📷 The last photo you took on your phone? More than half the time, the chip that turned that light into a digital signal was made by Sony. Sony commands over 51% of the global smartphone CMOS image sensor market — but its throne is finally wobbling, with Samsung breaking into Apple's iPhone supply for the first time. Then on May 8, 2026, Sony chose its counter-move. Not "go it alone." Not "double down on the US." Instead: a joint venture with Taiwan's TSMC. In an era when the US CHIPS Act, EU Chips Act, and Korea's K-Chips Act are redrawing borders around silicon, Japan and Taiwan are quietly building something different — a "silicon alliance."

What was announced on May 8

On May 8, 2026, Sony Group released its full-year earnings and dropped a major announcement alongside it. Its subsidiary Sony Semiconductor Solutions (SSS) and Taiwan Semiconductor Manufacturing Company (TSMC) — the world's largest contract chipmaker — signed a non-binding memorandum of understanding to form a strategic partnership for next-generation image sensors.

The key terms:

  • A joint venture (JV) in which Sony will be the majority and controlling shareholder.
  • Volume production at Sony's newly constructed fab in Koshi City, Kumamoto Prefecture, where development and production lines will be set up.
  • Additional phased capital investment by Sony at its existing Nagasaki plant, separate from the JV.
  • Premised on Japanese government support — METI has already approved up to 60 billion yen (~$380 million) for Sony's new Koshi fab, and the JV is expected to receive further public backing.

Sony President and CEO Hiroki Totoki told the earnings call he wants to "explore new business opportunities in automotive and physical AI."

The "next-generation image sensors" the JV envisions aren't simply a higher megapixel count. The plan combines Sony's strength in sensor design (the photon-capturing layer) with TSMC's leadership in advanced process technology (the AI logic layer), 3D-stacked at nanometer scale — the so-called "sensor + logic 3D stack" architecture.

Why Taiwan, not in-house — and not Samsung or Intel?

CMOS image sensors are split into a pixel layer (which captures light) and a logic layer (which processes the signal). Sony has long built both layers internally. Why outsource the most advanced part of the logic layer now?

Because the AI era has shifted the competitive axis from pixels to compute.

The "eyes" of modern smartphones, robots, and self-driving cars don't just capture — they recognize. They run AI inference on each frame in real time, distinguishing pedestrians, reading text, classifying objects. To do that efficiently, you need a cutting-edge logic chip stacked directly under the sensor, processing signals at the pixel level. And only three companies in the world can mass-produce leading-edge logic at 5nm, 3nm, and the looming 2nm node: TSMC, Samsung, and Intel.

If Sony tried to do the logic in-house at leading nodes, it would face trillions of yen in capex and unfamiliar yield risk. Pairing with Samsung means handing the playbook to a direct competitor — Samsung Foundry sits inside the same conglomerate as Samsung's own image sensor business. Intel has been struggling with the ramp of its 18A node through 2025–2026, raising questions about volume reliability.

That leaves TSMC: a pure-play foundry that doesn't sell sensors, has the most advanced process in the world, and — crucially — already has JASM (its joint venture with Sony, Denso, and Toyota) operating Fab 1 in Kumamoto, with Fab 2 upgraded to a 3nm-class node. The Sony Koshi fab and JASM Kikuyo fab are roughly a 15-minute drive apart, which makes pairing TSMC's logic wafers with Sony's sensors logistically ideal.

In other words: this isn't surrender of "in-house pride." It's a rational outsourcing call to keep up with the speed of the physical-AI market.

Sony's throne — 51% share, but the floor is shaking

Sony made this bold move because the throne it has held for over a decade is finally tilting.

According to research firm TechInsights, Sony held over 51% share of the smartphone image sensor market in Q2 2025, ranked first globally. Samsung System LSI was second, OmniVision third. Across the broader CMOS image sensor market — smartphones, automotive, industrial — Sony, Samsung, and China's GalaxyCore together control roughly 62%.

But Sony's largest customer, Apple, has shown the first cracks in its long-standing single-vendor sourcing of Sony sensors. According to a Nikkei Business report in early 2026, Samsung is expected to begin supplying Apple with CMOS image sensors as soon as 2027. If that holds, it would be the first time Sony has ever shared the iPhone sensor slot.

Samsung announced in early 2025 a three-layer stacked image sensor — separating the photodiode, transfer, and logic layers — explicitly aimed at the iPhone 18 generation. China's GalaxyCore and SmartSens are pushing aggressively in the mid-range. The market is tilting.

By building the Koshi fab and pairing with TSMC for sensor + logic 3D stacking, Sony is rebuilding its technical edge for the next decade — assuming Apple exclusivity is gone.

The bigger picture: Japan's all-of-government chip strategy

The Sony Koshi fab is itself the recipient of up to ¥60 billion (~$380M) in METI subsidies announced in April 2026. That figure, however small in absolute terms, is one piece of a much larger campaign.

Here is the cumulative scoreboard of Japan's semiconductor subsidies announced since 2022:

Recipient Cumulative subsidy (announced)
Rapidus (Hokkaido / 2nm logic foundry) ¥2.354 trillion (~$15B)
TSMC Kumamoto (Fabs 1 & 2 combined) up to ¥1.208 trillion (~$7.7B)
Micron Hiroshima (DRAM) up to ¥500 billion (~$3.2B)
Kioxia / Western Digital (NAND) up to ¥243 billion (~$1.5B)
Sony SCK Koshi fab (CMOS sensors) up to ¥60 billion (~$380M)

Total cumulative committed support is in the ¥3 trillion range. In absolute terms it lags the US CHIPS Act ($52.7B in direct funding plus ~$24B in tax credits), but as a share of GDP, Japan's effort is among the most concentrated on the planet.

