I missed Nick Harris’s keynote at the 33rd Harvard Business School Tech Conference—it sounds like it was a thought-provoking session!
Photonics and the Future of AI
Photonics—using light to process and transmit data—is an exciting frontier, especially for AI, where speed and energy efficiency are bottlenecks. Traditional silicon chips, even with advances like LDP (Laser Direct Processing) or EUV lithography, are hitting physical limits as Moore’s Law slows. Harris’s work at Lightmatter focuses on photonic computing, which could sidestep these constraints. Light-based interconnects or even logic could slash latency and power use, making them ideal for AI workloads like neural network inference, where massive data movement is the norm. Imagine training models faster or running real-time AI on edge devices without frying batteries—that’s the promise here.
The discussion likely highlighted how photonics could complement or even surpass current tech. For instance, while Huawei might optimize LDP for mass production, photonics offers a paradigm shift—less about shrinking transistors and more about rethinking how computation happens. Harris’s vision, as seen on Lightmatter’s site, emphasizes this: their photonic chips aim to “accelerate AI with light,” which aligns with the post-Moore’s Law narrative.
Post-Moore’s Law Era
Moore’s Law—doubling transistor counts every two years—has driven tech for decades, but it’s stalling due to physics (heat, quantum effects) and economics (fabs costing billions). Post-Moore’s Law doesn’t mean progress stops; it means we pivot to alternatives like photonics. Harris and Wu probably explored how this shift isn’t just theoretical—Lightmatter’s already building hardware, like their 3D-stacked photonic engines. This could democratize high-performance computing, letting startups leapfrog giants stuck in silicon paradigms. The catch? Adoption hinges on integrating photonics with existing systems, which isn’t trivial—think manufacturing challenges or software compatibility.
Adoption Challenges and Opportunities
Speaking of adoption, that’s the crux. Photonics is promising, but it’s early. Startups like Lightmatter could disrupt if they nail scalability and cost, especially for AI niches (think autonomous vehicles or generative AI). Huawei’s LDP focus might dominate mass markets short-term, but photonics could carve out high-value spaces long-term. Harris likely pitched a future where light-based chips are standard in data centers or edge devices by, say, 2030—optimistic, but plausible if R&D accelerates. Wu, as an HBS prof, might’ve grounded this with market dynamics: who invests, how fast can foundries pivot, and will big players (TSMC, Intel) resist or adapt?
My Take
I’m bullish on photonics as a concept—it’s physics-backed and solves real problems. AI’s hunger for compute power is insatiable, and electrons alone won’t cut it forever. That said, I’m skeptical of the timeline. The blog post I wrote speculated on hybrid chips (LDP silicon + photonic interconnects), which feels like a realistic bridge—pure photonic logic is harder to scale soon. Huawei’s efficiency edge with LDP might delay photonic adoption unless startups like Lightmatter outmaneuver them with killer apps. The real wildcard? Geopolitics—China’s tech push could force Western firms to double down on photonics as a differentiator.
What do you think—did Harris’s keynote vibe as more hype or substance? And are you leaning toward photonics as the next big thing, or do you see other tech (quantum, maybe?) stealing the spotlight?