What Is Optoelectronic Fusion? How AI Exposed the "Limits of Copper"
Optoelectronic fusion, put simply, is "a technology that melds circuits handling electrical signals together with circuits handling optical signals into a single chip or package." Engineers at NTT describe it as "a technology that integrates electronic circuits and optical circuits into a single system." Since this alone is rather abstract, let us break it down with a familiar example.
The "optical fiber" that brings the internet into our homes already carries information using light. However, that light is first converted into electrical signals by the modem in the home, and inside our PCs and smartphones, everything is processed by passing through copper wiring (electricity). Light for long distances, electricity for short distances—this has been the conventional wisdom for many years. Because converting between light and electricity requires dedicated components and energy, there was no point in going to the trouble of using light over short distances.
Generative AI has overturned this conventional wisdom from its very foundations. The GPUs (image-processing semiconductors) responsible for AI computation operate like a single giant brain, bundled together in the thousands. The problem is that GPUs exchange data with one another at a furious pace—and the copper wiring carrying that data falls into a triple bind of "heat generation, power consumption, and latency" the faster one tries to make it. Looking at concrete figures, NVIDIA's latest-generation GPU servers consume as much as 1.4 megawatts of power per rack. Current 1.6-terabit-per-second-class optical transceivers (the components that convert between light and electricity) consume about 30 watts each, of which the signal-processing chip (DSP) alone accounts for more than 15 watts. When thousands of these are lined up in a data center, the calculation comes out to megawatt-class power disappearing "just to move data." In fact, the proportion of "connecting components (interconnect components)" within data center capital expenditure has doubled, from 15% three years ago to more than 30% in 2026.
This gave rise to the idea of bringing light ever closer—even into the short distances that had until now been deemed "fine to keep as electricity": between device and device, between board and board, further between chip and chip, and ultimately inside the chip itself. This is optoelectronic fusion, and its leading edge is "Co-Packaged Optics (CPO)," which houses optical input/output engines right next to GPUs and switch semiconductors (within the same package). The effects of replacing electricity with light are dramatic: NTT's prototypes are projected to reduce power consumption to one-eighth of current levels, and ultimately to as little as one-hundredth. "Light is not only fast, but overwhelmingly energy-efficient"—this is the heart of why optoelectronic fusion has become an essential technology in the AI era.
Why Is It the "Most Critical Item" in Economic Security? — The Vital Point Is Indium Phosphide
There are two main reasons why optoelectronic fusion (photonics-electronics convergence) is called "the single most important item in economic security."
First, it is the very foundation of AI infrastructure itself. As we saw in the previous chapter, there is no scaling of AI without going optical. In an era where AI determines a nation's economic, military, and scientific power, having the core infrastructure of that heart controlled by another country carries the same significance as depending on another country for energy or food. Based on the Economic Security Promotion Act (2022), the Japanese government designated semiconductors as a "specified critical material" in December 2022, and in February 2026 it added advanced electronic components such as capacitors and filters to the designation. Furthermore, the "K Program" (Key and Advanced Technology R&D through Cross Community Collaboration Program), which backs the research and development of specified critical technologies with public funds, has been allocated a total of roughly 500 billion yen, operated through the funds of JST (Japan Science and Technology Agency) and NEDO (New Energy and Industrial Technology Development Organization). Optoelectronic fusion is now being positioned as precisely the centerpiece of this framework.
Second—and this is what this article most wishes to emphasize—the supply chain for optoelectronic fusion contains numerous "chokepoints" that only specific countries or companies possess. The biggest chokepoint is the wafer (substrate) of indium phosphide (InP), a compound semiconductor that serves as the foundation for laser light sources. Silicon may be good at conducting electricity, but it cannot emit light. To create light (lasers), special semiconductors such as InP or gallium arsenide (GaAs) are required. The global supply of these InP substrates—though estimates vary in range—is an extreme oligopoly market in which just the top three players account for over 90%: Japan's Sumitomo Electric Industries (estimated at roughly 40–60%), the U.S. firm AXT (including its Chinese subsidiary Beijing Tongmei), and JX Metals. Moreover, supply and demand are on the verge of collapse: while global shipments in 2025 remain at 600,000–700,000 units, demand reaches 1.5 million–2 million units, resulting in a supply shortfall of over 70%. The order backlog is filled through 2027 and beyond, and inventories are near zero. Even more serious is the fact that AXT's shipments are being constrained by delays in China's export licenses. This is precisely an economic security problem in which geopolitical risk directly strikes the supply of optical chips.
