AI Bought All the Optics: Could Your 2027 Network Upgrade Get Stuck Waiting for Transceivers?
Hyperscalers are locking up laser and optical-component capacity years in advance. 800G and 1.6T demand is exploding. Manufacturers are building new fabs to catch up. Does that mean the ordinary company planning a 100G, 400G or data-center upgrade in 2027 could find the switches installed — but the optics missing?
“AI bought all the optics” is obviously an exaggeration. But underneath the joke is one of the strangest supply-chain stories in networking right now. The companies building enormous AI clusters are no longer simply ordering transceivers when they need them. They are committing billions of dollars, signing long-term supply agreements and securing access to future manufacturing capacity for the lasers and optical components inside those transceivers. For an enterprise buyer, that creates a new question: when the biggest technology companies in the world reserve capacity years ahead, what happens to everyone else trying to upgrade a network next year?
What Happened? AI Turned Optics Into Strategic Infrastructure
GPUs get the headlines, but hundreds of thousands of accelerators are useless if they cannot move data between servers, switches, racks, buildings and campuses.
TrendForce estimates that the market for AI-focused optical transceivers will grow from roughly $16.5 billion in 2025 to approximately $26 billion in 2026. The research firm specifically identifies tight supplies of electro-absorption modulated lasers (EMLs) and continuous-wave lasers as constraints on production.
LightCounting’s estimates are similarly aggressive. In July it projected Ethernet optical-transceiver sales growing roughly 73% in 2026 and warned that shortages of indium-phosphide lasers could stretch into the middle of 2027.
The Number That Should Get Procurement’s Attention: 2029
In October 2026, Lumentum CEO Michael Hurlston said overwhelming AI demand had effectively committed much of the company’s advanced optical-component capacity well into 2029. Reports indicated significant gaps between demand and available production for certain products even as the company expands manufacturing.
The NVIDIA agreements are especially revealing. They do not merely involve buying finished transceivers. NVIDIA’s agreements with Lumentum and Coherent include multibillion-dollar purchase commitments and rights to future manufacturing capacity for advanced laser and optical-networking products.
That is what a strategic supply constraint looks like.
It does not mean NVIDIA owns all industry production. Both agreements were announced as nonexclusive. It does show how valuable guaranteed future optical capacity has become.
Why Can’t Suppliers Just Make More Lasers?
An optical transceiver may look like a small metal module that slides into a switch, but its supply chain is anything but simple.
Specialized Laser Capacity
High-speed modules rely on specialized EML, CW and other photonic devices. Indium-phosphide manufacturing capacity cannot be created overnight like adding another assembly shift.
Precision Packaging
At extreme speeds, optical alignment, packaging, testing and yield become increasingly demanding. TrendForce identifies high-precision alignment itself as another constraint on scalable production.
Qualification Takes Time
New fabs and suppliers do not instantly become interchangeable sources. Components must meet performance, reliability and customer-qualification requirements before large buyers can confidently deploy them.
So Is Your Ordinary Enterprise Optic Actually at Risk?
This is where the headline needs nuance. The shortage is not uniform across every speed and every transceiver design.
| Speed / Type | 2027 Supply Risk | AI Competition | Practical Buyer View |
|---|---|---|---|
| 1G / 10G | Low | Low | Mature supply chain with numerous suppliers. EOL or vendor-specific coding may create more risk than AI demand. |
| 25G | Low | Low | Mature enterprise and server technology. Broad AI optics demand does not automatically imply a 25G shortage. |
| 100G | Low–Moderate | Moderate | Record unit shipments show the ecosystem is producing huge volumes. Specific reach, vendor qualification or platform SKUs can still become constrained. |
| 400G Ethernet | Moderate | High | Increasingly mainstream but also heavily used in data-center and AI networks. Particular DR, FR, LR and vendor-coded modules deserve advance planning. |
| 400ZR / ZR+ | Moderate–High | High | Cloud DCI demand adds competition for coherent technology and high-speed component supply. |
| 800G | High | Very High | Core AI-fabric technology. Do not assume modules can be ordered shortly before deployment. |
| 1.6T | Very High | Extreme | Rapid production ramp, new component requirements and hyperscaler demand make early sourcing essential. |
Why 100G and 400G Buyers Should Still Pay Attention
The most tempting mistake is to say: “We aren’t building an AI cluster, so none of this applies to us.”
The issue is that enterprise and AI products do not live in completely isolated factories. Different modules can share upstream suppliers, laser technologies, packaging resources, testing equipment, substrate suppliers and engineering attention.
A manufacturer facing enormous demand for its newest, highest-value products has an economic incentive to allocate capital and production capacity toward those products. That does not automatically create a shortage in older modules, but it can change priorities throughout the supply chain.
Record Shipments Are Actually Good News
The supply story is not simply “everything is disappearing.” Manufacturers are shipping more optics than ever.
LightCounting reported that second-quarter 2026 optical-transceiver sales reached another record. Unit shipments of both 100G and 400G Ethernet transceivers reached new highs, while shipments of several 800G products rose as much as roughly threefold year over year.
This means the market is responding. Suppliers are adding capacity, increasing yields, qualifying additional sources and investing aggressively.
The Bull Case for Supply
- New InP manufacturing capacity is being constructed.
- Coherent and Lumentum are expanding production.
- Module suppliers are scaling 800G and 1.6T rapidly.
- Silicon photonics is broadening manufacturing options.
- Higher production volumes should eventually improve unit economics.
The Risk Case
- Hyperscaler demand continues to outrun forecasts.
- Strategic customers are reserving capacity years ahead.
- New fabs require time to qualify and ramp.
