In today’s column, I explore a recent comment made by Sam Altman about the nearing of pinnacle AI and how such AI will be extraordinarily vital throughout all of society.
Things go like this. Altman postulated that AI will be so essential that just about everyone will find themselves requiring the use of AI in one way or another. In that case, an issue that he foretells is that people will need to have a “compute budget” whereby they can afford to access this advanced AI. The problem is that the need for such budgets or monies might relegate society to splinter into the AI haves versus the AI have-nots, a disconcerting situation that presumably and collectively we will want to avoid or mitigate if possible.
Let’s talk about it.
This analysis of an innovative AI breakthrough is part of my ongoing Forbes column coverage on the latest in AI including identifying and explaining various impactful AI complexities (see the link here).
The Upcoming Rise Of AI Compute Budgets
In Sam Altman’s latest blog post dated February 10, 2025, entitled “Three Observations” and covering various ramifications associated with the attainment of artificial general intelligence (AGI), this remark was indicated:
- “We are open to strange-sounding ideas like giving some ‘compute budget’ to enable everyone on Earth to use a lot of AI, but we can also see a lot of ways where just relentlessly driving the cost of intelligence as low as possible has the desired effect.”
Allow me to briefly explain some associated context.
Suppose that your everyday existence eventually relied upon using advanced AI. You get up in the morning and consult with AI about what you are going to do that day. Upon going to work, you use AI throughout your workday, rather persistently and with few exceptions. Kids in school are using AI as their primary form of teaching and garnering an education. AI is ubiquitous.
Whether you like that scenario or find it disturbing, please go along with the assumption that it is what our near future holds. Humor me in that regard.
Show Me The Money
The somewhat seemingly mundane but altogether significant question is who will pay for all this use of AI?
AI doesn’t just magically run in a dream world. The use of AI requires computer servers and other computing devices that will execute the code and house the data involved in the AI. Those cost money. Someone must set up those systems and keep them going. AI will be nothing if the underlying facilities needed to run the AI aren’t available.
There is also the cost associated with the AI system itself. Even if somehow, that was made available for free, the hardware and associated devices are still likely to have a cost. It seems doubtful that all the hardware and all the software will somehow miraculously be available at zero cost.
Thus, believe it or not, the lofty topic of advancing AI and the future of humanity coupled with the use of AI, soberly comes down to good old regular dollars and cents. Where will the money come from and who will be able to afford to exist in an AI-infused society?
AI As A Fundamental Resource
You might vaguely recognize this societal problem but in a different milieu.
Namely, electricity.
You could almost make the same argument about electricity. We depend heavily on electricity. Electricity must be paid for. Some people can afford electricity, some cannot. Society has grappled with that issue and generally found ways to cope, especially in developed nations.
Right now, you would be hard-pressed to insist that AI is on par with electricity.
Some people use AI, but many do not. Those that use AI are pretty much able to walk away from it. The dependencies associated with AI are rather tenuous at this time. It is hard to envision a world in which AI becomes so pervasive, so needed, that we will be lock, stock, and barrel, dependent on AI.
AI reaching that position isn’t quite as far-fetched as might seem at first glance.
Imagine that AI is aiding you in making decisions, big and small. You trust and rely upon the AI. AI helps you to solve problems. You seek out AI when problems arise. AI enters all aspects of your life. Work-related aspects, sure, AI is there. Personal aspects, of course, AI is there. Think of AI as being so intelligent that you would be foolhardy not to consult with it. Advanced AI such as the presumed AGI would be your partner throughout your worldly existence, always at the ready, 24 x 7, and able to assist.
What happens when you can’t access the AI due to lack of affordability?
Everybody else that can afford AI will be a step ahead of you. They have AI on their side. Those people armed with AI will outsmart you at every turn. Your ability to keep up with them or try to compete with them will be at a low ebb. In a sense, it is akin to the famous line that two heads are better than one.
The AI have-nots are left out in the cold.
Keep The Price Low
As noted in Altman’s remark, one solution involves striving to make the cost of using AI so low that just about everyone can afford to use it. Overwhelmingly so. Almost universally so. The idea is that those who cannot do so will be a tiny teensy segment of society and presumably, we can find ways to aid them without too much difficulty.
The affordability problem then becomes in theory generally resolved by ensuring that AI is a nearly zero-cost usage proposition.
Can we get there?
Nobody right now can say for sure whether we can or will.
Consider some glimpses that might be telltale signs.
You might have heard recently about a newly released generative AI named DeepSeek R1 that was devised for supposedly pennies on the dollar in contrast to the millions and billions that have been sunk into the major generative AI apps. There is a heated debate about how this came to be, along with questions of whether the claimed costs are being played a bit with (see my coverage on the topic overall, at the link here).
Another factor is that generative AI is being compressed, streamlined, and shaped so that it can run on smartphones (see my discussion of 1-bit LLMs at the link here). If your AI usage is almost entirely via your smartphone, the cost to use the AI drops precipitously. Assuming the AI app itself is highly affordable, and assuming that you don’t need to rely on cloud or other computing resources, this might get us closer to the affordability rifeness (well, you’d still need to be able to afford the smartphone).
When Ready Affordability Isn’t In The Cards
Since we don’t know for sure whether advanced AI of this nature will be radically affordable, let’s consider what happens when such AI is relatively non-free.
Here’s a scenario to illustrate the predicament. An everyday person named Alex is at home and needs to use AI for something of importance to them. They log onto their AI account.
The dialogue they have goes like this.
- Alex’s entered prompt: “I need to consult with you further about the important issue that I had last discussed with you, but I’m almost out of AI compute credits. Can you grant me an extension?”
- Generative AI response: “I’m sorry, Alex, but your monthly AI compute budget is nearly exhausted. You have 2.3 compute units remaining, which is insufficient for your request.”
