# Open-Weight AI Models, Regulatory Shifts, and Compute Constraints

**Podcast:** The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis
**Published:** 2026-07-21

## Transcript

Today on the AI Daily Brief, why everyone is debating AI policy and the battle for the future of open source AI.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
All right, friends, quick announcements before we dive in.
First of all, thank you to today's sponsors, KPMG, Blitzy, Section, and Airtable.
To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts.
And if you want to learn more about sponsoring the show, send us a note at sponsors at AIDailyBrief.ai.
Today's episode is one of those ones where the headlines are all kind of connected to the main episode, so we're just doing it as one big episode.
We will be back presumably with our normal format between headlines and main tomorrow.
But for now, let's talk the battle around open AI.
Before we get into today's show, let me pitch you on why you should pay attention.
Now, if you're listening, maybe you don't need to be pitched, but my observation over the past few years of doing this show...
is that there is a big chunk of this audience who is primarily focused on how AI matters for you.
Specifically, this part of the audience cares about new models and changes and harnesses in the way that we access those models.
They are primarily interested in how AI is going to change their work and their careers and what they do.
They're interested in how AI opens up new opportunities for them.
And this is, of course, why a lot of this show is biased towards the highly actionable.
It's also why...
Almost inevitably, there is a little dip in engagement around shows that are more on the policy side or on the macro big picture geopolitics side.
Today's show is entirely about that.
But my very strong argument is that what we are discussing today is an area of policy and geopolitics that could have dramatic implications for what models you have access to.
how you access them, the cost at which you access them, the ways that you design systems for work, the tools that are available to you to design those systems.
The stakes, in other words, are extremely high, not just on some random theoretical level, but in terms of the AI that you actually access.
What's more, this particular area is a policy discussion which blends into democratic politics in the US in a major way.
And so I believe and hope is worthy of your engagement.
So what we're talking about today is, of course, the increasingly loud political questions around open source.
And specifically, but not exclusively, we're talking about Chinese open-weight models and whether they have a future in the US.
Now, the genesis for this, of course, is the release of Kimi K3, the model that was the subject of discussion in our Friday episode.
Now, at this point, you might be a little anesthetized to every time there is some new advanced Chinese model, everyone having a freakout in the style of the original DeepSeek freakout back from January 2025.
It's almost at this point kind of expected.
And yet, honestly, since that first deep seek moment back in January, I haven't seen something dominate the discourse quite as loudly as this has over the last few days.
Pretty much all anyone talked about all weekend was a tweet from Dean Ball, former policy advisor to the Trump administration on AI, now head of strategic futures at OpenAI, which became just incredibly contentious and controversial, as we will see, generating at this point 11 million views.
But before we get into Dean's post, let's take a step back and talk about what we know so far around the U.S.'s position over open source, as well as the latest out of China around their strategy.
Over the past couple of weeks, there have been a growing set of reports that suggest that the White House is taking a closer look at taking action against open models.
Last Wednesday, Washington Insider publication Semaphore reported that the administration was considering or at least not ruling out action on open source models.
with a senior White House official confirming that there was, quote, plenty of ongoing work that went beyond the cybersecurity executive order from June.
At the very least, it's very clear that the White House is paying close attention to open models.
Now, of course, the White House's relationship with models in general is in something of a flux moment.
On Friday, CNBC reported that the administration expects to limit the release of Western frontier models on an ongoing basis.
CNBC highlighted the limits around Fable 5 and GPT 5.6, but also tied the policy to the new AI clearinghouse announced last week named Gold Eagle.
It was originally believed and frankly framed like the clearinghouse would be mostly about sharing software vulnerabilities detected by AI, but sources indicate that Gold Eagle will also be the mechanism to determine which companies have access to new frontier models.
On the record, a White House official said the government doesn't require approval of AI models and any engagement is still voluntary, asserting that, quote, Decisions on timing and scope of releases rest entirely with the companies.
Now, I'm sorry to be cynical, but that is absolutely not the case anymore, as we have seen.
And the White House can pretend that this regime is voluntary all at once, but functionally speaking, that is no longer the case.
The only question is whether moving forward, the quote-unquote voluntary regime that seemed to be over the last month Howard Lutnick and Susie Wiles getting together and deciding when they thought Anthropic had eaten enough crow for them to allow them to re-release Fable 5 becomes actually something more formalized in the future.
Now, for those who think that basically anything formalized would be better than this weird de facto informal regime, over the weekend, Bloomberg noted that the White House is also considering a proposal for a self-governing body put forward by Demis Hassabis.
Financial industry regulator FINRA is considered the model, but then again, many have noted that financial services aren't exactly known for their rapid pace of innovation under this particular model.
