Perplexity Never Named AWS When Asked About GPU Clouds
A 24-answer test of ChatGPT, Gemini and Perplexity on GPU inference buying: the hyperscalers won two assistants, and Perplexity named neither AWS nor GCP.
Asked eight buyer-style questions about GPU inference, Perplexity never named AWS or Google Cloud once. ChatGPT and Gemini sent the buyer straight to the hyperscalers.
That split comes from a small, openly documented test published on Dev.to. One developer wrote eight questions about GPU inference with no brand names in them — cheapest option for prototypes, best for real-time low-latency inference, best MLOps integration, and so on — and sent each question once to each assistant through its official API on October 2, 2026. The models were gpt-5-mini, gemini-3.5-flash-lite and pplx-fast, giving 24 answers in total. A second model pass listed the products each answer recommended, in order.
What the 24 answers said
| Provider | Answers naming it (of 24) | ChatGPT | Gemini | Perplexity |
|---|---|---|---|---|
| Google Cloud | 13 | 6 | 7 | 0 |
| AWS | 10 | 7 | 3 | 0 |
| Microsoft Azure | 10 | 5 | 4 | 1 |
| Lambda | 9 | 3 | 3 | 3 |
| RunPod | 8 | 2 | 2 | 4 |
| CoreWeave | 5 | 0 | 2 | 3 |
Twenty products were named across the run. The pattern the author draws out: each assistant has its own favourite. ChatGPT named AWS in seven of its eight answers, Gemini named Google Cloud in seven of eight, and Perplexity named RunPod most often, at four of eight.
Three findings worth stealing
Being named often is not being named first. Google Cloud appeared in the most answers, 13, but came first in only two of them. AWS came first in five of its ten, at an average position of 2.0. Groq was named just three times and was first every time. If you track AI visibility for a product, the author's point is that mention count and position tell different stories, so track both.
Lambda was the only consistent name. It was named exactly three times by each of the three assistants, without topping any single one's list. Azure and RunPod were also named by all three, but very unevenly.
The assistants blur categories. Some answers to "GPU cloud for inference" recommended OpenAI, Anthropic and Cohere, which sell model APIs rather than GPUs, and Groq, which runs its own LPU hardware. The author left them in because that is what a buyer actually reads — a reminder that these answers mix "where do I run my model" with "whose model do I call".
Where Perplexity got it
Perplexity returns its sources, and the most-cited sites across its eight answers were gmicloud.ai, io.net, digitalocean.com and inworld.ai at four citations each, with dev.to, northflank.com, siliconflow.com and spheron.network at three. Most of those are vendor blogs and comparison posts rather than neutral reviews.
The author's reading — offered as interpretation, not proof from 24 answers — is that Perplexity answers from fresh web search, which is full of specialist providers' own comparison content, while ChatGPT and Gemini answer from training data, where the hyperscalers dominate.
How much to trust it
The author is explicit about the limits, and they matter. These are API answers from fast, cheap models; the consumer apps with browsing, memory and larger models may answer differently. Twenty-four answers is a snapshot, not a census, and a gap of one or two answers should be read as a tie. It measures what assistants say, not which provider is best — no price, latency or reliability was benchmarked.
A re-run of the same questions on September 28 kept the top two in place, Google Cloud and AWS, while the order below them shuffled. So treat the hyperscaler-versus-specialist split as the durable finding, and everything below it as noise at this sample size. The practical move for buyers: if your team is "just asking AI" which GPU cloud to use, ask more than one, because the answer depends heavily on which one you ask.
More from DangMua