Image sensors get a comparatively small slice. The symbolism is outsized: keeping the "eyes of physical AI" anchored on Japanese soil — for smartphones, cars, robots, drones, factories, and medical devices.

How it compares: CHIPS Act, EU Chips Act, K-Chips Act, and friends

United States — CHIPS and Science Act (enacted 2022; $52.7B direct + ~$24B tax credits). Targeted at major manufacturing builds: TSMC Arizona (~$6.6B), Intel (up to $8.5B), Samsung Texas (~$6.4B). 25% investment tax credit for fabs that begin construction by year-end 2026. Under the Trump administration, the political tone around the act has soured and contract terms are reportedly being renegotiated. Building cost remains 30–50% higher than Asia even after subsidies.

European Union — Chips Act (in force 2023; ~€43B aggregate target). Funds ESMC (TSMC + Bosch + Infineon + NXP, Dresden, ~€10B), Intel Magdeburg (~$11B in German subsidy), and others. Direct EU budget contribution is only €3.3B; the rest depends on member-state aid and private investment. In April 2026 the EU institutions agreed a Joint Roadmap pointing to a Chips Act 2.0 in Q2 2027.

South Korea — K-Chips Act. Up to 15–25% investment tax credit for large firms (raised from earlier levels), plus streamlined permitting for the 300-trillion-won (~$200B) Yongin mega-cluster anchored by Samsung. Korea's policy mostly defends domestic champions (Samsung, SK hynix), with little success attracting foreign entrants.

Taiwan — Statute for Industrial Innovation (Jan 2023). Up to 25% R&D tax credit, plus historic government support of 50% of land cost and 45% of facility cost for fabs. The strategy: keep the "sacred mountain" of leading-edge fabrication on the island.

China — National IC Industry Investment Fund ("Big Fund"). Phases 1–3 cumulative roughly $98B, covering design, manufacturing, equipment, and materials. Used as a counterweight to US export controls.

Japan. A two-pillar approach — recruiting foreign leaders (TSMC Kumamoto) while building a domestic foundry (Rapidus) — overlaid with targeted subsidies for areas where Japan already has world-leading position (CMOS sensors, power semis, materials, equipment, advanced packaging). Smaller in absolute dollars than the US or China, but more focused.

"Silicon Shield" — and the quiet rise of a Japan-Taiwan alliance

This deal isn't only commercial. It carries geopolitical weight.

Taiwan produces over 90% of the world's leading-edge logic (sub-5nm). That concentration has been called the "Silicon Shield" — the argument that any Chinese move on Taiwan would be so catastrophic for the global economy that it acts as deterrence in itself. Bloomberg Economics has estimated a Taiwan war could cost the world $10 trillion in the first year.

The US (CHIPS Act) and EU (Chips Act) are explicitly trying to disperse that shield by attracting TSMC fabs onshore. Japan's posture is different. Rather than competing with Taiwan or extracting its know-how, Japan is interlocking with it: JASM Fabs 1 and 2 in Kikuyo, the Sony-TSMC JV in Koshi, deepening collaboration on materials and equipment, and now joint volume production of sensor + logic stacks.

For Taiwan, Japan offers a friendly second site that doesn't compete with TSMC's home base. For Japan, Taiwan offers ongoing access to the leading-edge logic Japan can no longer build alone. The mutual dependency strengthens deterrence in both directions. Some analysts informally call this emerging structure a "silicon alliance" or a "second-layer Silicon Shield."

Physical AI — the next battlefield

Totoki's mention of "physical AI" wasn't industry jargon; it pointed to where the action is heading.

Generative AI's first wave was about data centers — GPUs and HBM. The second wave is about physical-world AI: robots, cars, drones, industrial machines, all of which need sensors (eyes) tightly fused with inference chips (brains). 3D-stacked sensor + logic is exactly that fusion in silicon.

If Sony and TSMC succeed, the JV's output will end up in self-driving ADAS cameras, humanoid robot eyes, high-speed factory inspection systems, and Earth-observation satellites. If they fall behind, Samsung's 3-layer stack could become the standard — and Sony risks being trapped as "large-share but legacy-tech."

Reactions inside Japan

Domestic reaction is layered. Some welcome it as a savvy alliance: pairing world-best design with world-best logic without bleeding capital chasing in-house leading-edge fabs. Others worry that putting downstream logic in Taiwanese hands deepens dependency on a single chokepoint. People in Kumamoto see another tailwind for the regional economy after JASM, but worries about water use, land prices, and labor competition continue. A familiar critique runs underneath: even with the world-leading 51% share, the image sensor market is far smaller than the AI chip market where Japan has almost no presence.

What about in your country?

"Build it at home, or partner with allies?" — almost every government is now wrestling with this question. The US piles on subsidies; the EU struggles with member-state coordination; Korea protects its champions; Taiwan defends the leading edge; China pursues self-sufficiency. Japan is taking a slightly different path — quietly tying its silicon future to Taiwan.

How does this look from where you are? What strategy is your country pursuing in semiconductors? Should governments subsidize domestic manufacturing, or bet on division of labor with trusted partners? We'd love to hear your view.

References