The picture surrounding next-generation materials is even more precarious. For "thin-film lithium niobate (TFLN)," said to have performance surpassing silicon, China holds a near-monopoly on wafer supply, and in terms of the number of patent applications in the optoelectronic fusion field, China nearly drew level with the United States in 2021 and, as of January 2025, is said to have reached more than double that of the U.S. (about 800 applications). In response to this trend, the U.S. has begun "removing China" from its supply networks. In November 2025, GlobalFoundries acquired Singapore's Advanced Micro Foundry (AMF), touting a "China-free" optical chip supply as its selling point. NVIDIA, too, in order to secure light sources, is investing 2 billion dollars each (roughly 310 billion yen each) in Lumentum and Coherent—a total of 4 billion dollars (about 620 billion yen).
Within this tense configuration, Japan's position is distinctive. It has not secured leadership over finished products (switches and GPUs) or over design and mass production. However, in the domain of "foundational technologies = components, materials, and equipment"—such as laser light sources, optical components, manufacturing equipment, and inspection equipment—Japan holds many chokepoints that everyone in the world is forced to rely on. To borrow the words of Nikkei xTECH, "Japan has strength in light sources," and in the phrasing of Mizuho Bank's Industry Research Department, "the winning path is a trinity." As discussed later, this is also precisely why Silicon Valley VCs are paying attention to Japan.
NTT's IOWN Initiative — In Fiscal 2026, Optical-Electrical Fusion Devices Finally Reach Commercialization
The banner under which Japan leads the world in photonics-electronics convergence is NTT's "IOWN" (pronounced "eye-own," Innovative Optical and Wireless Network) initiative. Proposed in 2019, the "IOWN Global Forum" was launched in January 2020 with NTT, Intel of the United States, and Sony as founding members, and by February 2026 it had grown into an international ecosystem involving more than 170 companies and universities worldwide. IOWN's ultimate goal is grand: by replacing network processing wholesale with optics, it aims to achieve "100 times the power efficiency, 125 times the transmission capacity, and one-two-hundredth the latency."
The All-Photonics Network (APN), the first stage of IOWN, has already been commercialized. In addition to being used at the 2025 Osaka-Kansai Expo for a demonstration connecting the venue with a data center, it has been put into operation, in partnership with Chunghwa Telecom, as an optical path spanning roughly 3,000 km between Taiwan and Japan with a latency of about 17 milliseconds. And in 2026, the initiative finally enters its second stage: "bringing photonics-electronics convergence devices inside the computer."
The key here is the photonics-electronics convergence device "PEC," which evolves generation by generation. Under NTT's roadmap, "PEC-2," which replaces the electrical wiring between a computer's boards (circuit boards) with optics, will be made commercially available during fiscal 2026 (by March 2027). Next, "PEC-3," an "optical I/O" that connects semiconductor packages such as CPUs and GPUs optically, will be introduced as a commercial sample around 2028. Ultimately, around 2032, the wiring inside the chip itself will be made optical, with the goal of reducing power consumption to one-hundredth. The closer it gets to optics, the greater the effect—but the technical difficulty also soars; this step-by-step conquest forms the backbone of the initiative.
PEC-2, the first wave of commercialization, is taking shape as a product that can be concretely envisioned. The core "light engine" is a compact component roughly 20 millimeters wide, and a switch carrying 16 of them achieves a total communication capacity of 102.4 terabits per second—the highest level on the market—at one-eighth the power of conventional designs. NTT describes this capacity as "more than seven times the 14.4-terabit-per-second chip-to-chip communication speed of the latest GPUs." Supporting this commercialization is a carefully assembled supply chain. The design of the LSI (large-scale integrated circuit) built into the switch is handled by Broadcom of the United States; the design and manufacture of the switch chassis that carries it by Accton Technology of Taiwan; and the semiconductor package substrate by Shinko Electric Industries. And responsible for the design and manufacture of the heart of the system—the light engine and switch module—is NTT Innovative Devices, which has a production capacity of 5,000 units per month per line and plans to expand to at least three lines going forward. Another strength unmatched by competitors is that NTT achieves miniaturization through its proprietary "membrane (thin-film) device" technology, which makes optical elements extremely thin.
