Overview of the 2026 Hyperscale Capex Landscape
Meta, Microsoft, Amazon and Alphabet together forecast more than $615 billion in capital expenditures for 2026 – a roughly 70% increase over the already lofty 2025 spend. The bulk of this money is earmarked for cloud infrastructure and artificial‑intelligence compute.
Why Nvidia Is Central to the Spend
Analysts estimate that up to 60% of the AI‑related portion of this capex will flow to Nvidia GPUs. Despite Amazon’s push for Graviton CPUs and Trainium accelerators, the market’s cost‑per‑token improvements are still driven by Nvidia’s annual product cadence.
- 12‑month GPU refresh cycle vs. 18‑24 month custom silicon cycles.
- Projected $225 billion run‑rate for Nvidia versus $10 billion for AWS silicon.
- Vera Rubin GPUs promise 5× performance, 10× throughput, and 0.1× cost per token.
AWS Strategy: Custom Silicon to Reduce Dependency
Andy Jassy highlighted three pillars:
- Scale‑up of Graviton CPUs – up to 40% better price‑performance than x86.
- Rapid deployment of Trainium2 – 30‑40% price‑performance advantage over comparable GPUs.
- Linking AI adoption to cloud migration, arguing that enterprises will pre‑train models on their own data.
While AWS aims to capture margin through silicon, the sheer volume advantage of Nvidia means the majority of AI‑driven compute will still rely on Nvidia’s GPUs.
Google Cloud: TPUs Complement Nvidia GPUs
Sundar Pichai repeatedly cited Nvidia as a “key partner” while emphasizing Google’s own Tensor Processing Units (TPUs). Google’s AI queries are now three times longer than traditional searches, driving a 100× higher inference cost per session.
- AI‑mode queries up 2× per user in the U.S.
- AI queries are 3× longer, with 1 in 6 being multimodal (voice or image).
- Cloud revenue grew 48% to $17.7 billion, with operating margin rising to 30.1%.
The combination of TPUs for cost‑effective workloads and Nvidia GPUs for cutting‑edge performance creates a complementary stack.
Microsoft Azure: Growth Decoupled from Capex
Azure’s capex rose, yet sequential growth slowed, prompting a market sell‑off. Azure added roughly $2.5 billion more incremental revenue than AWS in 2025, contradicting Jassy’s “incremental dollars” claim.
Investor Takeaways
- Capex alone is no longer a bullish signal; investors now demand a clear link between spend, growth, and margin.
- Nvidia’s rapid cadence and volume give it a durable low‑cost advantage that custom silicon cannot fully offset.
- Hyperscalers will continue to invest in proprietary silicon (Graviton, Trainium, TPUs) to improve unit economics, but these efforts are complementary rather than competitive to Nvidia.
- Monitoring Nvidia allocation, AI query cost trends, and incremental revenue versus base‑rate growth will be critical for assessing the next wave of AI infrastructure investment.