AI Chips
Can Nvidia Maintain AI Chip Dominance?
Don't just read what happened. See what could happen next.
Prediction
Nvidia remains the default training platform for the next product cycles, with share erosion at the edges from AMD and custom ASICs.
- Confidence
- 69%
- Horizon
- 12–36 months
- Impact
- High
- Direction
- Stable
Prediction changed -3 points in 21 days
Short answer
CUDA software lock-in, system-level networking, and relentless roadmap cadence still favor Nvidia. Challenges from AMD and hyperscaler custom chips are real — especially for inference — but displacing Nvidia as the training standard is a multi-year fight, not a single quarter event.
Why this question matters
Chip concentration shapes AI cost curves, geopolitical risk, and which companies can train frontier models.
What's happening now?
CUDA and ecosystem stickiness
Frameworks, kernels, and talent still optimize first for Nvidia stacks.
Signal · strong
Custom silicon at hyperscalers
In-house accelerators gain share for specific inference and training mixes.
Signal · moderate
AMD competitive pushes
Improved software and pricing pressure create credible second-source options.
Signal · moderate
Export controls reshape demand geography
Policy redirects who can buy what — and where alternatives get funded.
Signal · strong
What could happen next?
Scenario A
Entrenched leadership
Nvidia keeps commanding share of training GPU spend through the next architectures.
Scenario B
Inference diversification
Training stays Nvidia-heavy; inference fragments across AMD, ASICs, and NPUs.
Scenario C
Faster share loss
Software portability improves enough for large buyers to multi-home aggressively.
Key companies / entities
- Nvidia
- AMD
- TSMC
- Amazon
- Microsoft
Evidence
- company
GPU shipment and datacenter revenue reports
- news
Hyperscaler custom chip announcements
- policy
Export control updates affecting AI accelerators
Prediction history
- Sep 22, 202669%
- Sep 15, 202670%
- Sep 8, 202671%
- Sep 1, 202672%