Now, on Monday...
Axios published a provocatively titled article suggesting that there was a secret effort within the White House to curb Chinese AI.
They wrote, The Trump administration is showing signs it could ban cutting-edge Chinese AI models, a momentous move it could lock in dominance by OpenAI and Anthropic.
A source close to the administration detailed some of the plans.
They claimed that last year, the Commerce Department considered adding Chinese AI firms to their entity list, which would discourage domestic use in the corporate sector.
An executive order is also being considered which would require U.S.
tech companies to only host Chinese models if they can guarantee security and take liability for any breaches.
They said the Commerce Department has also circulated draft rules that would leverage supply chain security powers to crack down on Chinese AI.
Now, one important thing to note as we're discussing all of this is that the White House understands that they don't necessarily have to actually outright ban Chinese models to have their desired effect of prohibiting access to them.
One source familiar with the discussions described a push to highlight potential backdoors and a lack of security.
And frankly, anyone who's paid any attention at all to the Operation Chokepoint regimes that have happened across various administrations over the past few years with regard to undesirable areas.
So right now, it's very unclear where this is going to land, but also does seem clear that these conversations are being had.
Now, adding a bit of a twist to the discussion coming out of Washington, the Trump administration's head of the Center for AI Standards and Innovation has resigned.
Now, this organization was set up during the Biden administration and had its power curbed early in the Trump administration, although had seen a resurgence in recent months due to the role it played in assessing Fable during the ban.
Many saw the center as a way to have civilian input over AI regulation rather than leaving the matter entirely to the NSA.
On Monday, CNBC reported that Chris Fall, the head of the center, had resigned.
Fall had only been in the position for three months, so was a Trump administration pick for the role.
Axios characterized the departure as abrupt.
There is no immediate replacement with National Institute of Standards and Technology Director Arvind Rahman filling the role on an interim basis while he interviews candidates.
Analyst Max Weinbach wrote, I hope this is because what he was pushing was stupid and people called him stupid rather than the one I'm terrified of, which is he's fighting for the smart solution and it isn't working.
Now, as I've frequently said, over-reading a single personnel change is always a little bit dangerous, but the timing does make it at least a little bit notable.
Overall, USAI policy is in an extraordinarily confused place.
In a piece titled, Trump's AI Agenda Collides with Reality, the information wrote, Inside the White House, turf battles, clashing views, staff turnover, and hollowed-out offices have contributed to a chaotic environment for policymaking on AI.
Many have noted that the administration's apparent White House policy has gone from hands-off to extraordinarily heavy-handed in a very short period of time, and seems to be veering wildly between the two even inside.
But of course, when it comes to AI policy, it's not like the U.S.
government is sitting out there alone.
They have a counterparty in Beijing who is likewise going through its own process of evolution and codification of AI policy.
Chinese officials recently wrapped up the first World AI Conference in Beijing.
Now, heading into the event, it was clear that the purpose was to offer a Chinese-led AI future to the global South and U.S.
adversaries.
The Chinese Foreign Ministry touted 29 signatories to a new World Artificial Intelligence Cooperation Organization, including some of the U.S.'s favorite people in Russia, Indonesia, Pakistan, and Laos.
The event itself was headlined by a speech from President Xi Jinping, who endorsed an open-source approach to global AI.
She said that global AI governance must, quote, uphold openness and win-win cooperation to drive innovation and development.
She added, AI is a new engine of global economic growth and an accelerator in the transition from old to new growth drivers.
It is moving from the digital world into the physical world.
We must seize this rare historic opportunity.
encourage open source development, openness, cooperation, and sharing, and comprehensively advance technological innovation, industrial development, and real-world applications of AI.
We should coordinate efforts to transform and upgrade traditional industries, foster and expand emerging industries, and make forward-looking plans for industries of the future, thereby empowering all sectors through AI.
Now, there had been some questions of late around whether China was going to decide to go in the other direction and start restricting access to its top models that were coming out of labs like Moonshot.
But at the moment, at least, it seems like the narrative thrust is firmly focused on this open strategy.
Geopolitics commentator Arnaud Bertrand wrote, It's becoming clearer and clearer that China's AI open source strategy may end up being seen as one of the greatest strategic masterstrokes of all time.
They started with a clear resource and technological disadvantage, mainly due to the U.S.
semiconductor export controls, and have managed to change the rules of engagement in such a way that the U.S.'s own tech leaders and officials are now publicly siding with China's approach against their own companies, which is pretty extraordinary.
When you can't fight symmetrically, make the adversary's way of fighting obsolete and self-defeating.
