> is to be able to have finances to back a research division, where you can spend billions on building a new type of technology, if need be, that may not pan out. You don't need a monopoly to accomplish that
A company in an industry with very tight margins has much less money to invest in fundamental research. All the recent growth in generative AI has been driven by companies with very high margins; Google, Facebook, Amazon. If all those FANG were in tightly competitive markets and hence had low margins, they wouldn't have had billions of dollars to spend on the GPU compute necessary to develop modern language models. Which is evidenced by the fact that no companies in more competitive sectors have produced any large language models.
> company in an industry with very tight margins has much less money to invest in fundamental research
OpenAI has raised almost $18bn to date [1]. That puts it in the top 10 corporate R&D spenders globally, ahead of Intel and the entirety of big pharma [2]. (And OpenAI's gross margins for its API business are estimated around 40% [3]. Standard fare for tech. If anything, OpenAI subsidising its business with Apple and ChatGPT is behaving more like a tech giant than a start-up.)
The top of that list are the big 5 American tech companies, spending about $200bn annually on R&D. By coincidence, that's roughly the pace of U.S. VC spend [4]. The depth of American private capital markets make a solid case against favouring housing these long-shot bets inside tech giants. (Particularly absent non-compete and IP reform.)
A company in an industry with very tight margins has much less money to invest in fundamental research. All the recent growth in generative AI has been driven by companies with very high margins; Google, Facebook, Amazon. If all those FANG were in tightly competitive markets and hence had low margins, they wouldn't have had billions of dollars to spend on the GPU compute necessary to develop modern language models. Which is evidenced by the fact that no companies in more competitive sectors have produced any large language models.