- Advanced modules use increasingly demanding components.
- Double ordering can distort the real level of demand.
The Surprise Risk: A Shortage Can Turn Into a Glut
Buyers should not assume today’s scarcity will continue forever.
LightCounting has specifically warned about double ordering. When customers fear shortages, they place orders with multiple suppliers or reserve more components than they ultimately need. Backlogs become inflated. Manufacturers then add capacity to satisfy demand that may partly exist twice on paper.
If AI capital-spending growth eventually slows while new optics factories are coming online, the market can move surprisingly quickly from shortage to excess inventory.
Lumentum Is Building More Capacity — But Fabs Do Not Appear Overnight
In March 2026, Lumentum announced a new 240,000-square-foot manufacturing facility in Greensboro, North Carolina. The site is intended to increase U.S. production of advanced indium-phosphide optical devices, including CW and ultra-high-power lasers used in next-generation AI infrastructure.
Coherent is also directing capital toward expanded manufacturing. Its fiscal-2026 results described exceptional demand entering fiscal 2027 and continued investment in additional production capacity.
These expansions are important because they should eventually relieve parts of the bottleneck. But expanding a photonics supply chain requires equipment installation, process development, yield improvement, qualification and customer acceptance.
The Biggest Enterprise Mistake: Buying the Switch First
Imagine a company ordering a new pair of data-center switches with dozens of 400G ports. The switches arrive. Installation weekend is scheduled. Fiber has been tested. Old equipment is ready to be retired.
Then procurement discovers that the exact vendor-qualified LR4 or DR4 modules needed for production have a longer delivery window than the switches.
The expensive chassis becomes a very sophisticated rack decoration.
A 2027 Upgrade Should Start With an Optics Audit
Include both ends of every link, breakout configurations, inter-switch links, uplinks and DCI connections.
Do not design an expensive network where a failed optic leaves you waiting months for a replacement.
“400G optic” is not specific enough. Record reach, fiber type, connector, wavelength specification and form factor.
Determine whether the platform requires OEM-coded optics or supports qualified alternatives.
A switch delivery commitment is not useful if the required transceivers have a different shipment date.
“Normally in stock” is not a delivery commitment. Obtain dates for the actual quantities and part numbers.
Compatibility testing is easier during project planning than during an outage.
Moving from 100G to 400G or 800G can alter connector, lane, reach and fiber requirements.
OEM Optics vs Qualified Compatibles Gets More Interesting in a Shortage
OEM-Only Strategy
- Simpler support and warranty conversations.
- Clear vendor qualification path.
- Consistent firmware and telemetry expectations.
- Potentially fewer sourcing options during allocation.
- Often higher acquisition cost.
Qualified Multi-Vendor Strategy
- Creates additional sourcing options.
- May reduce replacement and spare costs.
- Can shorten exposure to a single supplier’s backlog.
- Requires genuine interoperability testing.
- Support policies must be checked before deployment.
Watch the Exact Module, Not Just the Speed
| Requirement | Why It Changes Supply Risk |
|---|---|
| SR / short reach | Technology and fiber requirements differ from long-reach single-mode products. Availability of one does not guarantee availability of another. |
| DR / parallel single mode | Heavily relevant to modern high-density AI and data-center fabrics. |
| FR / LR | Longer reach and wavelength multiplexing change both optical design and component requirements. |
| 400ZR / ZR+ | Coherent DSP and laser requirements put these in a different supply category from ordinary client Ethernet optics. |
| OSFP vs QSFP-DD | A module in the wrong form factor is not an emergency substitute. |
| Vendor coding | A physically identical-looking optic may not be supported or recognized by every platform. |
Five Warning Signs Your Project Is Exposed
One Exact Supplier
The design depends on one exact OEM part number with no qualified alternative.
400G+
The project is concentrated in the same high-speed ecosystem currently being pulled hardest by AI infrastructure.
Just-in-Time Purchasing
Optics are scheduled to be ordered only a few weeks before installation.
Large Port Count
A module that is easy to find in quantities of two may be much harder to commit in quantities of 128.
No Spare Pool
Every purchased transceiver is needed for production with no stock available for failures.
Switch and Optics Separate
Hardware procurement has confirmed switch availability but has never checked the full optical bill of materials.
What I Would Do for a 2027 Enterprise Upgrade
| Project | Suggested Procurement Posture |
|---|---|
| 10G / 25G refresh | Normal procurement discipline. Check EOL status and platform compatibility, but there is little reason for panic buying based solely on AI demand. |
| 100G enterprise core | Quote early, qualify alternates and hold reasonable spares. Current production volumes remain very large. |
| 400G core or data center | Treat optics as critical-path equipment. Obtain committed dates on the exact modules before finalizing implementation timing. |
| 400ZR / DCI | Source alongside routers and line systems. Verify coherent-module availability, software support and route engineering early. |
| 800G deployment | Engage suppliers well ahead of installation. Build an approved-component matrix and avoid assuming spot availability. |
| 1.6T early deployment | Procurement should participate during architecture design. Availability and qualification may influence which platform you select. |
Supply conditions vary substantially by manufacturer, form factor, reach, platform, contract and order quantity. References to industry shortages describe market conditions, not guaranteed lead times for any specific transceiver SKU.
2027 Optics Procurement Risk Checker
Enter the basics of a planned network upgrade. This does not predict whether a particular module will be unavailable; it highlights projects whose speed, volume, timing and single-source dependence justify earlier procurement.
The risk score is a planning heuristic, not a market forecast or inventory check. Actual availability must be confirmed with suppliers using the exact manufacturer part number, platform, reach, form factor and quantity required.
