Interconnects AI

Credo Technology Group and Oracle are working with the Open Compute Project community to advance standardization of optical transceiver telemetry. The collaboration aims to create common specifications for monitoring high-speed optical interconnects used in large data centers, improving interoperability and diagnostics across vendors' networking hardware.

The Advanced Photonics Coalition is promoting a set of manufacturing standards aimed at making photonics — using light instead of copper for data transfer in chips and data centers — cheaper and easier to produce at scale. The move targets a key bottleneck as AI-driven data center demand accelerates the shift toward optical interconnects, and industry watchers are debating whether standardization will genuinely lower costs and speed adoption.

Chip designer Marvell has proposed a $115 billion acquisition of optical manufacturer Lumentum, a deal that would bring chips, lasers and optical links under one roof. The move aims to solve a growing bottleneck in AI computing: faster processors are often left idle waiting for data, so high-speed optical connections are becoming as critical as the chips themselves.

Taiwan's chipmaker UMC has secured major orders from Chinese firms for AI optical chips, according to Businesskorea. The deal signals growing demand for optical interconnect technology used in artificial intelligence data centers, and highlights continued business ties between Taiwanese contract chipmakers and Chinese customers despite geopolitical tensions and export controls. Industry watchers will be looking at how cross-strait supply chains absorb rising AI-driven chip demand.

A Reuters analysis compares Marvell Technology and Lumentum Holdings, two chip and optical component makers, as competition intensifies to supply the optical interconnects and photonics needed for AI data centres. The piece examines how each company is positioned to benefit from surging demand for high-speed data transmission in artificial intelligence infrastructure.

Marvell is acquiring optical components maker Lumentum in an all-stock deal worth about $115 billion, paying roughly $1,132 per share, a 23.9 percent premium. The deal reflects Marvell's view that optical interconnects, not copper wiring, will become the critical bottleneck inside AI data centers, giving it control of the light-based technology underpinning AI compute infrastructure.

MCL, PVA TePla and ASE report that AI-based image enhancement improves detection of defects in flip-chip interconnects, the solder-bump connections that link chip packages to circuit boards. The companies say enhanced imaging makes tiny voids, cracks and misalignments easier to spot during inspection, supporting quality control as flip-chip packaging grows more common in advanced semiconductor production.

Nomura projects the market for optical laser chips will roughly triple to 1.5 billion units by 2028. The forecast points to surging demand tied to AI data centers and faster fiber-optic communications, where optical interconnects are becoming central to scaling bandwidth. Analysts' outlooks like this are being watched closely by chipmakers and investors positioning around the AI infrastructure buildout.

The author of the Interconnects AI newsletter is publicly revisiting his earlier arguments about whether open-weight AI models pose meaningful cybersecurity risks. The piece asks whether fears that freely available models would materially help malicious hackers have held up, reopening a debate that has divided AI policy researchers between open-weights advocates and safety-focused critics.