Why are AI photonics stocks collectively pulling back? LITE capacity sold out until 2029, AAOI still undervalued by the market?

Beijing time July 3, 2026, the other side of the traditional financial market presented a completely different picture—the US semiconductor index fell 11% in two days, the storage sector plunged sharply, and Sandisk dropped over 14%.

In this systemic correction of the AI hardware supply chain, the photonics and optical module sector became the hardest hit. Applied Optoelectronics (AAOI) closed at $120.95 on July 3, down 12.99% in a single day; Lumentum Holdings (LITE) closed at $728.32, down 9.09%. Coherent, Corning, and other optical communication-related stocks weakened in tandem.

For investors focused on the long-term narrative of AI infrastructure, a core question arises: is this decline a reversal of industry logic, or a phased mispricing caused by capital flow fluctuations?

What Happened in the Market: Photonics Sector Under Collective Pressure

From Beijing time July 2 to July 3, photonics and optical module-related stocks experienced a notable synchronized correction.

AAOI hit an intraday low of $113.41 on July 2, down about 18.4% from the previous close of $139.00, with a full-day range of 18.60%, trading volume of 15.9913 million shares, and a turnover rate of 19.93%. The stock’s 52-week high is $233.67, meaning even after this correction, its year-to-date gains remain substantial.

LITE also faced significant pressure. On July 2, the stock opened at $777.58, with a previous close of $801.16, trading in a range of $710.01 to $803.90, and closing at $728.32. It fell roughly 10% during the session. Notably, LITE’s Quant rating remains as high as 4.98, and its year-to-date stock price gain is about 130.99%—the substantial prior gains provide important context for this pullback.

From a sector perspective, this is not an isolated movement for photonics stocks. The iShares Semiconductor ETF (SOXX) fell about 6% over the same period. Chip stocks like Intel and AMD also weakened. The entire AI hardware chain—from computing chips to storage chips to optical modules—faced systemic selling pressure.

Direct Trigger of the Decline: Emotional Impact from Meta’s Computing Power Sale

The catalyst for this adjustment points to news from tech giant Meta.

Reports indicate that Meta plans to sell surplus computing power, sparking widespread market concerns about "AI computing power oversupply." Global tech stocks fell in response, and popular A-share tech sectors like storage chips, optical modules, PCBs, and MLCCs also declined.

The transmission mechanism of this logic chain is as follows: If Meta’s computing power is in surplus, does that mean the entire AI infrastructure buildout has exceeded actual demand? Will top cloud vendors’ capital expenditures slow down accordingly? Once capital expenditure contracts, all links in the AI hardware supply chain—from GPUs to optical modules—face the risk of order reductions.

However, whether this concern is fully supported by data warrants careful evaluation. Some industry insiders told the media that the computing power leasing market has not truly cooled. From an industry logic perspective, top cloud vendors’ capital expenditure decisions are typically based on multi-year planning, and a single company’s computing power allocation actions are unlikely to signal a reversal of industry trends.

What’s Happening with Fundamentals: Supply-Demand Gap, Not Oversupply

If market pricing reflects concerns about "computing power oversupply," then industry-level data tells a different story.

LITE: Capacity Sold Out Through 2029

As the global leader in EML lasers, Lumentum Holdings’ capacity status is a key indicator of the optical communications industry’s health. According to the latest information, LITE’s capacity is sold out through 2029. The current supply-demand gap for EML chips is about 30%. The company plans to increase EML capacity by over 50% by the end of 2026 compared to 2025, but it still cannot fill the gap.

Financially, LITE’s FY2026 single-quarter revenue reached $808.4 million, soaring 90.1% year-over-year, significantly exceeding market expectations. The company has raised its FY2026 Q4 revenue guidance to a range of $960 million to $1.01 billion.

Having capacity sold out through 2029 fundamentally contradicts the market’s fear of "computing power oversupply."

AAOI: Entering the AI Data Center Supply Chain with Clear Revenue Path

AAOI also shows strong industry momentum. The company is entering the market with U.S.-manufactured independent CW (continuous wave) laser capacity. Supply chain information reveals that AMD is in talks with AAOI for the procurement of high-power CW laser chips and external laser sources (ELSFP) to secure future capacity for AI servers and scale-up optical interconnects. AAOI is expected to become an important partner for AMD in building its own optical communications supply chain.

On the revenue front, AAOI expects to achieve quarterly revenue of $1.4 billion by the first half of 2027. Multiple brokerages have set target prices in the range of $159 to $210 based on revenue projections for 2026 to 2028.

Overall Industry: Tight Supply, Not Loose

From a broader industry perspective, the supply-demand landscape for the optical communications supply chain is not loose but continues to tighten. Upstream indium phosphide (InP) substrate supplier AXTI has gained 453%. AOI’s actual demand is nearly 40% higher than expected. Lumentum, Coherent, and AAOI—the three major North American optical component suppliers—are all benefiting from the AI-driven data center infrastructure buildout and are accelerating capacity expansion to meet strong orders for 2027.

Supply-demand gaps, capacity sold out, and order visibility extending to 2029—these industry-level data points reveal a significant factual deviation from the "oversupply" narrative being traded in the market.

Structural Industry Changes: The Third Phase of AI Infrastructure

The essence of this round of industry changes is that AI infrastructure construction is entering a new phase.

Phase 1: GPU. NVIDIA, as the core of computing power, dominated the early expansion of AI infrastructure.

Phase 2: HBM/DRAM. As model scale increases, memory bandwidth becomes the bottleneck, putting memory manufacturers like Micron and SK Hynix in the spotlight.

