Revenue from new product development is projected to become the biggest contributor to the industry, estimated to generate about $4.2 trillion to fund the booming technology’s global market within the next half-decade
This is false. Bain says that $4.2T of extra magic platforms/revenue needs to be fantasized, beyond what can theoretically be hoped for. Not that it will be. The primary fantasy he brings up is if the relative handful of leading scientific/health researchers spend $3T to $4.2T on tokens per year to make breakthroughs. Since this is absurd/impossible, the answer would be extending circular financing techniques where researchers can sell shares of their future profits for token credits today. Extreme discounts on those tokens would need to be given.
The $6T figure is based on simple 25% formula of what past web/cpu datacenters have spent on keeping up with competition/demand/replacements. GPUs are much more expensive, and included power infrastructure that is only profitable if the GPUs stay there. The electricity bill is much higher with GPUs and even if everything is as automated as web server farms, it is lower margins on those revenues, and the replacements are more expensive. So $8T to $10T in annual revenue is a more appropriate target, and an even more unattainable one.
That particular instance was more “using a public hint given by an expert” as a search path instead of theft. But, even when they say they don’t (and sometimes they explicitly keep prompts/responses for future training), they have the power to steal. An alternative to investing in R&D companies with tokens, mentioned above, is AI labs creating their own R&D divisions and hiring researchers directly. But that requires them coming up with even more money they don’t have.
Even at $6T (without my higher revision), it’s actual proof of AI bubble popping unless military/surveillance state buys $4T in tokens.
This is false. Bain says that $4.2T of extra magic platforms/revenue needs to be fantasized, beyond what can theoretically be hoped for. Not that it will be. The primary fantasy he brings up is if the relative handful of leading scientific/health researchers spend $3T to $4.2T on tokens per year to make breakthroughs. Since this is absurd/impossible, the answer would be extending circular financing techniques where researchers can sell shares of their future profits for token credits today. Extreme discounts on those tokens would need to be given.
The $6T figure is based on simple 25% formula of what past web/cpu datacenters have spent on keeping up with competition/demand/replacements. GPUs are much more expensive, and included power infrastructure that is only profitable if the GPUs stay there. The electricity bill is much higher with GPUs and even if everything is as automated as web server farms, it is lower margins on those revenues, and the replacements are more expensive. So $8T to $10T in annual revenue is a more appropriate target, and an even more unattainable one.
They will steal the scientific data anyway, just like Navier Stokes recenty.
That particular instance was more “using a public hint given by an expert” as a search path instead of theft. But, even when they say they don’t (and sometimes they explicitly keep prompts/responses for future training), they have the power to steal. An alternative to investing in R&D companies with tokens, mentioned above, is AI labs creating their own R&D divisions and hiring researchers directly. But that requires them coming up with even more money they don’t have.
Even at $6T (without my higher revision), it’s actual proof of AI bubble popping unless military/surveillance state buys $4T in tokens.