The greatest irony in all of this is had the U.S.
not done the export controls, there's a decent chance that not only China wouldn't have gone for the open source approach, but the U.S.
would have made an enormous amount of money selling compute to them.
Now they're getting neither the money nor the containment.
Now, according to reporting from the Financial Times, the consultations between the Chinese Ministry of Commerce and AI companies about possible export controls continues.
So I don't think that we should take anything of this moment as a given going forward, but it's clear from the speech that at least when it comes to the global story, China is positioning itself as the great defender of open AI.
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.
You've tried in IDE co-pilots.
They're fast, but they only see local silos of your code.
Leverage these tools across a large enterprise codebase and they quickly become less effective.
The fundamental constraint?
Context.
Blitzy solves this with infinite code context, understanding your codebase down to the line-level dependency across millions of lines of code.
While co-pilots help developers write code faster, Blitzy orchestrates thousands of agents that reason across your full codebase.
Allow Blitzy to do the heavy lifting, delivering over 80% of every sprint autonomously with rigorously validated code.
Here's a harsh truth.
Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.
Half of companies have AI tools, but only 12% use them for business value.
Most employees are still using AI to summarize meeting notes.
If you're the one responsible for AI adoption at your company, you need Section.
Section is a platform that helps you manage AI transformation across your entire organization.
It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.
The result?
You go from rolling out tools to driving measurable AI value.
Your employees move from meeting summaries to solving actual business problems.
And you can prove the ROI.
Stop guessing if your AI investment is working.
Check out Section at sectionai.com.
That's S-E-C-T-I-O-N-A-I dot com.
This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together.
New users get $1,000 in inference.
Forget local agents and chat workflows waiting on your laptop to be prompted.
HyperAgent deploys always-on agents in the cloud, doing real work across the tools your team already uses.
Marketing's agent turns competitor moves into landing pages.
Sales as agent enriches leads, drafts emails, and updates the CRM.
Ops agent chases the paperwork and tracks the budget.
Every agent has access to shared context and follows your rules about scope and approvals.
It's time you add agents that feel like teammates.
Hire yours at HyperAgent, built by the team at Airtable.
Claim your $1,000 in inference at hyperagent.com slash AI Daily Brief.
All of which gets us to the tweet, which triggered the latest round of conversations, not just on Twitter, but all across the actual AI policy world.
Dean Ball, again, former White House advisor and now head of strategic futures at OpenAI, began his tweet with mild praise of Kimi K3, recognizing that the model is pretty much on par with where the U.S.
labs were in Q1 of this year.
Ball then expressed surprise that the Chinese government was still allowing open-weight models to be released, given the risk posed by the new generation of ultra-large models.
Continuing with a line that would cause a ton of consternation, Dean wrote, Open-weight models are inherently decelerationist.
and I'm continually surprised to see the so-called accelerationists so excited about open-weight models.
I suspect the reason they are is that they know open-weight models are effectively ungovernable, and they simply like the overall cloak of ungovernability open-weight models creates over the whole of AI.
It's not a bad strategy.
It reminds me of James Scott's recounting of the Hill People in The Art of Not Being Governed.
Still, at the end, open-weight models deter further AI capex.
The argument, of course, being the same one that some market skeptics are making.
which is that if cheaper open-weights models can do everything or close to everything that Fable 5 and GPT 5.6 Sol can do, why the heck would customers spend a premium to buy those frontier models when they can get the good enough models much cheaper?
And by extension, if those customers weren't buying those models anymore, why would investors continue to fund the infrastructure build-out that is powering those companies?
And so on and so forth until all of a sudden we have a big market crash on our hands.
Now, of course, there is the less dramatic reading of this that doesn't necessarily implicate a full-on market crash.
but they can still recognize that on the margins, the availability of near-frontier open-weight models would potentially dampen revenues for open AI and Anthropic, and make further investment in both model training, as well as infrastructure build-out to support model training, more risky.
Dean then attempted to describe where he saw this going, writing, One probable outcome of an open-weight model-dominant world is full AI communism, which is precisely what China proposes.
Rather than a market product, AI is a public good, which will ultimately be provided by the state as a kind of digital public infrastructure.
Dean continues, This future strikes me as a dystopian hellscape, but I've never met an open weights model advocate who doesn't ultimately concede this is where things end.
You'd be surprised how many accelerationists lobbied me while I was in government to support an 11 or 12 figure federally funded data center so that startups could train models at a subsidy and then give them away for free.
There was no other way for AI to progress, they said.
Perhaps this is the logical end state of things.
Nonetheless, I find myself surprised to see supposed accelerationists excited about such an outcome.
I think many of them just don't know what they're doing.
Many accelerationists do not view the creation and serving of frontier models as a legitimate business.