Phase 3 (Current): Optical Interconnect/CPO/Photonics. As AI cluster scales continue to expand—with a surge in GPU count—traditional electrical interconnects (copper) hit physical limits in bandwidth, latency, and power consumption. Data centers need lower latency, higher bandwidth, and lower power interconnection solutions. Optical interconnect (Optical Interconnect/CPO) thus becomes the core bottleneck for next-generation AI infrastructure.

A recent Goldman Sachs research report quantifies this trend: the total addressable market for optical interconnects is expected to grow from about $150 billion to $1.54 trillion, an increase of more than 9 times. The core driver is the shift in data center architecture from scale-out to scale-up—the demand for high-speed optical interconnects within racks and super nodes is exploding.

This trend is clearly visible from NVIDIA’s product roadmap: the current production GB300 NVL72 has a network interconnect value of $315k per compute unit, while the expected Rubin Ultra NVL576 shipping in 2027-2028 is projected to reach $9.4 billion, a 29-fold increase. This leap stems from the expansion of GPU cluster size from 72 to 576 units, with interconnect layers extending from intra-rack to inter-rack.

On the technology path, CPO and pluggable optical modules are not substitutes but coexist in layers. CPO is suitable for scenarios extremely sensitive to bandwidth and power consumption, while pluggable solutions continue to dominate applications requiring flexible maintenance. Goldman Sachs estimates that by 2028, CPO penetration in scale-out scenarios will be about 29%, but the absolute volume of pluggable optical modules will still continue to grow.

Long-Term Logic: Temporary Misalignment Between Price and Fundamentals

Synthesizing the above analysis, the current collective correction in the photonics sector can be understood as a typical instance of "price and fundamentals misalignment."

In the short term, the market is digesting several factors: profit-taking pressure from excessive prior gains in the AI hardware sector; emotional impact from Meta’s computing power sale news; and systemic selling pressure on momentum stocks during the July seasonal pullback. These factors together drive the short-term decline.

However, long-term industry realities remain unchanged by short-term fluctuations: LITE’s capacity is sold out through 2029; AAOI’s revenue explosion path is clear; the optical interconnect TAM faces multiple expansions; the four major cloud vendors plan to invest a combined $725 billion in capital expenditure in 2026, up 77% from $410 billion in 2025.

Serenity’s response on July 3 pointed out this misalignment: CPO-related stocks have actual revenue support, in stark contrast to quantum computing concept stocks with “almost no revenue.” LITE’s capacity is sold out through 2029, and AAOI is entering the market with U.S.-manufactured independent CW capacity, expecting quarterly revenue of $1.4 billion by the first half of 2027, while its current market cap is only about $9.3 billion.

Conclusion

The logic of AI infrastructure construction has not changed due to short-term market fluctuations. From GPUs to HBM/DRAM to optical interconnects, the value center of the supply chain is migrating along the physical bottlenecks of computing power expansion. The current correction in the photonics sector likely reflects a temporary rebalancing of capital within the AI hardware supply chain, rather than a reversal of industry trends.

For investors focused on the long-term narrative of AI infrastructure, the core question may not be whether optical interconnects are still important—the expansion of AI clusters objectively requires higher bandwidth and lower latency interconnection solutions, an industrial trend with a solid physical foundation. What truly needs continuous tracking is the evolution pace of the supply-demand gap, the progress of capacity expansion execution, and the iteration direction of technology paths.

In the short term, the market is a voting machine; in the long term, it is a weighing machine. When price and fundamentals misalign, the final return direction is often determined by the latter.

FAQ

Q1: Why have optical module and CPO stocks fallen collectively recently?

From July 2 to 3, Beijing time, affected by news that Meta plans to sell surplus computing power, the market feared AI computing power oversupply leading to capital expenditure contraction. Global AI hardware sectors faced systemic selling pressure, and optical modules, storage chips, and other supply chain links corrected synchronously. AAOI fell 12.99% in a single day, LITE fell 9.09%.

Q2: Is LITE’s capacity truly sold out through 2029?

Yes. According to public information, Lumentum Holdings (LITE) capacity is sold out through 2029. The current EML chip gap is about 30%. The company plans to increase capacity by over 50% by the end of 2026, but it still cannot meet demand. FY2026 single-quarter revenue was $808.4 million, up 90.1% year-over-year.

Q3: What is AAOI’s fundamental situation?

AAOI is entering the AI data center supply chain with U.S.-manufactured independent CW capacity. AMD is in talks with it for high-power CW laser chip procurement cooperation. The company expects quarterly revenue of $1.4 billion by the first half of 2027. Multiple brokerages give buy ratings with target price ranges of $159 to $210.

Q4: How large is the long-term market for optical interconnects?

Goldman Sachs predicts the total addressable market for optical interconnects will grow from about $150 billion to $1.54 trillion, an increase of over 9 times. The core driver is the shift of AI data centers from scale-out to scale-up, with explosive demand for high-speed optical interconnects within racks. From GB300 NVL72 to Rubin Ultra NVL576, the network value per compute unit will grow 29 times.

Q5: Is the current pullback a buying opportunity or a trend reversal?

From industry data, fundamental factors such as LITE capacity sold out through 2029, AAOI’s clear revenue path, and multiple expansions of the optical interconnect TAM have not changed. The current pullback more likely reflects a temporary rebalancing of capital within the AI hardware supply chain, rather than a trend reversal. However, investors still need to monitor key variables such as the evolution of the supply-demand gap and the progress of capacity expansion execution.