The performance of leading companies in the artificial intelligence (AI) supply chain between 2021 and 2025 provides a clear, data-backed signal of how capital is being allocated and where the next phase of the market is heading.
Data shows that returns were highly concentrated with a small group of companies delivering outsized gains:
- Palantir Technologies: +876% (from $18.21 to $177.75)
- Vertiv Holdings: +549% (from $24.97 to $162.01)
- NVIDIA: +534% (from $29.41 to $186.50)
- Broadcom: +420% (from $66.54 to $346.10)
- SK Hynix: +397% (from $86.85 to $431.60)
- Arista Networks: +265%
- Micron Technology: +247%
- CrowdStrike: +129%
- Dell Technologies: +124%
- Alphabet: +116%
At the same time, some companies underperformed despite AI exposure:
- Intel: –28%
- Equinix: –9%
- Digital Realty: –13%
1. Returns Are Concentrated, Not Broad-Based
The top three performers were Palantir, Vertiv, and Nvidia, and delivered between 500% to 876% gains, while large-cap names like Alphabet returned just 116% over the same period.
This shows:
- AI gains are not evenly distributed
- Capital is flowing to specific high-leverage positions, not the entire sector
2. Infrastructure Is Outperforming Core AI Chips
A key data point:
- Vertiv (+549%) outperformed Nvidia (+534%)
Vertiv does not build AI models or chips. It provides:
- Cooling systems
- Power infrastructure for data centres
Interpretation:
- The bottleneck has shifted from compute to infrastructure
- Companies solving physical constraints are gaining pricing power
3. Memory Has Become a Critical Constraint
- SK Hynix: +397%
- Micron: +247%
These gains are tied to demand for:
- High-bandwidth memory (HBM)
- AI model training capacity
Insight:
- Memory is now a core limiting factor in AI scaling
- Not all semiconductor exposure is equal
4. The Market Is Penalising Indirect Exposure
Despite being in the same ecosystem:
- Intel: –28%
- Equinix: –9%
- Digital Realty: –13%
This shows:
- The market is filtering aggressively
- Only companies with direct AI revenue linkage are rewarded
New AI Models Are Scaling at Extreme Speed
CoreWeave provides a strong example:
- Revenue: $15 million (2022) → over $5 billion (2025)
- Share price: +93% post-IPO (March–December 2025)
This reflects:
- Emergence of GPU-as-a-service
- Shift toward outsourced compute infrastructure
What This Means for the Next Phase (2026 and Beyond)
The data points to a clear transition:
Phase 1 (2021–2024):
- Chipmakers dominate (Nvidia, Broadcom)
Phase 2 (2024–2026):
- Infrastructure and bottlenecks take over
(Vertiv, memory suppliers, GPU cloud)
Investors King Note
- AI is no longer a broad theme — it is a precision trade
- The highest returns are coming from:
- Bottlenecks
- Capacity constraints
- Critical infrastructure
The next phase of the market will be defined by:
- Who controls power, cooling, memory, and compute access
- Not just who builds AI models
