When I first started tracking US stocks from Hong Kong, I made the mistake of treating NVDA and AVGO separately. NVDA for AI exposure, AVGO as a dividend holding. Separate mental buckets.
That framing was wrong, and it cost me a few months of clarity.
The duopoly case
NVDA designs the GPUs that train and run large language models. AVGO makes the custom ASICs (Google TPU chips, Meta's MTIA chips) and the high-speed interconnects that link those chips together at scale. They're not competing -- they're working on different layers of the same infrastructure stack.
When a hyperscaler spends $50 billion building out AI compute in a year, both companies benefit. NVDA gets GPU orders. AVGO gets ASIC orders plus the networking hardware to connect everything. The revenue exposure is correlated, but the mechanisms are different enough that holding both gives you different risk profiles inside the same theme.
Four numbers, two for each
NVDA: Price-to-sales around 20x forward. The valuation assumes datacenter GPU demand continues at the 2024-2025 pace, which is the main risk. If hyperscaler capex slows, that multiple compresses fast.
AVGO: Forward P/E around 26x, with a dividend growing at roughly 15% annually. The dividend creates a floor psychology -- income-focused investors hold even in volatility, which damps drawdowns compared to NVDA.
Neither is cheap by traditional metrics. The question is whether AI infrastructure spending is durable, and over what timeframe.
Allocation math for HK/TW investors
For HK investors buying US stocks via a local broker: NVDA and AVGO are both accessible through Interactive Brokers HK, Futu Securities, and Tiger Brokers. The dividend withholding math differs between them.
NVDA pays about $0.04/share quarterly -- tiny relative to share price. Withholding on that is almost noise.
AVGO pays about $23.50/share annually (as of 2025). The 30% US withholding tax on dividends -- standard for HK residents without a W-8BEN or equivalent -- takes a meaningful bite. On a $10,000 AVGO position, roughly $35/year after withholding versus $50 pre-withholding. That changes the income math if you're holding for yield.
My current allocation: roughly 60% NVDA, 40% AVGO within this pair. Higher NVDA weight for upside optionality if AI compute demand continues scaling. AVGO for the income component and partial volatility damper.
What could actually break this pair
The bear case isn't "AI doesn't work" -- it's "customers build their own chips." Google already makes TPUs and buys AVGO ASICs. If Google internalizes more ASIC work, AVGO's custom silicon revenue narrows. If AMD closes the performance gap with H100/H200/B100, NVDA faces market share pressure.
The scenario where both underperform simultaneously: hyperscaler capex rotation -- they've spent heavily and are now in a digestion phase. This happened briefly in 2023. It could happen again.
What doesn't break it
"Nvidia is overvalued so both will fall" -- AVGO has a different multiple and a dividend. They don't always move together. AVGO's drawdowns in the last two years have been meaningfully smaller.
"AVGO is a legacy semiconductor company" -- AVGO's networking business (Ethernet switching, optical interconnects) is more relevant to AI infrastructure than it was five years ago. The custom ASIC revenue is growing, not declining.
I'm not recommending either position. This is how I think about the pair for my own portfolio, as someone managing investments in HK and tracking AI infrastructure themes. Your tax situation, broker access, and risk tolerance will differ.
The core point: treating NVDA and AVGO as a coordinated pair rather than separate decisions changes how you think about entry points, volatility, and income -- especially with the withholding tax asymmetry that matters specifically for HK/TW investors.
Top comments (0)