IOWN-related revenue is still very small when viewed against NTT's roughly 13 trillion yen in consolidated sales. However, starting from commercialization in fiscal 2026, photonics-electronics convergence devices for AI data centers could grow into a business worth anywhere from tens of billions of yen to the trillion-yen scale in the 2030s. By one estimate, building the APN and implementing CPO in servers together would require investment totaling 4.5 trillion yen. NTT is rapidly pushing the initiative into its "implementation" phase—with President Akira Shimada delivering a keynote at MWC Barcelona 2026 in March 2026, and the release of the "NTT IOWN Technology Report" in January.
Japan's Homegrown Semiconductor Maker Rapidus—The Bet on "Back-End Processing Premised on Optoelectronic Fusion"
Another leading player in optoelectronic fusion is Rapidus, the "Japanese national semiconductor" champion taking on cutting-edge 2nm (nanometer) generation logic chips. Founded in August 2022 by eight companies—Toyota, Sony, NEC, NTT, SoftBank, Denso, Kioxia, and MUFG Bank—the company launched a pilot line at its "IIM-1" factory in Chitose, Hokkaido, in April 2025, and in July of the same year unveiled a 2nm prototype chip on a 300mm wafer using the GAA (Gate-All-Around) architecture developed jointly with IBM. Compared with 7nm products as of 2022, 2nm is said to deliver up to a 45% performance improvement and a 75% reduction in power consumption, with mass production targeted to begin in 2027.
There is momentum on the funding front as well. On February 27, 2026, the company completed a total fundraising of 267.6 billion yen (approximately 1.7 billion dollars), of which 167.6 billion yen was raised from 32 companies including Canon, Fujitsu, NTT, SoftBank, the Development Bank of Japan, and Sony Group. Then, on April 11, the Ministry of Economy, Trade and Industry approved an additional 631.5 billion yen (approximately 4 billion dollars), bringing the government's cumulative research and development support to about 2.35 trillion yen. The U.S.-based Bloomberg has reported this as "Japan's roughly 16 billion dollar bet on Rapidus."
What is decisively important for the theme of this article is that Rapidus is venturing into "back-end processes (packaging and assembly) predicated on optoelectronic fusion." In the world of cutting-edge chips, the "back-end process" of combining multiple small chips (chiplets) like Lego blocks has become a new main battleground, and this packaging technology is precisely what holds the key to mounting optical I/O onto a package. Rapidus opened its back-end process base, "Rapidus Chiplet Solutions (RCS)," on April 11, 2026, and is advancing a world-first effort to prototype organic insulating-film RDL (redistribution layer) interposers using large panels measuring 600mm square. Furthermore, according to a report by TrendForce (May 26, 2026), the company has adopted Lam Research's panel-level packaging equipment and has also embarked on the development of 600mm-square glass interposers.
Yasumitsu Orii, Rapidus's executive vice president, has outlined a policy of evolving back-end processes in four stages, stating clearly that the company will begin mass production of multiple generations of back-end processes from 2028 and "in the future, will also handle optoelectronic fusion." This strategy of providing front-end processes (2nm manufacturing) and back-end processes (advanced packaging) as an integrated whole is also a bet to elevate Rapidus from a mere "contract manufacturer" into a "comprehensive player that handles everything up to the packaging of AI semiconductors."
The optical-electronic fusion (co-packaged optics) implementation project of Rapidus/LSTC, adopted by NEDO
Backing this bet as a national project is the research and development initiative led by LSTC (the Technology Research Association Leading-edge Semiconductor Technology Center), announced by the Ministry of Economy, Trade and Industry on April 11, 2026, and adopted by NEDO. Its official name is "Development of Semiconductor Packaging Technology to Accelerate Optoelectronic Fusion and Formation of an Advanced Back-End Process Foundation," positioned as part of the "Post-5G Information and Communications System Infrastructure Enhancement R&D Project / Development of Advanced Semiconductor Manufacturing Technology." LSTC announced this on April 13 and held a detailed press conference on April 17.
