Research record
Not reviewedNo reviewer, review date, or supporting citations are recorded for this demo thesis. Scores are editorial opinions and have not been verified against cited evidence.
Revision history
No dated revisions are recorded for this thesis.
Investment thesis
The thesis in one line
NVIDIA is the toll road of artificial intelligence — almost every frontier model is trained and served on its hardware and software stack.
The real moat is software
It is tempting to see NVIDIA as a chip company. The durable advantage is CUDA — the software layer that two decades of developers, libraries and frameworks have been built on top of. Competitors can match a transistor count; they cannot easily replicate the ecosystem. On top of that, NVIDIA owns the networking (NVLink, InfiniBand, Spectrum-X) that stitches thousands of GPUs into a single training cluster. It sells the system, not just the silicon.
Why it keeps winning
- An annual cadence (Hopper → Blackwell → next) keeps rivals a generation behind.
- Full-stack integration means performance-per-watt advantages compound at the data-centre level.
- Dominant share of AI training accelerators gives pricing power funding the next R&D cycle.
The honest risks
Revenue is concentrated in a handful of hyperscalers, several of which are designing their own in-house ASICs to reduce dependence. The biggest swing factor is the durability of data-centre capex: if the hyperscalers enter a digestion phase, a premium-multiple stock corrects fast. The bull case isn't that there are no risks — it's that demand visibility still outruns them.
Stage of the cycle
The undisputed leader. The market debate is no longer whether NVIDIA dominates, but how long the capex super-cycle runs. Historically the best entries came on broad-market or capex-scare pullbacks, not on breakouts.
Scorecard
Upside 8/10 · Safety 7.5/10 · AI exposure 10/10. The benchmark every other AI investment is measured against.
