Conversation with Ayar Labs CEO Mark Wade on Silicon…

PodcastThe CircuitMark WadeSep 21

Mark Wade argues that interconnect performance is becoming a constraint on AI data-center scaling, and that optical links could address electrical bandwidth-distance limits across chips and racks. He describes the packaging, integration, and supply-chain challenges involved, alongside a potential scale-up optics market worth tens of billions of dollars and possibly close to $100 billion. Wade expects a varied mix of near-term solutions, with wider foundry-plus-OSAT adoption and GPU-company developments unfolding from 2028 onward.

AI Infrastructure Semiconductors Technology Stocks

Preview · 15 more points in the app

Read the Full Summary

Key Takeaways

Wade argues that interconnect performance can limit revenue scaling for large AI data centers, because electrical links face a bandwidth-distance trade-off as they move more data between GPUs and accelerators. He sees optical connectivity as a way to address that constraint across chip-to-chip, rack-scale, and rack-to-rack connections.

Wade says optical connectivity is not just a photonics-chip problem: the optical engine must combine a leading-edge electrical IC and photonics through 3D integration and advanced packaging, while also adding an optical connector. He identifies getting fibers connected and light on and off the chip as especially challenging.

Wade says large packages are nearing physical limits in bandwidth and connectivity radix, creating a need for another scaling option. His core bet is that optical connectivity will move into advanced packages and the computing fabric through CMOS-like foundry manufacturing and existing foundry and OSAT models.

Wade estimates that optical connectivity for GPUs, AI accelerators, optically connected memory, and scale-up switches could serve a volume 10 to 100 times larger than the datacom market, with scale-up optics potentially worth tens of billions of dollars and approaching $100 billion under reasonable average selling price assumptions.

TECHNOLOGY

Interconnect Limits — Wade argues that interconnect performance limits revenue scaling in large AI data centers. He says better connections are needed across chips, racks, and between racks to keep AI road maps on track.

“So let me try to start with a one or two sentence framing, which is that large scale AI data centers are limited in their revenue scaling by the performance…”—

Optical Connectivity — Wade says electrical interconnects face a bandwidth-distance trade-off: increasing data rates between GPUs and accelerators limits transmission distance. He argues optical connectivity can solve this limitation from first principles.

“And you've got this issue in electrical interconnects, is that the faster you want to go, so that, you know, the more data you want to send between these GPUs…”—

Integration Challenge — Wade says a complete electronic-photonic solution combines technology domains that telecommunications historically split among specialized companies. Bringing them together in a working end-to-end product is part of the challenge.

“To really get a full end to end electronic and photonic solution to work. You're talking about the combination of multiple domains of technology and expertise…”—

MARKETS

Scale-Up Market — Wade estimates that optics for the scale-up opportunity could represent a market worth tens of billions of dollars, potentially approaching $100 billion, based on reasonable average selling price assumptions.

“You know, I won't give you our exact ASP targets, but, know, you do some reasonable ASP targeting. It's…”—

FORECASTS

Near-Term Mix — Wade expects a chaotic mix of technologies and form factors in the short term as companies pursue 2027–2028 stopgap solutions. He says solutions also need a supply chain that can ramp to the volume required for rack-scale deployments.

“You know, it forces everyone to scramble to find, you know, well, what's my twenty twenty seven, twenty twenty eight, you know, kind of stop gap solution going to be?…”—

Foundry-OSAT Shift — Wade expects silicon photonics to move into a foundry-plus-OSAT flow over the longer term. He says this could simplify the customer problem once high-volume delivery is proven, and describes the longer-term horizon as 2030 and beyond.

“But I think in the longer term, we're gonna land, you know, this new generation of silicon photonics into a, you know, TSMC like foundry plus OSAT flow. And we'll…”—

AI Interconnect Bottleneck

The hosts introduce the guests and frame interconnect as a growing constraint on AI compute, while Wade says scaling data-center performance requires optical links to overcome electrical bandwidth-distance limits.

Why Optical Is Difficult

Wade explains that optical products require expertise across photonics, electronics, materials, manufacturing, and supply chains, making end-to-end integration unusually challenging.

AI Creates Optical Demand

Wade says large-scale AI created a high-value market whose growth depends on connectivity, giving optical interconnect the commercial demand that earlier telecom and datacom markets lacked.

Co-Packaged Optics Manufacturing

Wade describes the added difficulty of combining optical engines with leading-edge chips and advanced packages, including the need to develop scalable testing and production processes.

Near-Term Interconnect Mix

Wade expects a varied set of interim technologies and form factors as customers seek solutions before optical connectivity becomes a high-volume foundry and OSAT offering.

Digital Twin Manufacturing Model

Wade describes using coding agents to model the full manufacturing flow, where detailed production and test choices could improve yields and enable volume scaling.

Ayar Labs' Foundry Bet

Wade says the company’s core thesis is that optics will move into compute packages and must be manufactured through leading-edge foundry and advanced-packaging supply chains.

TSMC COUPE Integration

Wade argues that TSMC’s value lies in delivering a complete advanced package and scalable ecosystem, not simply a photonics wafer, while noting fiber coupling remains a key challenge.

Customer Adoption Timeline

Wade says leading GPU companies are advancing adoption, while broader accelerator customers will need proven, low-risk supply and integration paths, especially through ASIC partners.

Ayar Labs' Investor Strategy

Leiter explains how strategic and institutional investors have helped fund development and build commercial relationships, while the team prepares for a possible IPO path.

Scale-Up Optics Opportunity

Wade estimates the scale-up optics opportunity could reach tens of billions of dollars and approach $100 billion, with a potential deployment and IPO window around 2028–2029.

Investor Questions And Visibility

Wade says investors are increasingly probing manufacturing details, while successful execution could provide stronger visibility into revenue ramps and make the eventual IPO story more financially grounded.

AI Cluster Scaling Outlook

The hosts emphasize that large unified compute domains remain an early-stage need for AI, thank the guests, and preview a future discussion of lasers.