What people zeroed in on was how limited and selective these controls really are. Several comments pointed out that being on the Entity List does not mean a company disappears. It mostly cuts off US sellers and service providers, while purchases from the listed company can still continue. Others added that
GPU controls already leak through resellers and neighboring countries, so the policy often raises costs rather than shutting access down. That made the DeepSeek question look less like a clean security decision and more like industrial policy by other means.
The strongest consensus was that cheap Chinese models are puncturing the economics of US frontier labs. Many users said DeepSeek is good enough for coding help and routine development work at a tiny fraction of Claude or GPT pricing. That made the blacklist debate feel to them like an attempt to create an artificial moat around expensive US APIs. At the same time, plenty of commenters drew a distinction between Chinese model providers and Chinese models. The safer route, in their view, is to self-host or use third-party infrastructure outside China. That framing matters because it turns the issue from "use DeepSeek or not" into "where do you run the weights and who sees your prompts."
Another theme was hypocrisy around
IP and model
distillation. Reuters cited claims that DeepSeek and other Chinese labs extracted capabilities from Claude. The reaction was blunt. If US labs trained on huge swaths of copyrighted material and defend that as normal learning, they have weak moral standing to call model-output scraping illicit. The more practical point underneath the dunking was that model outputs are becoming training data and pricing pressure is becoming strategic. Cheap
open-weight releases do not just win customers. They commoditize
inference and shift value away from the labs trying to own the model layer.
The mood was hostile to the blacklist idea overall, though not especially pro-CCP. A lot of people simply do not trust either government and see monopoly prevention, lower prices, and local deployment as the useful frame. The clearest throughline was that AI is moving into the same bucket as chips, cars, and telecom gear. Once that happens, "national security" becomes the language used to decide who gets market access, who captures margin, and which parts of the stack stay open.