The aim of this project is to establish optoelectronic fusion packaging technology that surpasses the limits of electrical wiring. The numerical targets it sets out are concrete: achieving ultra-high-bandwidth transmission on the order of 10 terabits per second per millimeter, along with a reduction in power consumption of 40% or more. To this end, it will develop technology to "hybrid-bond" optical engines and optical RDL interposers at an extremely fine pitch of 6 micrometers or less, and will establish the world's first advanced back-end-process open-innovation hub capable of handling 300mm-square panels. The technology development is broadly composed of three pillars: (1) technology to bond optical engines and optical RDL interposers with high precision, (2) the optical RDL interposer technology itself, and (3) the formation of a back-end-process foundation supporting 300mm-square panels.
What deserves attention is the project's implementation structure and location. Participating in the project, alongside LSTC, Chitose Institute of Science and Technology (a public university), Tohoku University, Hokkaido University, and Yokohama National University, is Belgium's imec—one of the world's foremost semiconductor research institutions—which joins as a technical collaborator. The R&D hub will be built within the campus of Chitose Institute of Science and Technology, that is, adjacent to Rapidus's RCS, with completion targeted for fiscal 2028. With Rapidus handling the front-end process, the LSTC hub handling the back-end process, and universities and international research institutions all concentrated in the Chitose area of Hokkaido, a configuration that might well be called a "Japanese silicon photonics cluster" is now taking shape.
Masters of the Upstream ①—Compound Semiconductors and Laser Light Sources (Sumitomo Electric Industries / Furukawa Electric)
From here, we will examine the upstream choke points that this paper calls "Japan's winning hand," focusing on specific companies. First up are the compound semiconductors and laser light sources that generate the "light itself" of optoelectronic fusion.
Foremost among them is Sumitomo Electric Industries. As noted above, the company holds a world-leading share (estimated at roughly 40–60%) of indium phosphide (InP) substrates, which form the foundation of laser light sources. Because silicon does not emit light, making a laser requires the "field" that is InP—and one can understand it as Sumitomo Electric being the company that owns more of that field than anyone else in the world. The company is also strong in compound semiconductor technologies such as the EML (electro-absorption modulator integrated laser), regarded as the industry leader for the transmitting-side light source in AI data centers, and GaN (gallium nitride) HEMTs, which can also be used for next-generation wireless. Investment is accelerating as well: in early 2025 the company announced it would invest approximately 14 billion yen to expand production of optical devices for communications, and at its Yokohama Works a new R&D building is set to be completed in March 2026 and to begin operating in July, with a plan to raise optical device production capacity in fiscal 2028 to several times the fiscal 2024 level. It also plans to increase InP substrate production capacity by about 40% by 2027. The more InP becomes the "choke point" of AI, the more the strategic value of Sumitomo Electric, which owns that field, rises.
Another such company is Furukawa Electric. The company has built a global position in DFB laser diode chips for signal light sources and in the pump lasers indispensable to optical fiber amplifiers. Furukawa Electric's brand "FITEL" has prided itself on being No. 1 in 1480 nm-band pump light sources since 1999, and for CPO it supplies an "external light source (ELS)" that places the light source outside the chip. In CPO, designs that deliberately separate and mount heat-sensitive lasers outside the chip are becoming mainstream, and Furukawa is positioned to supply that "battery of light." Its investment in response to the surge in demand is also large: it is investing a total of 38 billion yen in Iwate Prefecture and Thailand to raise DFB laser chip production capacity to more than five times the fiscal 2025 level (targeted for 2028), and its second plant in Thailand was completed in February 2026. The company has set a medium-term target of growing the operating profit of its data center business to 200 billion yen in the fiscal year ending March 2031—8.5 times the previous year—and expects sales of its optical communications products for data centers in fiscal 2025 to be about double those of the previous year. In NICT's B5G commissioned research "BRIGHTEN," it is also participating in the development of an optical transceiver for CPO at 56 gigabits per second.
Masters of the Upstream ②: Precision Optical Components and High-Density Connectors (Japan Aviation Electronics Industry, Hirose Electric)
Light ultimately always needs a "socket." Connectors that join optical fibers to one another, or optical components to substrates, and ferrules (support components) that align the optical cores with micron-level precision, are unglamorous yet irreplaceable parts—and here, too, Japanese companies command a global presence. The more widely optics spread, the more the demand for these connection components snowballs.