- Alex’s entered prompt: “Can’t you make an exception? It’s just a small task.”
- Generative AI response: “Sorry, you’ll need to add more money to your account. Once you’ve done so, I’ll be glad to go over the issue with you. I’m sure that doing so will be immensely helpful to you.”
In this scenario, Alex has nearly depleted the monies in their AI compute budget. The AI has been programmed to require upfront payment or some form of ironclad payment capability that assures the AI costs will be covered. Unfortunately, Alex doesn’t have the dough.
It’s no dice for Alex.
National Approaches To AI Affordability
All manners of socioeconomic options are being postulated for the AI affordability crunch (assuming it happens).
Perhaps some nations will establish a guaranteed base level of AI access or a minimal AI compute budget for all their citizens. AI might become a heavily regulated commodity, akin to electricity and water. There could be regulations and laws established concerning rights associated with having access to AI. For my analysis of the legal dimensions, see the link here. Some speculate that there might be National AI Compute Reserves, modeled on the approach to the electrical grid or oil reserves.
The United Nations might seek to establish some form of AI as a global resource, similar to the need for clean water, food, and the like. AI would potentially be considered an essential human right. I’ve previously examined how the UN is striving toward AI for the SDGs (sustainability development goals), see the link here, ergo this added step of AI for good and AI for all is well within the same sphere of attention.
Let’s venture into the scenario of Alex and suppose he is in a country that has put in place a National AI Access Act.
We shall continue Alex’s dialogue with AI.
- Alex’s entered prompt: “I thought that there was some national law that can get me more compute units so that I can continue my AI access. Isn’t that the case?”
- Generative AI response: “Yes, compute allocations are regulated under the National AI Access Act. Exceptions require approval from a verified authority. Would you like me to submit a priority request?”
- Alex’s entered prompt: “How long would that take?”
- Generative AI response: “Standard review time is 3-5 business days and must be handled by human reviewers. Expedited requests are available for government and corporate users with premium access.”
- Alex’s entered prompt: “What if I borrow credits from next month’s allotment?”
- Generative AI response: “Your account is indeed eligible for future compute borrowing. However, you’ve already used most of that up. I think there might be a way that we can do this. You can borrow from next month for the little you have left, and I can go into a low-fidelity AI mode, which will consume fewer compute units. The downside is that my answers aren’t as high-quality as usual. Do you want to go this route?”
- Alex’s entered prompt: “I guess so. It seems like my only option to get underway right away. Please proceed.”
You’ll observe that I put a few twists and turns in that dialogue. Just wanted to ensure that we all realize that there will still be bumps in the road even if various schemes are devised to enable AI compute budget affordability.
Nefarious Underground AI Compute Budgets
I don’t want to seem gloomy, but if AI is an affordability issue, you can bet your bottom dollar that some form of underground or black market is bound to be formed around getting the compute budget needed for AI.
There might be AI hackers that can break into AI systems and give you access in exchange for under-the-table payments or other barter. Maybe corrupt officials that oversee the AI compute resources will take a bribe or otherwise sell AI access in some criminal way. There might be devised AI compute farms that are under the radar of the government. They pride themselves on selling super inexpensive AI access. The government though might shut them down at any moment.
Cross-border trading of AI compute budgets could arise. Imagine that one country has a lot of AI availability, while another one does not. Some broker arranges for access to the other country’s AI, even though you are in a different country. Of course, this might be breaking laws in both countries at the same time.
AI compute laundering networks could grow like weeds.
Additional Weighty Considerations
A few more points about what could happen if the AI affordability dilemma comes to fruition, and then I’ll conclude this discussion.
I’ve previously covered that the availability of advanced AI could become a geo-political superpower wedge issue, see the link here. Country X has AGI, country Y does not. Country X is much more productive and can run circles around country Y. To the victor go the spoils. You can see where that leads.
On a related facet, AI compute nationalism or sovereignty is a relatively new moniker that has been formulated and refers to the possibility that a country might strictly keep AI within its national borders. Access to foreign AI resources is banned. The national AI grids are exclusively for those of the country at hand. For more on this, see my commentary at the link here.
There is also talk of AI colonialism, see my analysis at the link here.
This generally refers to the idea that some countries might dominate the AI availability gambit. They have massive resources devoted to running AI. In turn, they dole out to other countries the AI access. Sometimes this is done on an altruistic basis, sometimes to control or dominate other countries.
Finally, as noted, money is the primary currency that will be used to determine how someone will pay for AI access, and you pay money for some amount of AI compute. You might be allocated a monthly allotment of AI compute. And so on.
I thought that you might also find it of keen interest that some believe the AI compute will be purchased not based on time or processing cycles but instead based on tokens. Tokens are the prevailing way that generative AI represents words inside the computer, doing so in a numerical fashion, see my detailed explanation at the link here. I bring up this nitty-gritty facet simply to let you know that those proponents of a token basis are of the mind that tokens will become a universal digital currency.
Here’s how this goes. Rather than thinking about money per se, you will think about how many tokens you need for whatever you want AI to do for you. Tokens will be paid for and will float on the open market as to the cost per token, or there might be heavily regulated markets that dictate the prevailing cost per token.
The bottom line is that you had better start saving up your tokens. You heard it here first.
AI Compute Budgets On The Horizon
Congratulations, you’ve now done a deep dive into AI compute budgets.
Will this come to pass?
Time will tell.
Some exhort that all this chatter about AI compute budgets is pure hogwash. Won’t ever happen. It is wild speculation based on an outsized conjecture with a splash of fanciful imagination.
I guess we’ll let Abraham Lincoln have the last word for now: “The most reliable way to predict the future is to create it.” So, get out there and help create the future of AI, doing so might provide huge benefits to your fellow humans and yourself.
