# Navigating AI Access Inequality and Compute Scarcity

**Podcast:** The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis
**Published:** 2026-05-17

## Transcript

Today on the AI Daily Brief, AI inequality.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
All right, friends, quick notes before we dive in.
First of all, thank you to today's sponsors, KPMG, Robots and Pencils, Superintelligent and Blitzy.
To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts.
To learn more about sponsoring the show, send us a note at sponsors at aidailybrief.ai.
And when you are at AIDealyBrief.ai, you can check out everything else going on in the ecosystem.
We've got our newsletter.
We've got companion experiences that come out pretty frequently alongside different episodes.
And of course, we've got our live listing for a growth engineer.
You can find that at jobs.aiDealyBrief.ai.
Pretty much anything I do is going to be there in some way, shape, or form.
Now, today we are talking about AI inequality, but in a slightly different way.
In this case, I'm talking about the possibility of haves and have-nots.
as dictated by their access to AI models.
So this is not a some people will have jobs, some people won't have jobs argument, but an exploration of how we could quickly be going into a world where the quality of the AI that people have access to is far more unequal than it is today.
The background context for this and what got me thinking about it is two parts.
The first is the business model constraints that have been a theme running throughout the show for the last couple of weeks.
One of our major points of discussion has been what happens to the business models of AI companies when the demand for tokens greatly starts to exceed the supply of tokens.
Now, why, of course, we are facing this issue now is the broader shift from assisted to agentic AI, where a single individual can be consuming billions of tokens, thanks to all the agents they have running around building things and doing things on their behalf.
This really came home to roost this week, specifically with Claude Code's big change to how it builds.
effectively ending the subsidy for people who aren't building within Anthropics-owned harnesses.
The second bit of context for this exploration has been the way that the introduction of mythos has impacted the way that the U.S.
government thinks about models.
These two things are combining in ways that I think are quite potent when it comes to the landscape of AI access, with some not tremendously great outcomes potentially on the horizon.
Now, this is, of course, a long-read-slash-big-think episode, and so to set up the conversation, We're going to read a piece by Antoine Licht, who has been writing some great stuff about the political economy of AI recently, in a piece that he called Cut Off.
Summing it up quite simply in the subheader, soon, Antoine writes, access to frontier AI will be scarce and selective.
So first we'll read Antoine's piece, and then we'll build from there.
He writes, There's a common mantra in the outskirts of AI policy thought.
Driven by market pressures and overheated capital markets, AI tokens will soon be abundant.
and the future belongs to those who can use them best.
The further you get away from San Francisco, the louder this mantra grows.
It reaches a fever pitch in the peripheries.
The many middle powers of the world still caught up in a plan to navigate the AI revolution on the basis of merely good enough models.
That view requires important AI capabilities to be widely accessible.
Defenders have access to models before attackers do.
Firms in all domains compete based on access to the same AI capabilities.
Recent events have thrown that view for a loop.
And it now seems clear that access to frontier AI will soon be limited by economic and security constraints.
In early April, Anthropic announced it had developed Mythos, a leading cybersecurity model, and that it would only make its considerable ability to patch extant vulnerabilities available to a select few companies.
Cybersecurity startups in the Mission District, systems integrators on the eastern seaboard, and allied capitals on the Atlantic and Pacific all had a similar experience.
scrolling down the page to see the list of privileged partners only to find a limited selection of U.S.-based corporations.
Perhaps you were hopeful that OpenAI was going to stick to its preferred method of rollout, that it would release GPT-5.5 Cyber, a model reportedly similar to Mythos and Capabilities, more broadly.
And yet, it did not.
In their Daybreak initiative, OpenAI too committed to a limited release, dispelling hopes that this was a fluke or doomer marketing.
Even worse, while it's not quite clear to anyone, including the U.S.
government, what exactly the US government will do about all this, by all reports it's at least planning to do something at some point.
And while it's easy to dismiss this as a confluence of current events, the mythos moment actually reveals structural trends that have been ramping up for a while.
Mythos and reality.
Three trends, compute, security, and US government involvement, will further constrain the availability of frontier AI in the future.
They compound and reinforce each other and have dramatically accelerated in recent weeks and months.
Everyone outside the inner circle of US-based developers needs to grapple with that fact.
Security and Distillation The first and most obvious constraint on widespread availability is the one we've seen in the Mythos context.
Security considerations prevent developers from providing top-tier capabilities to every paying customer.
The canonical story starts with misuse risks.