Japan Aviation Electronics Industry (JAE) is a major manufacturer handling a wide range of connectors—from PCs and mobile devices to automotive and industrial equipment—and it has accumulated technology in the field of multi-fiber optical connectors such as MPO/MT, which are indispensable for the high-density wiring of data centers. It has also laid the groundwork with an eye toward optoelectronic fusion: in October 2023, it entered into a capital and business alliance with AIO Core, a startup that handles compact components for converting optical signals and electrical signals, for the development of automotive components. Meanwhile, its current results show that although net sales for the fiscal year ending March 2026 rose 3% year on year to 227.8 billion yen, soaring raw material prices and the costs of launching new products squeezed profits, with operating profit falling sharply by 43% to 8.9 billion yen. This reveals a transitional phase in which the cost factor of high material prices and the growth factor of optical adoption are in tension.
Hirose Electric is known for its high-functionality micro-connectors touted as ultra-compact, low-profile, and high-speed, and it supplies various high-speed connectors as well as waterproof connectors for optical fibers and high-frequency coaxial use, aimed at the high-speed networks of data centers. The company, too, formed a capital and business alliance with AIO Core in September 2024, embarking on the joint development of optics-compatible connectors for use in automobiles and communications equipment. The fact that the two leading connector makers, Japan Aviation Electronics and Hirose, both invested in the same optoelectronic-fusion venture is suggestive. Rather than finished products, they are moving early to secure the "standards and components of connection"—a strategy befitting upstream players seeking to claim a share of the initiative at the turning point from electricity to light.
Masters of the Upstream ③—High-Precision Manufacturing and Inspection Equipment (Disco, Advantest)
What supports the manufacturing front lines of optoelectronic fusion are the equipment makers that "grind, cut, and inspect" tiny chips. Here, Japan boasts overwhelming strength.
Disco is the master of back-end process equipment, holding a 70–80% global share in dicing equipment that cuts apart semiconductor wafers and a 50–70% share in grinders that shave them thin. It has a strong reputation for nano-level processing technology, including its proprietary "stealth dicing," which uses lasers to create internal flaws in a wafer so it can be split apart. In optoelectronic fusion, there will be a growing need to process new and brittle materials other than silicon—such as glass interposers, silicon photonics chips, and InP—without damaging them. In fact, the company's dicing saw "DFD6450" also supports the cutting and grooving of glass for optical components. Processing chips that are thinned to the extreme, like HBM (high-bandwidth memory), is also the company's exclusive domain, and the tailwind is strong. Business performance is excellent: revenue for fiscal 2025 (April 2025 to March 2026) rose 11.1% year on year to ¥436.8 billion, surpassing the ¥400 billion mark for the first time and reaching a record high, while operating profit of ¥184.9 billion and net profit of ¥135.5 billion were also record highs, marking a sixth consecutive year of record earnings. For the April–June 2026 quarter as well, net profit is expected to rise 24% year on year to ¥29.5 billion, a record high for that quarter. Optoelectronic fusion will further expand Disco's role through the increase in new materials.
Advantest stands at the top of the world in "testers (inspection equipment)" that determine whether manufactured chips work correctly. Against the backdrop of surging demand for high-performance semiconductors and HBM for generative AI, revenue for the fiscal year ending March 2026 rose 44.7% year on year to ¥1,128.6 billion and operating profit rose 118.8% to ¥499.1 billion—both record highs—and its share of the tester market reached 65% (up 7 points year on year). For the fiscal year ending March 2027, it has set an even more bullish target of ¥1.42 trillion in revenue. Advantest matters to optoelectronic fusion because the "good-product determination" for CPO is orders of magnitude more difficult. Light is extremely sensitive to slight positional misalignment, and testing CPO, in which electricity and light coexist, has been pointed out as one of the biggest bottlenecks to widespread mass production. Integrated optical-electrical inspection technology using optical monitors and the like will become a new growth frontier for Advantest, the strongest player in testers.
