Customer concentration in the hyperscaler segment is practically unavoidable, when a handful of customers represent such a large percentage of the market.
In 2025, four hyperscalers (Amazon, Google, Microsoft, Meta) represented as much as 70 percent of all capital investment in the technology business, for example.
Amazon, Google, Meta, and Microsoft control a massive portion of the world's total computing infrastructure, particularly in artificial intelligence. Together with Oracle, these hyperscalers own over 66 percent of global AI compute capacity.
To be sure, some suppliers (especially those which historically sell to enterprises, mid-market or small business) might not be as dependent on those few customers.
Still, for some products (graphics processing units, high-performance memory and accelerators, for example), there are just a few big volume buyers.
And as much as any supplier might wish to reduce reliance on just a few customers, that is hard to do in markets where a handful of buyers dominate:
Google: Dominates the single-largest share of pure AI compute, holding about 25 percent of global capacity. While rivals rely on Nvidia, Google’s capacity is largely driven by its proprietary Tensor Processing Units (TPUs). In the broader public cloud infrastructure market, Google Cloud holds approximately 14 percent market share
Microsoft: Holds an estimated 21 percent share of the global cloud infrastructure market. Through massive deployments of Nvidia GPUs and its strategic partnership with OpenAI, Microsoft accounts for an outsized share of enterprise AI workloads in the cloud
Amazon (AWS): Leads the global cloud computing market with 28 percent market share. While historically focused on broad, general-purpose enterprise computing, Amazon is rapidly scaling its proprietary AI chips (Trainium and Inferentia) to capture a larger portion of specialized AI computing workloads
Meta: Operating the largest internal compute footprint, Meta owns an estimated 10 percent of the world's total AI compute. Unlike the others, Meta's compute is largely dedicated to internal workloads, such as powering the massive recommendation algorithms for Facebook and Instagram, alongside its proprietary AI models.
So, like it or not, customer concentration in some parts of the AI value chain is unavoidable. There are only a few “whales” on the buyer side.
And some might argue there are similar few whales in the anchor tenant category as well. Consider the role OpenAI and Anthropic play as anchor customers for the hyperscaler compute “as a service” providers.
No comments:
Post a Comment