A highly capable new model seems realistically useful for conducting some sort of dangerous activity, such as cyberattacks or biological weapons design.
Instead of rolling it out to the general public right away, you might first distribute it to defenders who can use their early access to shore up vulnerabilities, like we've seen in the case of Mythos.
You continue by rolling out some models only to customers of which you are reasonably sure they won't outright abuse the model for criminal purposes, and perhaps only after the model is no longer state-of-the-art, you roll out to everyone.
Already now we're seeing the second stage.
The US government realizes that this sort of restricted access is better both for national interest and national security.
and starts flirting with the idea of making the virtuous early example into a general rule.
There are many reasons for the national security apparatus to do this.
Perhaps they don't trust AI developers to keep dangerous capabilities away from just as dangerous criminals, non-state actors, and adversaries.
Or perhaps they'd rather like to know which exploits the new models are about to reveal, so they can use them themselves first, as they've done before.
Put differently, If I were the NSA and sitting on a bunch of zero days, I'd also love to know which of them Mythos can find, so I could use them to my advantage before everyone else gets their patch online.
Next to misuse risks, there's another dimension that might motivate even more straightforward crackdowns on availability.
Risks of model theft, espionage, and distillation.
The former would make developers wary of where to host models, waits in an unsecured data center would pose a substantial vulnerability, and many countries outside the US haven't even started thinking about securing data centers.
But the latter, distillation, is the more pressing concern.
Multiple reports indicate that part of the success story of so-called fast followers, model developers six to nine months behind the frontier like China's DeepSeq, is based on distillation practices that require more or less unfettered access to API tokens.
Distillation is not tenable for model developers in the long run.
It will be very hard to capture sufficient revenue if you have to recoup all R&D investments in the six months until someone distilled your model.
That point is extremely salient to politicians, and plays right into latent concerns on US-China competition and industry espionage.
So, I'd expect distillation crackdowns if not from the government, then from developers.
More burdensome KYC, more restrictive default access, more geopolitically motivated access conditions.
None of those bode well for broad-based frontier access.
Compute crunches.
But the trouble does not stop with security concerns.
More fundamentally, providing access to a frontier model is a zero-sum game.
Veterans of the tech industry and European sovereignty hawks both like to invoke the parallel to software licenses.
That yes, software innovation came with some marginal dependencies, but that the logic of consumer market size prevailed in the end.
Microsoft and others faced low marginal costs compensated at full market prices for rolling out their software for everyone.
But not so with Frontier AI.
Providing access to AI models, especially those at the bleeding edge, takes massive amounts of computational resources.
The marginal compute demand to service another thousand tokens is high.
So high, in fact, that leading developers time and time again face compute crunches, reduce offerings, and struggle to balance subsidizing their consumer subscriptions against the real constraints on the chips they have.
So dire is the compute crunch for Anthropic specifically, that the firm is now shopping around for ad hoc access deals to less well-utilized data centers, such as one with rival firm XAI.
It seems likely that this situation would get worse, not better.
If AI systems really do rival the output of human workers in a few months, the amount of tokens required to reproduce that much human activity would be staggering.
The often invoked hope that the efficiency curves will compress token costs quickly doesn't save us here.
Efficiency curves mean that next year, Mythos-level capabilities might be very cheap.
They don't mean that Mythos 2 will be cheaper than Mythos.
The opposite is the case.
Frontier capabilities have grown more expensive month to month for years now.
And if you, like me, believe that competitive dynamics between economic rivals and attackers and defenders mean you need not only good enough AI but the best AI, efficiency curves will not bail you out.
That means the marginal cost of providing access to a new user, country, or firm is high.
There's still value in expanding your coverage.
Inroads into new markets for when your capacity expands, more demand to increase prices, goodwill with governments, and so on.
But these benefits trade off against costs.
compliance costs of entering new markets, product design costs of catering to new customers, and the costs in terms of security in relationship to the U.S.
government described in this piece.
The market power isn't entirely inverted, but it's strongly diminished.
You cannot count on your role as interested buyer to carry much weight in securing your access.
This is complicated even further by the fact that, faced with this trend, competition around who gets access to these tokens will emerge.
The U.S.
will be protective of its domestic economy, And I think we might see a comeback of the same logic that motivated the GAIN Act proposal a few months ago.
Back then, advocates were toying with the idea of giving Americans right of first refusal to American chips.
Soon, perhaps American firms will be declared buyers of first resort of American-produced tokens of intelligence.
