TOWER's $1.3 billion order is expected to ignite silicon photonics—Is this a structural turning point for AI computing power?

```

Under the nearly stringent requirements for low latency, low power consumption, and large-scale scalability demanded by AI large models (LLMs) on cluster bandwidth, optical modules—the "nerve endings" of data center internal communications—are facing unprecedented pressure for technological iteration. Tower Semiconductor recently announced a $1.3 billion silicon photonics wafer contract for 2027, along with a $290 million advance payment from its largest customer, signaling that silicon photonics technology (SiPh) has officially entered its "explosive growth year."

This order sends an extremely clear signal to the capital market: silicon photonics technology is rapidly accelerating from an "alternative" to the "mainstream choice" for AI data center optical interconnects, and the slope of this process far exceeds the most optimistic institutional forecasts from a year ago. Short-term catalysts are highly focused on three dimensions: the structural pull exerted by doubled 800G demand and explosive 1.6T growth for CW light sources by 2026; Coherent's confirmation that 1.6T modules will simultaneously produce both EML and silicon photonics solutions with similar gross margins; and LightCounting's prediction that silicon photonics will, for the first time in 2026, account for more than half of global optical module shipments, implying a leap in industry status.

1. What happened? TOWER's $1.3 billion mega order

Tower has received not just a contract, but a "capacity insurance policy." In the semiconductor wafer foundry field, the customer's $290 million prepayment (about 22% of the total contract value) to lock in capacity three years ahead is extremely rare in history.

Highlight 1: Extremely high certainty. The existence of a prepayment means the execution of this order has both legal and financial compulsion, greatly hedging the downstream optical module manufacturers’ risk of default during demand fluctuations.

Highlight 2: Demand reflected ahead of time. Note that the $1.3 billion is only the "reserved capacity" base price, and the actual purchasing volume will likely be adjusted upwards with the full outbreak of 1.6T optical modules in 2027.

Tower revealed that its active silicon photonics customers now exceed 50, meaning silicon photonics is no longer just a play for giants like Intel or Cisco. This "from points to surfaces" expansion indicates that over the next three years, second and third-tier optical module manufacturers and cloud service providers (hyperscalers) developing their own chips will generate huge foundry demand.

In the 800G era, traditional EML solutions dominate, but as single-wave rates approach 200G, EML faces physical limits, declining yield, and raw material supply bottlenecks. Scale effect: Silicon photonics adopts mature silicon-based CMOS processes, enabling mass manufacturing on 12-inch wafers, with unit chip cost declining logarithmically with increased production, in sharp contrast to EML which relies heavily on manual coupling and precision packaging. Integration and reliability: Silicon photonics PIC (photonic integrated circuit) can integrate modulators, waveguides, detectors, etc. onto a single chip. This not only reduces packaging size but also drastically improves long-term link reliability by minimizing connection loss.

Strong endorsement from Coherent and AXT: Industry giant Coherent clarified that in its 1.6T solutions, silicon photonics and EML solutions have converging gross margins. When silicon photonics is no longer synonymous with "low price but low margin" but offers comparable profitability as advanced solutions, module makers will be more inclined to scale deployments. Meanwhile, AXT is expanding InP capacity to supply CW light sources needed for silicon photonics, solving the key bottleneck of "external light source" in silicon photonics solutions.

This involves an investment logic that needs deep understanding: The jump in CW light source demand from 200 million → 400 million, compared to incremental EML demand from 300 million → 500 million, reflects increasing silicon photonics share in the AI optical interconnect industry chain, rather than pure terminal demand growth. LightCounting predicts that by 2026, over half of global optical module shipments will use solutions based on silicon photonics modulators—in other words, silicon photonics will occupy "half of the landscape" for the first time.

2. Why is it important? Quantifying global supply structure and supply-demand gap

The global supply structure of the silicon photonics industry chain features three mainstream forms:

Type 1: IDM vertical integration. Coherent, Lumentum and other established optical communication leaders span the vertical chain from optical chips, silicon photonics PICs, and optical modules, and hold a complete product matrix covering EML/CW light sources and silicon photonics integration. Their advantages are technological closed-loop, supply chain security, and high gross margins; the disadvantage is huge pressure on capex and R&D for multi-link expansion.

Type 2: Dedicated foundry. Tower Semiconductor is the most typical case, its advantages lie in standardizing and modularizing silicon photonics manufacturing processes to service large numbers of fabless customers efficiently. Tower’s active customer base exceeding 50 is proof of this model’s large-scale success. TSMC and GlobalFoundries are also rapidly building silicon photonics foundry platforms.

Type 3: Pure design and system integration. Including China's leading optical module companies like Accelink and Eoptolink, whose core competitiveness lies in advanced packaging and system-level integration of silicon photonics modules, finishing manufacturing by externally sourcing CW light sources and PIC wafers. Some optical chip design companies focus on CW light source and PIC fields.

Looking at the global foundry landscape, Chinese companies like TeraHop and Hisense have shipped millions of optical modules and are rapidly narrowing the technology gap with Western leaders, forming a new wave of global competition and cooperation.

Supply-demand imbalance is currently one of the most important structural variables in the silicon photonics industry chain. Lumentum management pointed out that the current EML supply-demand gap is greater than 30%. The market predicts the global high-speed optical chip supply-demand imbalance will persist until 2027. Supply constraints come from several layers:

High layer—manufacturing yield barriers: Optical chip overall yield is much lower than integrated circuits, with huge yield differences between manufacturers for the same products. Yield is affected by multiple process factors; gas mix, temperature changes, equipment angles, etc. can cause yield fluctuations directly affecting actual delivery capability.

