News for hackers. Stuff that matters.
Re:DeepSeek V4 Flash on a Single AMD MI300X by segmondy on Tuesday August 04, 2026 @09:13AM (#49168491)
no they won't , the bubble is a financial thing. the demand is real and not going away.
Re:DeepSeek V4 Flash on a Single AMD MI300X by atwrk on Tuesday August 04, 2026 @10:48AM (#49169774)
The big question is whether the demand will stay if the subsidized pricing ends. That's what the bubble talk is about. Right now all the players compete for market share and don't care about the losses (hence the debt). But what happens if no one wants to lend them anymore?
Re:DeepSeek V4 Flash on a Single AMD MI300X by jack_pp on Tuesday August 04, 2026 @11:55AM (#49170689)
I don't think inference is subsidized, it's the training. So what happens is, there's no new models anymore or are released slower.
Re:DeepSeek V4 Flash on a Single AMD MI300X by FeepingCreature on Tuesday August 04, 2026 @02:34PM (#49172921)
API inference is probably not subsidized. Coding plans absolutely are.
Re:DeepSeek V4 Flash on a Single AMD MI300X by vehemenz on Tuesday August 04, 2026 @12:28PM (#49171160)
Good distinction. The demand is partially driven by the low costs, which are only low because the major providers are losing money.
Re:DeepSeek V4 Flash on a Single AMD MI300X by qwytw on Tuesday August 04, 2026 @04:00PM (#49174226)
There is no evidence they are losing money on inference, though?

Also if they are keeping the price low because they want to gain market share and reduce the competitiveness of Chinese models they won't be able to raise prices without providers serving open models (at cost + low margin) severely undercutting them.

Re:DeepSeek V4 Flash on a Single AMD MI300X by vehemenz on Wednesday August 05, 2026 @11:52AM (#49184596)
There's no evidence they're making money, and we already know from the projected datacenter capacity in a few years that there will be for more supply than demand, so the major providers will have to repay that debt. Even if they are making money on inference, it's nowhere near enough to cover the bill. It's a losing proposition either way, especially with Chinese models now in play.
Re:DeepSeek V4 Flash on a Single AMD MI300X by qwytw on Wednesday August 05, 2026 @03:30PM (#49187765)
There is. Specifically the pricing for open models from third party providers on OpenRouter since inference is a "commodity" at this point. Unless we think that Opus/GPT-5.6 are many times less efficient than GLM 5.2 or Kimi Openai and Anthropic are making money from inference. > it's nowhere near enough to cover the bill

Obviously it does not cover R&D, marketing and other spending but nobody has ever claimed that here.

Re:DeepSeek V4 Flash on a Single AMD MI300X by vehemenz on Wednesday August 05, 2026 @04:43PM (#49188729)
I see your point, but Anthropic and OpenAI are not only less efficient, they have much higher operating expenses because of their massive AWS/GCP spending (due to not owning it) and have more capital expenditures than everybody else. It's not comparable to third-party providers making some profit on the margin.
Re:DeepSeek V4 Flash on a Single AMD MI300X by onoesworkacct on Wednesday August 05, 2026 @01:36AM (#49178949)
Demand is money.

No bubble, where money?

Re:DeepSeek V4 Flash on a Single AMD MI300X by SwedishDungeon on Saturday August 08, 2026 @08:48PM (#49227305)
A financial bubble drives a significant amount of demand. Take away the financial incentive and the demand will cool. I'm not expecting a pop though.