Or the competition turns purely economic, margins shrink and become razor thin, and only those who can shoulder the cost, or most effectively turn API tokens into revenue, are able to afford them.
Who would that be?
My bet is neither on governments that haven't internalized the logic of million-dollar AI subscriptions, nor on European businesses constrained in their ability to generate software revenue by many, many adverse conditions.
The U.S.
government is here to help.
Lastly, what starts as restrictions motivated only by genuine concerns doesn't always stay that way.
Once it has a more formal role in overseeing the flow of frontier tokens, the U.S.
government might wield its access control to pursue its political and strategic interests.
That starts with security concerns.
Revisiting the NSA example, it's clearly not in the interest or mission of the NSA to ensure the equitable diffusion of AI capabilities throughout the world.
Instead, it's closer to the intelligence community's DNA to limit any potential adversary's access, even to the detriment of softer upsides like economic productivity or ally relationships.
And it doesn't stop with the security questions.
The Trump administration's signature style of international engagement is to wield American leverage as a bundle.
Deadlocks in trade negotiations are broken by threatening to withhold intelligence.
Tech deals are stalled by reference to food safety standards.
And so, I don't know when a U.S.
administration would choose to leverage its seemingly inevitable pre-deployment authority over frontier models to secure its broader interests, but I'm sure it would in due time.
That means that even if we do everything quote-unquote right on the security and economic side, frontier access is still fundamentally contingent as long as there will be divergences between government strategic interests.
The next equilibrium.
In that new world, access to unlimited APIs is the exception, not the norm.
A new frontier model might first make it to the U.S.
national security apparatus, where embedded interests might decide to stall its deployment for security reasons, wield it first to plug defenses or attack its adversaries.
The model might then be handed back to the developers, with the implicit understanding or explicit demand that it would first be rolled out to trusted defenders, U.S.
firms and perhaps a few internationals if we're so lucky.
If the risks are cybersecurity, the defenders might be quick to resolve them.
If they're thorny or biological or agent-autonomy-driven risks, they might take another few weeks.
Once that phase is over, the circle of unfettered access might expand again to firms that have cleared high KYC bars and U.S.
security concerns.
Everyone else—enthusiastic consumers, scrappy startups, and nervous governments all over the world—might never get clean API access, but draw their access through fundamentally limited product layers.
Maybe the chatbot and coding agent interfaces of today may be the few big startups that could afford to hire the lawyers and lobbyists to make the good list.
A few months after development, the model will have made it into the hands of everyone, but not everyone will have enough tokens to use that capability well, and most might only get to deploy it in ways that trusted vendors have charted out for them.
Only when the next generation has already entered the same pipeline would everyone have access to the de facto unlimited access to Frontier AI that we still enjoy today.
Unevenly distributed futures.
This is not a future we should welcome.
AI tokens will be strategically and economically central to all future societies, so we should do our best to enable their free flow.
If we fail, we'll bear costs, economic and geopolitical.
Economically, I think the accelerationists in their criticism of Anthropic have it right.
Restricting frontier model access to startups and ambitious deployers is antithetical to innovation and economic growth.
Iterative deployment unleashes our ability, call it Heike and a few absolutely must, to actually figure out how we want to live and work with AI at scale and capture its benefits.
But it's not Anthropik's purported quest for nationalization that's at fault.
It's the market dynamics and security implications of advanced AI that send us barreling toward a world where that's no longer possible.
In that world, we'll also see geopolitical rifts opening.
Countries will be divided into the frontier haves and have-nots.
I don't mean to exaggerate when I say...
that those living in the former might be much wealthier and safer than the latter, with access to better public services, greater economic opportunities, and shielded by security agencies that actually operate at the state of the art.
If AI will be as big as I, and many readers of this publication believe, there's no telling what these suddenly emerging asymmetries do to global order.
In the past, when the fruits of industrial revolutions were unevenly distributed, the resulting shifts in relative wealth, security, and power have prompted mass migration, reopen dormant conflicts, and destabilize democracies.
I hope it never gets this far, and there are still many technical and economic trends pulling towards broader diffusion that could bail us out.
But we'd be naive to disregard the dangers of asymmetrically distributing transformative technology in an unstable world order.
One of the most important AI questions right now isn't who's using AI, it's who's using it well.
KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising.
The highest impact users aren't better prompt engineers.
They treat AI like a reasoning partner.
They frame problems, guide thinking, iterate, and push for better answers.
And the good news?
These behaviors are teachable at scale.
If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time.
Learn more at kpmg.com slash US slash sophisticated.