Global Development Competition and the Silicon Valley VC Perspective
Let us now shift our perspective to the other side of the Pacific. The initiative over the "finished products" of optoelectronic fusion still rests with the giants of the United States and Taiwan. Broadcom has rolled out the "Tomahawk 6 Davisson," a 102.4-terabit-per-second Ethernet switch that represents its third-generation CPO. As the core of the "Vera Rubin" platform unveiled at GTC 2026, NVIDIA is introducing the CPO-equipped "Spectrum-X Photonics" Ethernet switch and the "Quantum-X Photonics" InfiniBand switch, with the former expected to ship in the second half of 2026 and the latter in the first half of 2026. NVIDIA claims that, compared with conventional pluggable transceivers, CPO improves power efficiency by 3.5 to 5 times, boosts reliability (link uptime) tenfold, and reduces the number of lasers to one-quarter. On the manufacturing front, Taiwan's TSMC silicon photonics platform "COUPE" will enter mass production in 2026, Samsung is targeting a CPO turnkey offering in 2029, and Intel still remains in the R&D stage. Silicon photonics modules are projected to account for more than 50% of the optical transceiver market in 2026 (up from about 33% in 2024).
Silicon Valley VC money is pouring into this gold rush in a torrent. Optical I/O pioneer Ayar Labs raised a $500 million (approximately ¥78 billion) Series E in March 2026, reaching a valuation of about $3.75 billion (approximately ¥580 billion). Its investors include, in addition to the corporate arms of NVIDIA and AMD, Sequoia-affiliated funds, Insight Partners, ARK Invest, and the Qatar Investment Authority, while its early backers include the renowned deep-tech VC Playground Global. Photonic computing firm Lightmatter raised $400 million (approximately ¥62 billion) to reach a valuation of $4.4 billion (approximately ¥680 billion), bringing its "Passage" optical interposer to market. As for Celestial AI, Marvell acquired it for about $3.25 billion (approximately ¥500 billion, rising to a maximum of $5.5 billion = approximately ¥850 billion upon meeting performance milestones) before it could list as an independent company. In February 2026, Mesh Optical Technologies raised $50 million (approximately ¥7.8 billion) in a round led by Thrive Capital, with an investment thesis that was explicitly the "mass production of American-made optical links."
Taking a bird's-eye view of these moves, the worldview of Silicon Valley VCs becomes visible. First, they have resigned themselves to the conviction that "Copper is dead." This is the belief that, as long as AI keeps growing, the shift to optics is inevitable. Second, the center of gravity of investment has clearly moved from "research-oriented ventures" to "companies with realistic mass-production roadmaps." Third, the idea of turning geopolitics into value—"China-free" and "American-made"—has become a factor in investment decisions. And fourth, and this is decisive for Japan, the value that VCs are fighting over most fiercely actually resides in the "upstream pressure points" of light sources, substrates, packaging, and inspection. Both Ayar Labs and Lightmatter ultimately have no choice but to rely on Japanese and surrounding components and equipment for their light sources, substrates, and inspection. When Nikkei xTECH wrote that "Japan has a strength in light sources," and when a certain practitioner described it as "a structural path to victory in which, rather than the finished product, you hold the most critical component of AI infrastructure," they were pointing to this very structure. Conversely, Japan's weaknesses are the "void of mass-production manufacturers" and "not holding decision-making authority over design and packaging," and whether NTT and Rapidus can fill that void will be the watershed determining whether Japan can be promoted from a "parts supplier" to the "leading player in packaging."
Reporting from each newspaper and institution, and the timeline ahead
Coverage of optoelectronic fusion (photonics-electronics convergence) heated up rapidly entering 2026. Nikkei Tech Foresight and Nikkei xTECH have run a succession of features such as "The Optoelectronic Fusion Industry Map: Where 37 Companies Stand," "The Industry Resembles a Bubble; Japan's Strength Lies in Light Sources," and "China Leads in Next-Generation Materials," carefully portraying the overall picture of the industry along with Japan's strengths and weaknesses. In a report dated March 31, 2026, Mizuho Bank's Industry Research Department argued that "the winning formula lies in a trinity approach, advanced data center development, and the cultivation of services unique to optoelectronic fusion," while Toyo Keizai Online posed the question of whether NTT "can establish a presence as a device maker through its greatest strength, 'optical' technology." Optoelectronic fusion took center stage at international conferences as well: at OFC 2026, held in Los Angeles, USA in March 2026, engineers from OpenAI and Samsung Electronics took the stage, and exhibits of optical chiplets electrified the venue. In securities firms' analyses of related stocks, Sumitomo Electric and Furukawa Electric are positioned as the "high-probability group certain to grow earnings in AI data center optical wiring," while Hamamatsu Photonics, Advantest, and Lasertec are positioned as the "unsung kings" that profit from the manufacturing, inspection, and packaging of optical semiconductors.
