1 hour ago
18
Jensen Huang has a number he likes, and he is sticking to it. At the Goldman Sachs Communacopia + Technology Conference on September 10, 2026, Nvidia’s CEO reaffirmed his forecast that global AI infrastructure spending will reach somewhere between $3 trillion and $4 trillion annually by the end of the decade. Notably, he made the exact same call at the exact same conference a year earlier.
The numbers behind the confidence
Nvidia reported fiscal Q2 2027 revenues of $96.2 billion, a figure that represents 106% growth year-over-year. Of that total, $89 billion came from data center operations alone.
Huang attributed the acceleration partly to the death of Moore’s Law. Traditional chip improvements, the reliable doubling of transistor density every couple of years, have effectively stalled. That forces companies chasing AI performance to buy more specialized hardware rather than wait for the next silicon generation to bail them out.
Supply, however, is not keeping up with demand. Huang flagged memory chip availability as the primary bottleneck currently limiting Nvidia’s ability to ship more product.
On the product pricing side, the numbers tell their own story. Nvidia’s Hopper-generation GPUs run around $18,000 per unit. The newer Blackwell architecture comes in at roughly $25,000 per chip. The forthcoming Vera Rubin platform is priced at approximately $40,000.
Hyperscalers are spending like it’s their job (because it is)
Hyperscalers are expected to increase their combined capital expenditures from roughly $800 billion in 2026 to approximately $1.3 trillion in 2027, a year-over-year jump of around 60%.
Nvidia has also partnered with major financial firms to help funnel more than $500 billion into AI data center projects.
For fiscal 2028, Nvidia is projecting revenue growth of around 70%. The analyst community consensus sits closer to 45%.
What this means for the broader AI trade
Within the semiconductor sector, the supply chain constraints Huang identified create a secondary investment thesis. Companies producing high-bandwidth memory and advanced packaging stand to benefit as bottlenecks attract capital. The shortage also provides a window for competing chip architectures to make inroads with hyperscalers looking to diversify their hardware suppliers and reduce dependence on a single vendor.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·