That's kpmg.com slash US slash sophisticated.
Today's episode is brought to you by Robots & Pencils, a company that is growing fast.
Their work as a high-growth AWS and Databricks partner means that they're looking for elite talent ready to create real impact at velocity.
Their teams are made up of AI-native engineers, strategists, and designers who love solving hard problems and pushing how AI shows up in real products.
They move quickly using RoboWorks, their agentic acceleration platform, so teams can deliver meaningful outcomes in weeks, not months.
They don't build big teams.
They build high-impact, nimble ones.
The people there are wicked smart with patents, published research, and work that's helped shape entire categories.
They work in velocity pods and studios that stay focused and move with intent.
If you're ready for career-defining work with peers who challenge you and have your back, Robots & Pencils is the place.
Explore open roles at robotsandpencils.com slash careers.
That's robotsandpencils.com slash careers.
It is a truth universally acknowledged that if your enterprise AI strategy is trying to buy the right AI tools, you don't have an enterprise AI strategy.
Turns out that AI adoption is complex.
It involves not only use cases, but systems integration, data foundations, outcome tracking, people and skills, and governance.
My company, Superintelligent, provides voice agent-driven assessments that map your organizational maturity against industry benchmarks against all of these dimensions.
If you want to find out more about how that works, Go to besuper.ai.
And when you fill out the Get Started form, mention Maturity Maps.
Again, that's besuper.ai.
Weekends are for vibe coding.
It has never been easier to bring a passion project to life, so go ahead and fire up your favorite vibe coding tool.
But Monday is coming, and before you know it, you'll be staring down a maze of microservices, a legacy COBOL system from the 1970s, and an engineering roadmap that will exist well past your retirement party.
That's why you need Blitzy, the first autonomous software development platform designed for enterprise-scale codebases.
Some over-determined solutions.
So, we might want to do something about all this and avert the cutoff scenario.
The solutions at hand are not new ideas, but they are frequently misunderstood.
Sometimes, accelerationists that should agree with my wanting to avert restrictions on model diffusion think they're safetyist plots.
And sometimes, safetyists think they are accelerationist vehicles to hasten dangerous development in disguise.
Neither is the case, and I think all these policies are and should be areas of convergence.
First, we can make the world safer, so the need for security-motivated restraints is less pressing.
Despite the warranted cynicism on instrumentalized access controls, genuine concerns are still what portends future restricted access.
The reason Mythos has scaled the world into action is that firms actually felt vulnerable to the exploits it could find.
If we could harden the world against the most obvious pathways for biological agents to cause harms, build resilience, screen the manufacturing of protein structures, and so on, the Mythos for Bio moment might not move us further down this path, and so on.
By the same token, There are many Frontier Lab employees that argue dealing with distillation is a straightforward technical fix to usage policy and monitoring.
If that's the case, we might want to start deploying fixes soon, before anyone else chooses a more heavy-handed way to deal with salient distillation concerns.
The same goes for global proliferation of advanced models.
Would-be importers would do well to improve their own cybersecurity, including specifically data center security, to not make it a huge risk for American firms to run their models outside the U.S.
Second.
We should simply build a lot of data centers to alleviate the coming compute crunch.
This is not complicated, just fairly hard.
Much has been said about how to accelerate the buildout in the US and elsewhere.
And every GPU we get online this year makes a more equitable diffusion in three years more likely.
If you've objected to rapid buildouts on minimizing the risks grounds, you should think again.
Third, non-US countries should build out compute in exchange for access.
I've made this argument in greater detail before, but in short...
U.S.
allies can offer American hyperscalers favorable terms for data center buildouts in return for frontier access guarantees.
Subsidized energy prices or even outright energy access can be provided to these data centers in return for contractual guarantees to always provide frontier capabilities.
If the hyperscalers or labs renege, they're on the hook for their capital expenditure on the now-unpowered data center.
If the U.S.
government tries to force access restrictions, It faces an angry lobby of domestic tech companies that would rather make do and take the revenue from their international infrastructure investments.
The incentive for going for these deals would have to be commensurate to this downside risk for the investors.
But amidst a frenzy to capture markets and get compute online, I feel optimistic that we'll find a middle ground.
And last, while I'm still confident they cannot resolve this issue by outright building their own frontier systems today, Middle powers will still need contingency options to secure frontier capabilities for the edge cases in which all the above fails and frontier AI access does become the privilege of the few.
Much of it has to do with leverage.
But some will have to go through retaining some ability to build as well.
But that is a much deeper problem.
