NVIDIA invests $2B in Marvell to scale AI optical interconnects
This Photonics Spectra issue for June 2026 features various advancements in photonics technology, covering optical communication, SWIR imaging, transparent ceramics, and ultrafast lasers. Several companies, including Vector Photonics, Bruker-imec, Teledyne, Mojo Vision, and VIGO Photonics, announced new technologies, funding rounds, or acquisitions relevant to the photonics industry. The issue also highlights how photonics is pacing AI's growth through high-speed optical interconnects and new product launches from companies like Arista Networks, Marvell, and NLM Photonics.
Key Takeaways
- NVIDIA invested $2 billion in Marvell to secure a long-term role in the scale-up and scale-out AI networking stack
- Marvell launched the Structera S 30260, a 260-lane CXL switch designed to expand data center memory bandwidth
- Eoptolink debuted a 6.4T near-packaged optics (NPO) module for high-density AI data center interconnects
- Vector Photonics completed the first public demonstration of PCSEL technology for real-world free-space optical communication
- Monarch Quantum secured a $55 million growth round to scale production of integrated photonics for quantum infrastructure
Why It Matters
The $2 billion NVIDIA-Marvell deal marks a pivot from treating photonics as a peripheral networking component to a core architectural requirement for AI clusters. As GPU clusters outpace the thermal and reach limits of copper cabling, integrated optical solutions like NPO and CPO must move from sampling to high-volume manufacturing. For competitive video delivery providers and data center operators, the adoption of these 800G and 1.6T transceivers will be the primary driver for reducing per-bit power consumption. Watch for Marvell's next quarterly earnings to see if AI-related bookings for 1.6T optics meet expectations.
Additional Context
The strategic alignment between NVIDIA and Marvell comes amid a structural shift in the semiconductor market where data transmission has replaced raw compute as the primary AI performance bottleneck. Per Investing.com (June 2026), NVIDIA's $2 billion stake has already seen significant valuation growth, reflecting market confidence in Marvell's role as a primary alternative to Broadcom in the custom AI accelerator (XPU) market. This partnership explicitly leverages NVIDIA’s NVLink Fusion platform, a rack-scale architecture that allows hyperscale clients to build semi-custom infrastructure while remaining locked into NVIDIA’s software stack and licensing protocols.
Simultaneously, the broader industry roadmap is accelerating toward 1.6T and 3.2T data rates to satisfy large language model (LLM) training requirements. At OFC 2026, several vendors confirmed that 1.6T pluggable modules have moved into confirmed mass production, per Hengtong Global (April 2026). While pluggable modules remain the standard for current upgrades, the emergence of 6.4T near-packaged optics from vendors like Eoptolink indicates that the industry is preparing for a transition to more dense, silicon-photonic integrated solutions that reside closer to the switching ASIC to minimize signal degradation.
Investment in the supporting hardware layer is also reaching a fever pitch outside of traditional data centers. For example, Monarch Quantum’s $55 million round highlights a growing trend of venture capital flowing toward 'infrastructure-as-a-product' in the quantum and photonics sectors. Per Quantum Computing Report (March 2026), Monarch is positioning its Quantum Light Engines to replace complex laboratory optical setups with manufacturable silicon. This movement toward industrial-scale photonics suggests that the technologies currently being built for AI networking will soon find secondary applications in secure quantum communications and subsea imaging systems.
Read full article at photonics.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source