Nvidia Adopts CPO for AI Data Centers as Eight Taiwanese Suppliers Eye 2026 Growth

Nvidia is moving to integrate Co-Packaged Optics (CPO) into its next-generation architecture to solve power and latency bottlenecks in AI data centers. As commercial deployment begins in late 2026, analysts have identified eight Taiwanese suppliers—including LuxNet, WNC, and Accton—poised to benefit from the shift toward light-engine integration.

The artificial intelligence boom has pushed traditional copper-wire data transmission to its physical limits. As data centers struggle with soaring power consumption and signal delays, industry leaders including Nvidia and Broadcom have turned to Co-Packaged Optics (CPO) and silicon photonics to maintain the performance required for massive AI training workloads.

From Concept to Commercialization: The 2026 Timeline

While CPO has been discussed as a future technology for years, the industry is entering a phase of actual commercialization, according to market analysis. The transition is expected to move from concept verification to mass production by the second half of 2026. A major driver of this shift is the adoption of TSMC’s COUPE platform by both Nvidia and Broadcom, which allows for the integration of light engines directly alongside switching chips.

Photo: Cnyes

The scale of this transition is significant. Market forecasts suggest that light engine shipments could reach the millions-to-ten-millions scale by 2027. This shift is not just theoretical; Nvidia’s roadmap, including the Vera Rubin platform, explicitly incorporates silicon photonics to achieve higher performance per watt. As analysts note, the move is designed to address the “AI traffic jam” where data centers currently operate like highways congested with millions of vehicles on a single lane.

Eight Taiwanese Suppliers in the CPO Supply Chain

As the industry pivots to CPO, focus has shifted toward Taiwanese manufacturers capable of handling the precise requirements of epitaxial growth, laser packaging, and optical fiber alignment. Analysts have highlighted eight key firms that are now seeing these technological shifts translate into tangible order books.

Photo: Chinatimes
  • LuxNet (3081): As a leader in Indium Phosphide (InP) epitaxy, the company is a critical supplier of light sources for silicon photonics. It has successfully turned to profit and is expanding capacity to meet demand from US-based cloud service providers.
  • WNC (3163): Known for its expertise in passive optical components, including splitters and fiber arrays, it is positioned to gain as CPO penetration increases.
  • Accton (2345): A global player in AI switches, the company is already seeing volume growth for 800G products and is actively positioning for the 1.6T market.
  • LuxShare (3450): The company has doubled its capacity for high-speed laser die packaging, with high-end products now making up a significant portion of its optoelectronics business.
  • Ascenx (3363): Specializes in Fiber Array Unit (FAU) technology and high-yield optical connectivity, components deemed essential for silicon photonics packaging.
  • VPEC (4991) and IET-KY (4971): Both firms are recognized for their deep technical expertise in compound semiconductors and high-end epitaxy, which are critical for the growing demand for 1.6T high-speed optical modules.
  • FOCI (4979): Focused on high-speed optical transceivers and active optical components, the company is currently seeing a recovery in its operations due to data center upgrade cycles.

Nvidia’s Market Performance and the Second Half of 2026

Despite a frustrating first half of 2026 for investors, the company holds a $1 trillion order pipeline for its Blackwell and Vera Rubin architectures. Nvidia’s financial outlook remains aggressive, with revenue projected to jump 82% this fiscal year to $392 billion.

Nvidia goes on investing spree to bolster supply chain empire

Analysts at Capital Group have stated that artificial intelligence cannot function without the physical economy.

This physical reality is driving massive capital expenditure. Hyperscalers such as Microsoft, Amazon, Meta, and Oracle are expected to spend well over $700 billion on capital equipment in 2026. According to analysts at JPMorgan, the upward revision of earnings estimates seen this year is a phenomenon typically observed only following an economic shock or post-recession recovery. While the market remains volatile, the consensus among financial observers is that the integration of CPO and silicon photonics is no longer a “dream” scenario but a foundational requirement for the next phase of AI infrastructure.

Related reading


Discover more from Archyworldys

Subscribe to get the latest posts sent to your email.