All in all, frontier AI access is not a new problem, and it does not require particularly clever new solutions.
It just requires taking much more seriously what the center of the AI policy discussion has long suggested we should do.
build infrastructure that can host advanced AI systems at scale, and build a world that can handle them without coming apart at the seams.
If you still needed a wake-up call to execute on that agenda, the Mythos moment should be it.
Otherwise, we really are headed for the end of the Andy Warhol era of AI access, where the rich and poor will no longer have access to the same AI capability.
Back to NLW here.
If after hearing that you are not subscribing to Anton's blog Threading the Needle, I don't know what to tell you.
So to sum up.
The central thesis is that for a set of reasons, security considerations, compute constraints, and geostrategic necessity, the now golden period of fundamentally equal access to state-of-the-art models could very, very soon be coming to a close.
Indeed, one can argue that now that Mythos is in the hands of some and not everyone, we are already there.
Certainly, holding aside all of our annoyances that we don't have the Mythos toy to play with, you're already seeing some amount of this AI inequality show up in the numbers.
Microsoft recently released its Global AI Diffusion Report and found that in the last six months, not only was AI usage higher in the Global North versus the Global South, 27.5% of the population versus 15.4% of the population, AI use in the Global North was also growing at more than twice the rate of the growth in the Global South.
This is, of course, downstream of lots of other types of access issues, including limited access to electricity, internet connectivity, and more.
But the point is that this is already happening.
Now, there is so much to dig into here, given that this is both an economics and a politics argument.
But I just want to talk about the economic side, since that's where we've been for the last several weeks.
Even if the governments of the world were totally committed to equivalent access, in the world we live in right now, as we have discussed, AI companies no longer have the ability to serve the market as many tokens as it wants to consume at anything resembling current prices.
What that means, as we've seen this week, is prices go up.
And in fact, it's not just prices that go up, but the entire structure of access that changes.
The reason the developers got so angry about the changes to Claude's pricing model this week wasn't just because numbers went up a little.
It's because the new model makes the way that they had been building before in many cases completely uneconomically viable now.
It wasn't, oops, this is going to be more painful to me.
It's, I guess I don't get to do that thing that I was doing anymore.
Now, if you want a broader conversation about the risks of building on platforms and all that sort of stuff, you can go back to the conversation we had a couple days ago.
But the point is, these chickens are not just coming home to roost, they are firmly in the coop right now.
Which gets us to the policy conversation.
One of the great ironies of the moment is that the politicians who are loudest in wanting to ensure equitable distribution for AI in the future are also the ones behind a policy most guaranteed to decrease that equitable access and increase AI inequality.
In December, when he started really engaging with these issues, Bernie Sanders tweeted, I will be pushing for a moratorium on the construction of data centers that are powering the unregulated sprint to develop and deploy AI.
The moratorium will give democracy a chance to catch up and ensure that the benefits of technology work for all of us, not just the 1%.
Now, I want to be careful not to misrepresent Bernie and AOC's position here.
This is not just a misunderstanding of economics.
If you are in the camp, that AI is going to take all of the jobs, then slowing down AI so that it can't get to the point where it takes all the jobs becomes politically rational.
However, to the extent one believes, as it kind of seems like Bernie does, that this technology is inevitable.
And to the extent that you actually want to, as he puts it, ensure that the benefits of technology work for all of us, not just the 1%, there is almost no policy less likely to do that than pausing data center construction.
You guys are smart, so I don't need to really explain supply and demand to you.
But the TLDR of the pathway is less data center construction means more scarce compute, which means access gets rationed and costs more.
And so users that are already rich and big companies get to keep access and students, independent builders, small businesses, and the less resourced in general get pushed to weaker tiers.
Now, if Bernie and AOC are doing an almost Trumpian art of the deal thing here where they introduce a moratorium.
so that they can enshrine into law all the ways that data center construction has to just pour value and resources into the communities where those data centers are being constructed, then God bless and good luck.
But the policy on the face of it is going to make things worse, not better, if it came to pass.
None of this ultimately is completely inevitable, with the possible exception that in the short term at least, we do not have enough compute to give everyone access to all of the AI they want.
But when it comes to how we build from here, how we resolve that situation, and all of what Anton discusses in his essay.
As Lawrence of Arabia said, nothing is written, at least not yet.
This is quietly going to be one of the most significant political moments of our lives, and I'm glad we get to talk about it here.
For now, that's going to do it for today's AI Daily Brief.
Appreciate you listening or watching, as always, and until next time, peace.