If OpenAI needs gpu to succeed, and can't raise any more capital to pay for it without dilution/going past MSFT's 49% share of the for-profit entity, then the corporate structure is hampering the company's success.
Sam & team needed more GPU and failed to get it at OpenAI. I don't think it's any more complex than that.
Somewhere closer to a GPU source. E.g. a new company that can trade unlimited equity for GPU time from a hyperscale cloud vendor, or work for the vendor itself.
Or, just maybe, this architecture just isn't going to get to where they wanted to go (a viable product, much less real AI), and the excuse was "we just need more GPU". In reality, this company came out with, as others before me have called it, a better autosuggest, aka stochastic parrots. That's interesting, and maybe even sometimes useful, but it will never pay for the amount of firepower required to make it run.
- Sam Altman _briefly_ went on record saying that openAI was extremely GPU constrained. Article was quickly redacted.
- Most recent round literally was scraping the bottom of the barrel of the cap table: https://www.theinformation.com/articles/thrive-capital-to-le...
- Plus signups paused.
If OpenAI needs gpu to succeed, and can't raise any more capital to pay for it without dilution/going past MSFT's 49% share of the for-profit entity, then the corporate structure is hampering the company's success.
Sam & team needed more GPU and failed to get it at OpenAI. I don't think it's any more complex than that.