When it comes to market size forecasts, it is important to note that the figures vary widely depending on the research institution. Nikkei reports that the key components of optoelectronic fusion will become a market of roughly ¥8 trillion in 2035 (8 times the 2025 level), while IDTechEx puts the global market for photonic integrated circuits (PICs) at $54.5 billion (approximately ¥8.56 trillion) in 2035, and Mizuho Bank estimates the global market for optoelectronic fusion devices at ¥1.8 trillion in 2035. Because the order of magnitude changes depending on "what you count in the denominator," it is wise to grasp the figures as a range rather than taking any particular number at face value.
Future developments are best "tracked" along roughly the following timeline. 2026 is the year the starting gun for industrialization sounds, with NTT's commercialization of PEC-2, NVIDIA's shipments of Quantum-X (first half) and Spectrum-X (second half), volume production of TSMC's COUPE, the launch of Rapidus's back-end pilot line (April), and the start of mass production of 1.6-terabit optical modules all converging. 2027 is expected to see the start of Rapidus's 2nm mass production, a rush of CPO products at the semiconductor level, and the next phase of the IOWN Global Forum. 2028 is a milestone year in which Rapidus's multi-generation back-end mass production, the completion of the LSTC Chitose site adopted by NEDO, NTT's commercial samples of PEC-3, and the ramp-up of Sumitomo Electric and Furukawa Electric all peak. Around 2030, the focus will be on the realization of IOWN/6G and the spread of China-led next-generation materials such as TFLN into CPO, and around 2032, the final stage of bringing light inside the chip itself (reducing power consumption to one-hundredth) comes into view. Concrete observation points include the proportion of optical and data center business in each company's earnings, supply-and-demand trends for InP substrates, the construction progress of NEDO sites, and announcements at international conferences such as OFC, ECOC, and ECTC.
Conclusion: Japan's Path to Victory—"Seize the Vital Point, Not the Finished Product"
Optoelectronic fusion (photonics-electronics convergence) is, at present, the most realistic solution for breaking through the "power wall" that AI has run into. That is precisely why it is the foundation of infrastructure for the AI era, and precisely why it is positioned as a top priority for economic security. Within this larger picture, Japan's position is paradoxical. When it comes to finished products such as switches and GPUs, and to leadership in design and mass production, Japan lags behind the American and Taiwanese giants. And yet, upstream, Japan holds many of the "vital points" that no one can replace overnight: indium phosphide, which forms the basis of laser light sources (Sumitomo Electric); lasers for CPO (Furukawa Electric); multi-fiber optical connectors and ferrules (Japan Aviation Electronics, Hirose Electric); and the equipment for micro-fabrication and inspection (Disco, Advantest).
To sum it up in Silicon Valley VC terms, this is a structure in which "even if you can't seize the platform, the picks and shovels belong to Japan." In the gold rush, the ones who most reliably made money were not those who dug for gold, but those who sold picks and jeans—and that old adage applies directly to the modern gold rush that is AI infrastructure. When NVIDIA and Broadcom build colossal fortunes on light, the ones who ultimately generate that light, connect it, cut it, and inspect it are, after all, Japan's component and equipment makers.
That said, this path to victory carries clear risks. First, the "void of mass-production makers"—even with strength in upstream components, Japan is short on finished-product players who can bundle them together and mass-produce at a global scale. Whether NTT and Rapidus can fill this void is the question. Second, there is the upstream risk in materials and next-generation materials such as InP and TFLN; depending on geopolitics, even Japan's strengths could be swallowed up by supply constraints. Third, there is the structural weakness of not holding the decision-making power over design and implementation. Taking these into account, 2026 to 2028 will be the true critical moment that determines whether Japan can climb a rung up from "the unseen parts supplier" to "the protagonist of implementation." NTT's commercialization of PEC-2, Rapidus's back-end-process strategy, and the optoelectronic fusion hub of NEDO and LSTC clustering in Chitose—whether these three mesh together will determine whether Japan can evolve, in optoelectronic fusion, the very heart of economic security, from "a country that holds the vital points" into "a country that decides the flow."