SemiAnalysis reported on October 16 that several neoclouds — smaller providers that rent out GPUs — were turned away from GTC Berlin, held October 20–22, for adding AMD GPUs or Google TPUs. No names and no public response from Nvidia, so this is a report, not a confirmed fact; still, it illustrates a real risk for anyone buying compute.
What Was Reported (and What Wasn't)
According to SemiAnalysis, as picked up by TechSpot and Crypto Briefing, several neoclouds learned just days before the event that they would not be able to attend. The firm attributes the pressure to Nvidia's interest in keeping smaller providers running Nvidia-only fleets. Hyperscalers such as AWS and Azure, with more purchasing power, reportedly stayed outside that pressure.
Why Neoclouds Use AMD and TPUs
A Crypto Briefing summary says that as of August 2026 many neoclouds were splitting orders between AMD and Nvidia at close to 1:1, with at least two deploying 10,000 GPUs of each brand; the reason cited is AMD's price advantage, while Nvidia keeps the performance lead. Those figures come from different dates and from a secondary source, so treat them as indicative.
Nvidia vs. the Alternatives: How to Read It
Nvidia keeps its edge in software (CUDA), interconnect and ecosystem availability; AMD and TPUs compete on price and available capacity. When the dominant vendor also controls the allocation of new chips, a small customer that diversifies risks — according to fears SemiAnalysis had already gathered in June from neocloud executives — losing delivery priority or support for a future funding round or IPO. This is a market-power dynamic, not a technical one, which is why the end buyer should know about it.
What It Means for a Company Renting AI Compute
If your GPU provider depends on Nvidia's goodwill, so do your price and your capacity. Ask about the hardware mix, about capacity contracts with delivery dates, and about what happens to your workload if the provider switches chips. And design your applications so they aren't tied to a single manufacturer: frameworks with ROCm and XLA support in addition to CUDA, and periodic testing on alternative hardware.
Sources: SemiAnalysis (via TechSpot and Crypto Briefing, October 16, 2026).