Goldman Sachs Says Optical Networking Is AI's Next Trillion-Dollar Opportunity. Lumentum May Be the Biggest Winner

Goldman Sachs argues AI capex is shifting from GPUs to AI networking, projecting AI networking value per computing unit rising from ~$315k (GB300) to ~$9.4M (Rubin Ultra), ~29x higher. It also forecasts the AI networking TAM growing from ~$15B to ~$154B by 2028, with co-packaged optics at ~$91B (~59% of spend) and pluggable optics seeing ~10x growth in Goldman’s framework. The article positions Lumentum (LITE)—a near-pure optical components supplier—as a high-leverage beneficiary of the optical buildout, though it flags CPO adoption timing and hyperscaler qualification/CapEx discipline risks.

Analysis

The market is likely to overreact to the headline TAM expansion while underestimating the real constraint: qualification speed and yield. That makes LITE the highest-beta expression, but also the most timing-sensitive; the first leg of the trade is multiple expansion on narrative, while the second leg requires actual order conversion over the next 2-4 quarters. If hyperscaler capex stays elevated, optical suppliers can see operating leverage faster than diversified peers because mix shifts into higher-value modules and lasers rather than just unit growth.

The competitive read-through is more nuanced than a simple "optics up, GPUs down." NVDA still benefits if larger clusters are built, but networking intensity can also create temporary bottlenecks that delay node ramps and distort inventory at the system level. AVGO and MRVL should participate, but their diversified revenue bases will likely cap upside versus LITE; that argues for relative-value exposure rather than a blind sector basket. A second-order loser is any legacy interconnect chain exposed to copper-heavy architectures, especially if CPO adoption accelerates faster than expected.

The contrarian risk is that investors may be pulling forward 2028 economics into 2026 stock prices. If adoption slips, the market will punish pure-plays first because their valuations depend on a clean conversion from design wins to shipments, not just a larger TAM slide deck. The thesis is falsified if LITE does not show sequential backlog/revenue improvement over the next two earnings prints, or if hyperscaler networking capex commentary rolls over while compute capex holds steady.\n\n

Key Decisions for Investors

• Go long LITE on a staggered basis over the next 1-2 weeks; use weakness to build, not strength chase. Base case is a 3-6 month rerating if optical order cadence inflects; cut if the next two quarterly prints fail to show sequential improvement in orders or margin mix.
• Enter a relative-value pair: long LITE / short MRVL or AVGO in equal dollars to isolate pure optical beta from diversified AI exposure. This is a cleaner expression if the market starts pricing the whole networking complex too uniformly.
• Use NVDA as a hedge, not a short: if cluster buildout remains healthy, NVDA likely still captures the primary capex dollar even as optics gain share. Consider reducing NVDA only if management commentary shows network bottlenecks are delaying system shipments.
• Set an alert on hyperscaler capex guidance and LITE customer commentary over the next 1-3 months; if spend is disciplined or qualification timelines stretch, expect the stock to retrace quickly and the trade becomes a sell-the-rally setup.
• Avoid treating the report as an automatic buy signal for the broader AI basket; if the market rotates into AVGO/MRVL on the same theme, LITE likely remains the highest-upside but also the highest-volatility name.