AI will absolutely be cheaper than it is now in the future, because you’ll actually be able to run it locally - look at the recently announced Nvidia powered Surface Ultra as an example of what’s coming.
Companies are trying to get their whole AI pipeline more efficient and “smaller” to reduce costs, because their costs to build and run data centres are astronomical. Where it will likely go is a subscription service to access their models that you run locally. Passes most of the cost on to consumers, and is recurring revenue which they all dream of.
I don’t think there’s a chance that happens either at the consumer level or mass adoption at the enterprise level. The estimated price of that nvidia surface ultra is expected to be around $3000. Only a handful of power users will pay that, and others may just buy it for the computer power. Right now for the basic user it’s about $30 a month for basic AI plans. How many people that are willing to pay that do you think are willing to pay 100 years up front to then pay a subscription to then manage a local model?
Also, there are a lot of viable free models available right now that people can download for free. I have a recycled i5, 8gb ram machine that can run some basic models that are cable of doing 90% of the stuff the average person wants. My desktop with far better specs can run far more powerful models. The thing is, most people don’t know these exist, and if they do they don’t know how to set them up or are too intimidated to try.
That’s just for consumers. Trying to host that at an enterprise level is much more difficult. Assuming the business has the tens of thousands of dollars of hardware to support 100s or thousands of users, configuring governance on LLMs and everything associated with enterprise management that is currently handled at the current companies will fall back on the business. That will probably require hiring specialist, or at least expanding the work force. If it takes 1 $3000, unreleased and untested laptop, imagine the cost to support a company, especially one doing software development.
Even with all of that, training the models isn’t going to get cheaper. It’s going to get more expensive as more data centers are built. Operating costs will go up, and unless these companies start charging upwards of 3k a month per user, they won’t be profitable on AI alone. Obviously they are making money on other things like selling user data, but none the less, costs have to go up for even the notion of making money for these companies.
AI will absolutely be cheaper than it is now in the future, because you’ll actually be able to run it locally - look at the recently announced Nvidia powered Surface Ultra as an example of what’s coming.
Companies are trying to get their whole AI pipeline more efficient and “smaller” to reduce costs, because their costs to build and run data centres are astronomical. Where it will likely go is a subscription service to access their models that you run locally. Passes most of the cost on to consumers, and is recurring revenue which they all dream of.
I don’t think there’s a chance that happens either at the consumer level or mass adoption at the enterprise level. The estimated price of that nvidia surface ultra is expected to be around $3000. Only a handful of power users will pay that, and others may just buy it for the computer power. Right now for the basic user it’s about $30 a month for basic AI plans. How many people that are willing to pay that do you think are willing to pay 100 years up front to then pay a subscription to then manage a local model?
Also, there are a lot of viable free models available right now that people can download for free. I have a recycled i5, 8gb ram machine that can run some basic models that are cable of doing 90% of the stuff the average person wants. My desktop with far better specs can run far more powerful models. The thing is, most people don’t know these exist, and if they do they don’t know how to set them up or are too intimidated to try.
That’s just for consumers. Trying to host that at an enterprise level is much more difficult. Assuming the business has the tens of thousands of dollars of hardware to support 100s or thousands of users, configuring governance on LLMs and everything associated with enterprise management that is currently handled at the current companies will fall back on the business. That will probably require hiring specialist, or at least expanding the work force. If it takes 1 $3000, unreleased and untested laptop, imagine the cost to support a company, especially one doing software development.
Even with all of that, training the models isn’t going to get cheaper. It’s going to get more expensive as more data centers are built. Operating costs will go up, and unless these companies start charging upwards of 3k a month per user, they won’t be profitable on AI alone. Obviously they are making money on other things like selling user data, but none the less, costs have to go up for even the notion of making money for these companies.