The Optical Fiber Communications (OFC) trade show in late March this year featured a focus on optical communications roles in the future of artificial intelligence (AI) infrastructure, notably coming on the heels what has been called “AI Woodstock,” the annual Nvidia GTC conference that took place a week earlier. The line to register for the OFC exhibits on March 26th was nearly reminiscent of OFC 1999 in Baltimore and there was more positive energy at the show in decades, preceding the optical communications business community darkness that came in the 2000s. Our recent solar eclipse at least was predictable and short-lived.
The Monday program at the Optica Executive Forum provided insightful, energetic talks and networking. Yet to serve the expected AI boom, the optical sector faces many challenges. The hyperscale datacenter and AI processing vendors call on the optical community for reduced power consumption, higher speeds-lower latency, better reliability, and smaller footprints not in incremental steps, but in 10x+ improvements and ASAP. The ask is obvious, but the ability to execute on this is questionable. How will efforts be funded to reach such goals—are the talents and production capabilities aligned?
For a perspective on industry dynamics, I presented that the “Big 6 of AI/Datacenters” (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia) have seen significant revenue and stock price gains in recent years and now have a combined market value of approximately $13 trillion, plus $500 billion of cash on their balance sheets with stock analyst projections to reach nearly $2 trillion in cash-investments in 2028. All have robust profit margins, with Nvidia gross margin in excess of 75%—an envy for hardware device makers.
The consensus among analysts, the hyperscale and AI processor and communications equipment vendors, and end-users is the optical component community is not moving fast enough and more innovation is needed to serve the AI and datacenter sectors. In spite of the progress of silicon photonics and photonic integrated circuits (PiCs) and the roadmaps for co-packaged optics (CPO), the theme in public talks and among private conversations is there is a long way to move from a status of TRL 6-7 to MRL 10 in VERY high volumes required to go live in datacenters and networks. Electronics, ethernet, and software keep marching forward at hyper-speed while optics is often mired at academic speed and incumbency think vs. innovation acceleration.
In contrast to the “Big 6” noted, if we look at 6 of the larger optical communications suppliers (Lumentum, Coherent, Ciena, Infinera, Corning, and Cisco), we see flattish to even declining market values the past year and much lower margins that don’t attract much investor enthusiasm. Nvidia, having bought Mellanox in 2021 for its Infiniband networking capabilities when its revenue was $1.7 billion per year, has ramped it to a run rate of $17 billion projected in 2024. During OFC, we witnessed the IPO of Astera Labs, which provides connectivity products for high-performance cloud and AI infrastructure. As a private firm, it had raised investments from Nvidia’s past VCs plus Fidelity, raising over $700 million in the IPO. Having supply agreements directly to Amazon, it now has a market value of over $10 billion; i.e., greater than Lumentum, Coherent, Infinera, and Ciena.
A question might be why couldn’t or didn’t any of the incumbent optical communication vendors innovate both product and business models to enter this market opportunity? Did they lack the skills, customer/market insights, or the capital? Similarly, why has Microsoft acquired IP and built its own hollow-core fiber plant rather than work with long-established vendors in fiber? In private dialogs and cocktail chatter during the week, I heard several comments on optical sector “C-level” management being “stodgy,” lacking innovation, and tired. This quote from one executive tells the story that others echoed: “While I’ve been in the industry a long time and enjoyed it, I would not want my 20-30 year-old kids or friends to join any of these established optical comm firms.”
The fundamental observation that came out of these OFC sessions is about market power and how as a supplier, one needs to understand and adapt. The hyperscalers, the communications user communities, and the largest investors in the world are driving optical vendors to “show me your better selves,” not tell me you are from Missouri and can’t do what we want. Like the state of Missouri, the optics sector can be both stubborn as a mule and conservative in making moves in innovation and ramping scale to meet customers. AI and datacenter customers are aggressive and innovative in comparison to the historic telecom network customers of the optical communications ecosystem. If we as customers have to, we will do it ourselves, by hiring your best and brightest from across the world, funding new companies with the best and brightest, and creating new operating models and supply chains. We have the cash to invest and the margins to absorb near-term expenses.
If you can’t serve us in new ways, be prepared to be relegated to a legacy business model and not be among the innovators that win new high-volume opportunities. We will find and support the innovators coming out of universities and labs around the world and given our volumes, we either need to have them get in corporate hands that can scale reliably or do it ourselves.
