NVIDIA Corporation told investors on July 29, 2026 that hyperscale cloud customers have committed multi-year purchase agreements for Blackwell Ultra accelerators covering deliveries into late 2027. The update landed days before month-end as Microsoft and Amazon earnings reinforced that AI capital expenditure is still rising, not peaking.
Why Supply Visibility Matters
NVIDIA already controls an estimated 80% of the market for AI training accelerators. What changed in July is visibility. Rather than quarter-to-quarter purchase orders, Microsoft Azure, Amazon AWS, Google Cloud, and Meta are locking allocation windows further out, reducing the risk of idle data-centre shells waiting on chips.
Chief executive Jensen Huang said remaining performance obligation for data-centre products continues to expand as customers secure capacity for inference clusters, not only training runs. That shift matters for investors who worried that 2025–2026 training buildouts would leave a cliff in demand.
Competitive Context
AMD's MI350 family and custom silicon from Google, Amazon, and Microsoft are winning niche workloads, but Bernstein still estimates NVIDIA handles the majority of cloud AI FLOPS. Export controls on advanced chips to China remain a residual risk, though domestic U.S. and European demand has more than offset restricted China volumes so far in 2026.
Stock and Market Impact
NVDA remains one of the three largest companies in the S&P 500 by market capitalisation. For live pricing, see Yahoo Finance. Semiconductor peers including SK Hynix and Samsung HBM suppliers also traded firmer on the backlog commentary.
Investors will next watch NVIDIA's formal quarterly report for confirmation that data-centre gross margins hold as Blackwell Ultra ramps and as networking, CUDA software, and DGX cloud services contribute a larger share of revenue.