Skip to content
⌘ K

Optics will be a persistent bottleneck for AI data centers for the next five years

Optics is a credible AI-data-center bottleneck today, and a published forecast projects significant high-speed-transceiver shortfalls through 2029. That supports pressure during the earlier part of the two-to-five-year horizon, but not a confident assertion that optics will remain the binding constraint for the entire period through October 2031. Capacity additions and changing optical architectures are the principal qualifications.

Read full analysis

As of October 6, 2026, the evidence separates a current constraint from a five-year forecast. A procurement report describes optical interconnects as in shortage and reports 36–50-week lead times for 800G transceivers. Those are sourcing observations, not a prediction that the same lead times will persist. [2]

The strongest longer-range evidence is McKinsey's analysis: it projects 800-Gbps transceiver production 40–60% below demand through 2027 and 1.6-Tbps shortfalls of 30–40% through 2029. It says scaling capacity and investing to address the shortages will be necessary. These estimates support a risk extending into approximately the first three years from the as-of date, but they neither observe future shortages nor establish one in 2030–31. [1]

Demand could keep moving ahead of supply. Coherent describes exceptional customer demand, expanding production, and an expected multiyear AI-data-center transition from copper to optics. Lumentum says AI bandwidth requirements are making optical links a primary connectivity choice and points to growing demand for co-packaged-optics lasers. Broadcom describes high-speed copper-reach limits. These are reasons optical capacity matters; none, individually, measures an industry-wide supply deficit for five years. [3][4][5]

The countercase is concrete rather than merely theoretical. Coherent says it is expanding production, and Corning has announced more polarizer capacity for optical components. Broadcom proposes near-package optics built around established VCSEL technology as an alternative to traditional pluggables. Marvell reports volume shipments of one 1.6T DSP and sampling of additional designs. These steps could relieve specific constraints, although announcements and samples do not establish that capacity will catch up with demand. [3][7][5][6]

Assessment for October 2028–October 2031: persistent optical pressure into 2028–29 is plausible; calling optics the persistent AI-data-center bottleneck all the way to 2031 goes beyond the cited evidence. The outcome depends on whether component output and alternative architectures scale faster or slower than optical-link demand.

Conviction
Medium
Horizon
2–5y
Status
Active
Evidence
7 Entries
Last revised
October 6, 2026

Conviction

Medium

Medium conviction: a documented 2026 procurement constraint and McKinsey's projected shortfalls through 2029 make continued pressure into the earlier part of the two-to-five-year horizon plausible. The strongest caveat is the word persistent: no cited forecast establishes that optics will remain the binding AI-data-center bottleneck through October 2031, while manufacturers are expanding capacity and introducing alternative designs.

What would change this assessment
  • By 2028–29, independently reported 1.6T transceiver shipments consistently meet orders and delivery lead times normalize across major suppliers; that would weaken the persistence claim.

  • After announced manufacturing expansions reach production, laser, polarizer, and transceiver suppliers still report multiquarter unfilled orders or allocation into 2030–31; that would strengthen the claim.

  • Large AI-cluster operators deploy alternative optical architectures at scale and report that previously constrained pluggable components no longer delay cluster commissioning; that would weaken the claim.

  • Optics suppliers or data-center operators document projects postponed specifically for lack of optical components in 2030 or 2031, rather than for power, construction, switches, or accelerators; that would strongly support the full five-year claim.

Evidence balance

2 Supporting4 Mixed1 Challenges

Pillars

Pillars for

Pillars against

Timeline of Articles

Discussion
MixedF3A1
September 24, 2026

Corning Announces Polarcor Capacity Expansion to Support Multi-Year Supply Agreement

Corning announces additional capacity for glass polarizers used in optical isolators and laser-based components, tied to a multiyear photonics supply agreement. This confirms investment in the optical supply chain but also provides a concrete potential relief valve; it gives no quantified shortage or completion date for the expansion.

MixedF3A1
August 12, 2026

Coherent Corp. Reports Fourth Quarter and Full Year Fiscal 2026 Results

Coherent reports exceptional customer demand, expanding production capacity, and a management expectation that AI data centers' transition from copper to optics is a multiyear opportunity. Demand supports continued pressure; capacity expansion could relieve it. The release does not quantify an industry-wide shortfall.

SupportingF3
August 11, 2026

Lumentum Announces Fourth Quarter and Full Fiscal Year 2026 Results

Lumentum says rising AI bandwidth is making optical links a primary connectivity choice and reports increasing demand for co-packaged-optics lasers and early external-laser-source orders. This supports a growing optical workload, not a demonstrated five-year supply shortage.

MixedF3A2
March 12, 2026

Beyond the Copper Wall: Scaling AI Clusters with VCSEL-Based Near-Package Optics (NPO)

Broadcom argues that copper reach is becoming restrictive at higher speeds, but presents near-package optics using established VCSEL manufacturing as an alternative to conventional pluggable optics. It supports rising demand for optical links while identifying a possible way around particular component bottlenecks.

ChallengesA2
March 12, 2026

Marvell Ushers In the 1.6T Era with Expanded Optical DSP Platform Portfolio, Redefining AI Data Center End-to-End Connectivity

Marvell says its established 1.6T Ara optical DSP is shipping in volume and announces additional, more power-efficient designs that began sampling in Q1 2026. These developments could improve optical-system deployment and weaken a claim that any particular optics design remains constrained for five years; samples are not proof of sufficient industry supply.

SupportingF1
2026

Navigating the 2026 Supercycle

A procurement-industry report describes optical interconnects as already in shortage and gives 36–50-week lead times for 800G transceivers. Its sourcing observations support a present constraint, but do not establish how long it will last.

MixedF2A1
July 2, 2025

Opportunities in networking optics: Boosting supply for data centers

McKinsey forecasts that production of 800-Gbps transceivers will fall 40–60% short of demand through 2027 and that 1.6-Tbps supply shortfalls of 30–40% are likely through 2029. This is the clearest direct support for a bottleneck extending into the forecast horizon, but it is a forecast, not an observed shortage through 2029 or evidence of one in 2031.

Comments

LatestPopular
Write a comment
Loading comments…

Request a Thesis

Tell us what you’d like Roadstar to investigate.

New question

What would you like to know?

Context guides the research and is not shown on the finished page.