If you own NVIDIA (NVDA), you already own the AI build-out leader. The question for 2026 is whether one name should carry your entire AI thesis.
Mega-cap concentration is a real portfolio risk. Spreading across Broadcom (AVGO), Taiwan Semiconductor (TSM), Marvell (MRVL), and ARM Holdings (ARM) covers more of the supply chain.
This piece walks through why a 4-name basket makes sense now, how each ticker fits, and how to size it across a small US-stock sleeve.
Why Add Beyond NVDA Now
Goldman Sachs forecasts that ASIC demand will catch up to GPU demand in AI data centers by 2027. That implies a 50/50 split between custom silicon and merchant GPUs. According to The Motley Fool, the breadth of AI-semi beneficiaries has widened well beyond a single GPU vendor.
The Big Five hyperscalers (Microsoft, Alphabet, Meta, Amazon, and Oracle) plan combined infrastructure spend up 36% year over year in 2026. Per SemiAnalysis, that capex is being spread across custom accelerators, networking silicon, and host CPUs, not just NVDA-branded GPUs. Most of those dollars flow through chips that are not NVDA branded, and the MSFT and GOOGL repositioning piece walks through how the two largest customers are reallocating that spend.
Concentration risk also cuts both ways for a single name. One export rule change, supply hiccup, or customer reset can erase a year of gains overnight.
A basket smooths those single-name shocks. It also captures distinct layers of the AI stack at the same time, from raw wafers to design IP.
The supply chain is wider than the GPU itself. Networking silicon, custom accelerators, and CPU hosts all sit alongside NVDA.
The 4-Name Basket
AVGO: custom silicon for 6 hyperscalers
Broadcom posted Q1 FY2026 AI revenue of $8.4 billion, up 106% year over year. Total revenue hit $19.31 billion, up 29%, with semiconductor solutions at $12.5 billion, up 52%.
The company now has 6 named custom-silicon customers, including a new OpenAI engagement. Management projects above $100 billion in AI chip revenue by 2027.
VMware adds a software annuity on top. The unit booked $9.2 billion of TCV in Q1 alone, with revenue up 13% year over year.
TSM: the foundry chokepoint
Taiwan Semiconductor reported Q4 2025 revenue of $33.73 billion, up 20.4% year over year. About 77% of wafer revenue now comes from 7nm and below nodes.
TSM holds 72% of foundry market share per Counterpoint Research. It serves 534 customers, including every leading AI chip designer in the world.
2026 capex is committed at $52 billion to $56 billion. That spend is the gating factor for leading-edge AI capacity worldwide, and TSM owns the only seat at the table.
MRVL: hyperscaler design wins
Marvell is up 89% year to date in 2026, and 172% on a trailing 12-month basis. Analysts expect AI to account for 40-50% of revenue by year-end.
Custom silicon momentum is building at Microsoft and Amazon. Marvell sits in the design layer right next to AVGO, with its own slate of hyperscaler programs.
The stock is the most volatile of the four, which means position sizing matters more here than elsewhere in the basket.
ARM: data-center CPU shift
ARM benefits from the shift toward Arm-based server CPUs in cloud data centers. Its Neoverse cores power AWS Graviton and similar in-house silicon at Microsoft and Google.
Every new AI cluster needs a host CPU to feed the GPUs and accelerators. More of those CPUs are now Arm based rather than x86, and ARM collects royalties on each one.
Sizing the Basket
Equal weight across 4 names
A clean starting point is roughly 2-3% per name. That puts total basket exposure near 8-12% of a US-stock sleeve.
This sizing keeps each position large enough to matter, but small enough that one disappointing print does not blow up the portfolio. It also leaves room to add NVDA separately as the GPU anchor.
Tilt AVGO heavier on backlog visibility
If you want a tilt, AVGO has the clearest backlog signal. The $100 billion AI chip target by 2027 is anchored on signed customer programs, not just market sizing.
A 4% AVGO weight with 2% each across TSM, MRVL, and ARM is a defensible alternative. Either path beats a 12% NVDA overweight on diversification math.
Catalyst Calendar Through Year-End
AVGO reports its next quarter in early September. Watch the AI revenue run-rate against the $100 billion 2027 target, plus any new custom-silicon customer additions.
TSM publishes monthly sales data on the 10th of each month. Year-over-year growth above 20% supports the leading-edge AI capex case for the back half of 2026.
MRVL prints late August. The key number is the AI share of revenue, and any new hyperscaler custom-silicon disclosure.
ARM reports in early November. Royalty rate progression on Neoverse v3 cores is the metric to track for the data-center thesis.
Conclusion
NVDA is still the most direct AI compute name in the market. But a 4-name basket of AVGO, TSM, MRVL, and ARM captures more of the build-out and reduces single-stock risk in a noticeable way.
The Goldman 50/50 ASIC versus GPU forecast for 2027 is the structural reason this basket exists. Hyperscaler capex up 36% in 2026 is the near-term fuel for every name in it.
Gotrade Global lets you build this basket with fractional shares. Trade US stocks from $1, so even a 2% weight in each of the 4 names fits a small portfolio.
FAQ
Why exclude AMD from this basket? We exclude AMD here since it has its own framework in the AMD vs NVDA setup piece, where it competes more directly with NVDA on GPUs rather than diversifying the supply chain.
Is TSM exposed to Taiwan geopolitical risk? Yes, and that risk is part of the concentration tradeoff, but TSM's 72% foundry share makes it hard to avoid for any AI exposure.
How often should I rebalance the basket? Quarterly rebalancing after each earnings cycle is a sensible cadence for most retail investors.
Can I buy fractional shares of all 4 names on Gotrade? Yes, all 4 tickers are available as fractional shares starting from $1 per trade.
