The Most Important AI Stock You're Not Watching
Everyone is comparing chips from Nvidia, Alphabet, AMD, and custom accelerators from Broadcom. That debate matters. But there's a simpler signal for where AI capex is headed: follow the factory that makes most of those chips.
That factory is Taiwan Semiconductor Manufacturing (TSMC). It's the largest foundry by revenue and the quiet beneficiary of AI spending, no matter who wins the compute race.
Why TSMC sits at the center of AI spend
- Hyperscalers are guiding for more record data center capex through 2026.
- Chances are high those compute dies are produced at TSMC's nodes, then packaged and shipped into accelerators months later.
- TSMC's revenue mix is a direct read-through for AI demand: high-performance computing (HPC) was roughly 57% of revenue in a recent quarter; smartphones around 30%.
Because of the lag from wafer start to live data center deployment, chips made now may not power workloads until late 2026 or 2027. That makes TSMC's growth one of the cleanest leading indicators of AI infrastructure spend.
The tell: revenue momentum
Recent results showed 41% year-over-year growth in the third quarter. That pace suggests hyperscaler demand is still strong. If that growth stalls or flips negative, consider it an early warning that AI buildouts are pausing.
In short: TSMC's top line is a proxy for AI infrastructure cycles. Monitor it like you'd monitor bookings in software.
What to watch next (practical checklist)
- Monthly revenue trend and YoY comps from TSMC's investor page (TSMC IR).
- HPC mix: stable-to-rising share confirms ongoing AI demand.
- Packaging capacity: CoWoS/advanced packaging is a gating factor for AI GPUs.
- Node ramps: N5/N4 for volume, N3 and below for next-gen accelerators.
- Customer commentary: capex updates from Microsoft, Alphabet, Amazon, and Meta ripple straight into TSMC's backlog.
Valuation and positioning
TSMC trades around 28x forward earnings. It isn't cheap on an absolute basis, but it's still at a discount to many AI-adjacent peers. If you want exposure to AI compute without betting on a single winner, TSMC is the "picks and shovels" approach that often holds up across cycles.
Framed differently: you're buying the factory that benefits whether Nvidia keeps leading, AMD gains share, Alphabet leans into custom silicon, or Broadcom secures more accelerator wins.
Risk factors you should underwrite
- Geopolitical and supply chain concentration in Taiwan.
- Customer concentration and order volatility.
- Cycle risk if AI buildouts moderate after large 2025-2026 waves.
- Advanced packaging constraints and substrate bottlenecks.
Bottom line for finance teams
If you track AI capex, track TSMC. Rising revenue and a strong HPC mix point to continued buildouts; a slowdown flags caution ahead. You don't have to predict the next top-performing accelerator-TSMC gets paid either way.
Related resource
Exploring practical AI tools for FP&A, risk, and analytics? Here's a concise roundup: AI tools for finance.
Your membership also unlocks:
