Expo, remain niche — commercial deployment on backbone lines is not expected for at least five years.
Thus, the shortage of traditional fiber creates a dual pressure: it simultaneously requires expanding old capacity and diverts resources from developing new technologies. This is a classic innovator's dilemma at the physical infrastructure level.
Second hidden story: The geopolitical dimension. The second largest driver of fiber optic demand is military applications. FPV drones with fiber optic control consume 10–20% of the global market. This is not just a statistical curiosity: military procurement operates outside market mechanisms, with priority access to supply and non-market pricing. Civilian AI projects compete for glass with the defense sector, and this competition is only intensifying.
Third point: Corning is no longer a cyclical supplier but an AI infrastructure company. Guaranteed contracts with prepayment for capacity change the very nature of the business. Previously, optical fiber was a commodity market with volatility; now it is a contract market with long-term predictability. This is a structural shift that analysts have not yet fully priced in.
Forecast: 30 Days and 90 Days
30 days (until mid-June 2026):
- Corning will announce a fourth major hyperscaler contract — I expect it to be Oracle, which is aggressively expanding its AI cloud. GLW shares will gain 8–10% on this news.
- Fiber prices will reach $6.80–7.00/km by end of June — another 8–10% increase. Manufacturers will start requiring prepayment for capacity reservations for 2027.
- Nvidia and Corning may announce a second round of investment in new plants — another $700–800 million, aiming for a 15x capacity expansion instead of the original 10x. Jensen Huang will use Nvidia's upcoming earnings report on May 20 to announce the expanded partnership.
90 days (until mid-August 2026):
- The shortage will begin to affect even hyperscalers. One of the FAANG companies will report a quarter delay in data center commissioning due to lack of fiber optic infrastructure. The market will perceive this as a negative signal for the entire sector.
- Startups in optical interconnects — Ayar Labs, Lightmatter, Celestial AI — will see a wave of interest from venture investors. Valuations of companies working on CPO and silicon photonics will rise 2–3 times over the summer. At least two major rounds of $200–400 million each will occur.
- Corning will raise its annual revenue forecast in the July Q2 report. Optical Communications will show growth above 40% year-over-year. GLW shares will reach an all-time high.
Key signal: if within 90 days reports emerge of fiber optic disruptions for AI clusters under construction in Asia (especially Japan and South Korea), this will confirm that the shortage has become global. Then the entire industry will begin reassessing the AI infrastructure deployment timeline, and analyst forecasts for computing power growth will be revised downward.
The glass we have perceived as a commodity for decades is becoming as strategic a resource for AI as uranium is for nuclear power. Whoever controls the preform factories controls the speed of AI deployment for the next decade. Corning understands this. Nvidia understands this. The question is when the rest will understand.
— Editorial Team
No comments yet.