Mid layer—long cycle for production line expansion: Delivery and debugging cycles for core equipment like MOCVD and EBL run over a year, high-end InP substrates are monopolized by foreign companies and expansion cycles are long. Dual constraints of equipment and materials make industry-wide expansion extremely difficult.

Low layer—hard client verification threshold: Optical chip client verification process is rigorous, from internal testing to module manufacturer validation to end-user certification and mass production, the full cycle is about 2 years. Client stickiness is extremely high, new entrants breaking into high-end supply chains face great difficulty, so the industry competition is characterized by few but elite players.

According to Yole, the global silicon photonic chip market will grow from $6.8 million in 2022 to $61.3 million by 2028, with a CAGR of 44%. Moreover, silicon photonics will accelerate in 800G, 1.6T, and even faster optical modules and extend to next-generation connection schemes like CPO, becoming the key force driving technological iterations in the optical communications industry.

Silicon photonics PIC design is the segment with the highest technical barriers in the silicon photonics industry chain; traditional advantages are held by design giants such as Intel, Broadcom, Marvell. However, domestic manufacturers are catching up fast: leading optical module companies like Accelink have accumulated strong capabilities in module-level system design and deep client relationships through custom PIC design for AI clients like Nvidia. Tianfu Communication's optical engine yield reaches 95% and is Nvidia’s exclusive supplier, occupying a key packaging value link in the silicon photonics module chain.

Optoelectronic co-packaging (especially CPO technology) is becoming the next major growth engine in the silicon photonics ecosystem. Its core value is reducing data transmission power by 80%, lowering latency to the picosecond level, and increasing bandwidth density fivefold. As CPO technology enters mass deployment from 2028 to 2030, silicon photonics will leverage its high integration as the core platform technology. According to Yole, silicon photonics transceiver market size could reach $5.413 billion in 2027, with the CPO optical engine market reaching $259 million.

Facing the doubling annual AI compute spending by giants like Nvidia, traditional solutions can no longer match the hunger for 800G/1.6T modules. Silicon photonics leverages existing logic chip wafer foundries (such as Tower, GlobalFoundries, TSMC) to rapidly achieve production leaps from tens of thousands to hundreds of thousands of wafers.

3. What's next? Evolutionary pace of industrial trends

Based on current industrial progress and global technology roadmaps, we believe the silicon photonics industry will evolve along the following path:

Phase One: 2026-2027—Rapid penetration. Silicon photonics will dominate 800G/1.6T optical modules, with penetration rates expected to reach 50% and 60% or more, respectively. Foundries like Tower will operate at full capacity, CW light sources will be structurally tight. The Chinese silicon photonics chain will make the critical leap from "sample verification" to "mass delivery." Sustainable volume and quality ramp-up of 1.6T modules this year will be a key marker of chain maturity.

Phase Two: 2027-2029—New tech fusion and deepening localization. CPO/NPO and other new packaging technologies will be gradually introduced, heterogenous integration of silicon photonics + thin-film lithium niobate will emerge in the 3.2T era. China's silicon photonics chain will approach international advanced completeness and global AI infrastructure supply will see rising Chinese module share. Domestic silicon photonic chip production lines of $10 billion scale will fully come online, with mature processes and significant cost advantages, forming structural substitution for foreign supply chains.

Phase Three: After 2030—Era of silicon photonics domination and platformization. Silicon photonics will move from optical modules to broader applications such as optical interconnect, optical computing, and optical sensing. With the implementation of OIO (optical input/output) technology, optical chip usage will soar, and the silicon photonics platform will integrate InP, silicon nitride, thin-film lithium niobate, and other photonic materials.

The current silicon photonics industry chain faces a threefold resonance of overall industry growth, structural silicon photonics penetration, and rising domestic share. Specifically:

First (total growth): Explosive demand for 800G/1.6T high-speed optical modules. 800G+1.6T modules are projected to reach 70 million units in 2026 and 100-150 million in 2027, creating massive incremental demand for silicon photonics.

Second (structural substitution): Accelerated penetration of silicon photonics, rising from 20-25% in 2025 to an expected industry status of 50-60% in 2027. LightCounting predicts that in 2026, silicon photonics optical module shipments will occupy half of the market—a fundamental revaluation of the silicon photonics industry chain's value.

Third (local substitution): China's silicon photonics industry chain completeness is rapidly improving; all segments—CW light sources, silicon photonics PIC, silicon photonics foundry (8-inch lines), electrical chips, optical engines/packaging—have mature local enterprise coverage. Increasing localization rate should further enhance profitability and global competitiveness of China's silicon photonics industry chain.

Conclusion: Non-linear leap in industry value

In summary, Tower’s mega-contract is a milestone marking the leap of the optical module industry from the "analog era" to the "fully integrated era."

This means the future model of the optical module industry needs to be reassessed—not just by order volume, but by its "degree of siliconization." Silicon photonics is no longer a distant vision, but a core track that is already here and will contribute significant revenue certainty in 2027. The domestic silicon photonics industry chain’s advantages in cost control and supply chain response speed will enable it to continuously gain excess returns in the global AI compute infrastructure wave.

Risk Disclaimer and Exemption ClauseThe market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their circumstances. Invest at your own risk. ```