Broadcom Q3: Hardware Could Be The Missing Piece In An AI Portfolio

Broadcom (AVGO) is presented as a compelling buy on the back of AI hardware demand and valuation support, with the argument that AVGO’s customized ASICs and CPO (co-packaged optics) deliver superior energy efficiency versus GPU-based architectures. The article highlights AVGO’s Ethernet switch momentum, projecting Ethernet revenue to be ~3x InfiniBand by 2029, driven by AI inference workloads. Overall, the thesis is bullish and suggests favorable risk/reward rather than a specific earnings or policy catalyst.

Analysis

AVGO is the cleaner way to express AI capex durability if the market’s center of gravity shifts from raw compute scarcity to system-level efficiency. The key second-order effect is that networking, interconnect, and custom silicon typically become more important as hyperscalers optimize for inference economics and power density; that supports margin resilience even if GPU growth rates normalize. In that regime, AVGO can take share in the “picks and shovels for AI plumbing” bucket while NVDA remains the higher-beta compute expression.

The main risk is that the bullish narrative conflates architecture share gains with perpetual capex acceleration. If enterprise inference workloads migrate more slowly than expected, or if hyperscalers push back on power/network buildouts, AVGO’s revenue mix could still benefit but the multiple may not expand because the market will question the durability of the spend cycle. Watch for any evidence that custom ASIC wins are lumpy or that CPO/networking adoption is being delayed by thermal, supply, or interoperability issues; those are the first places the thesis breaks.

Consensus appears to be underpricing the portfolio effect: owning AVGO alongside NVDA is not pure diversification if both are fed by the same AI capex pool, but AVGO is less exposed to narrative de-rating when the market starts demanding proof of ROI. The contrarian call is that AVGO may be the better long even if AI spend cools modestly, because infrastructure-efficient names can win share in a slower-growth environment. What would falsify that view is a 1-2 quarter stretch where AI order growth decelerates and networking/custom silicon commentary softens at the same time.

Key Decisions for Investors

• Go long AVGO vs short NVDA on a 3-6 month horizon as a relative-value expression of AI efficiency over pure compute beta; target a 10-15% spread if the market rotates toward infrastructure winners.
• Add AVGO on any post-earnings pullback that is driven by multiple compression rather than order-book deterioration; use the next 1-2 earnings cycles as the catalyst window.
• Monitor hyperscaler capex commentary and networking/optics lead times closely; if those data points weaken, cut the AVGO overweight quickly because the thesis depends on sustained AI infrastructure buildout.
• If you want convexity, use AVGO calls financed by selling downside in NVDA only if GPU sentiment is already crowded; this keeps exposure aligned with the relative-value view and limits outright AI beta.