Turning back to the U.S., Ball continued, and honestly, this is where he really stepped in it.
I would guess that the Trump administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models.
You don't need to quote-unquote ban open source.
You just need to direct every agency to issue soft law that creates FUD.
For example, a Federal Reserve advisory bulletin found that there may be backdoors in Chinese AI models.
It needn't be that well justified.
You just create enough regulatory risk that every regulated enterprise backs off.
You probably don't want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models.
This will just drive startups to sketch your providers.
There's a happy middle ground here.
I'd assume they will do some version of this.
Now, this was the section that really had people's flabbers gasted.
As they took it as Dean effectively advocating for some version of Operation Chokepoint-style tactics, to use the soft power of the government to soft ban these models by creating so much risk around them that companies would effectively just ban themselves from using those models.
People jumped all over this.
Epic CEO Tim Sweeney responded, Picture an executive of a taco company saying this sort of thing about a new brand of tacos coming onto the market, speculating about the geopolitical and societal disruptions they anticipate as a result of advances in tacos.
Cloudflare engineer Dylan Mulroy wrote, Actually an insane thing for OpenAI's head of strategy to publicly say.
Deep Dish Enjoyer wrote, LMAO, OpenAI, admits it does not want fully automated luxury space communism.
They are openly admitting they want techno-feudalism where they own everything.
Entrepreneur Brian Atwood wrote, This is grotesque.
I bet Dean is a good, smart guy, yet he is trying to convince you that hosting an LLM in your basement, private and sovereign, as the founding fathers would have wished, is dystopian communism.
Now, why would he do that?
And of course, as you can see, a lot of the critique is not just around Dean's argument, it's the fact that Dean is now...
an extension of OpenAI making that argument.
Ben Norton wrote, This guy who works at the poorly named OpenAI, more accurately ClosedAI, laments that China's open source models could lead to, quote, Norton continues, But of course, people who work at OpenAI claim it would be a dystopian hellscape because they would not have a monopoly on AI and could not become trillionaire techno-feudal lords by forcing everyone to pay them digital rents.
Others made the comparison to Steve Ballmer comments on Linux back from the beginning of this century.
Qualia Script writes, It's 2001.
Open source Linux is better than Windows on servers.
Steve Ballmer calls Linux cancer, communists, and asks for it to be regulated away.
It's 2026.
Open source LLMs are better than ChatGPT and Claude on costs.
They are called decelerationist and communist.
Now, after all of this, Dean later came back, both to recognize that he effectively doesn't get to tweet the way that he used to anymore now that he works for OpenAI, but also to try to tidy up a few points.
He tried to clarify that this wasn't supposed to be a prescription for the Trump administration, just what he sees as the most likely scenario.
He also described his longstanding support of open source software and gratefulness for what it's brought to the world.
However, he concluded...
I think it's pretty clear that we are approaching the point I described, the point where, absent a major technical safety breakthrough, the national security implications of frontier open-weight model distribution are simply too severe.
I don't think we're there yet, but the direction of travel is clear, and an analyst must be honest about this.
Governments will realize these risks eventually, and when they do, they will have much lower risk tolerance than I have.
We see this today with the Trump administration, which once proudly championed open source AI, and now has a de facto licensing regime for frontier AI that I suspect will make it challenging if they still end up enforcing it to release the weights of models of the mythos tier.
Every government will be safetyists once they understand themselves to be in the foxhole.
You don't have to like this, I don't, but it is the reality as I see it.
Now, speaking of the Trump administration, they were among the folks to jump all over Dean around these posts.
In comments very clearly befitting the importance of his role, Undersecretary of War Emile Michael called Dean Ball the supreme village idiot for AI.
While former AI czar David Sachs wrote, I'm not sure whether Dean Ball is confessing to a regulatory capture strategy or simply predicting this will happen, he now says the latter.
Either way, the weaponization of regulatory uncertainty as a competitive tool should be completely unacceptable.
He argues there's no need to ban Chinese open source models, just direct agencies to issue soft law warnings that create enough FUD so regulated enterprises back off.
Wrong.
Regulatory decisions should always be well justified and grounded in facts, logic, and evidence, not the deliberate exploitation of fear and uncertainty.
Implementing a surreptitious policy through manufactured doubt, rather than strong and explicit justification, corrodes the rule of law and invites future abuse against anyone.
We are at a critical inflection point in AI policy.
The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition.
They have laid their cards on the table.
It's time for the rest of Silicon Valley.
the vast majority that still values open competition, to do the same.
Dean Ball clapped back, Now, for whatever influence David Sachs still has at the White House, he seems to not be in favor of the sort of bannings that are being discussed.
Talking about Chinese models and their advanced capabilities, he said, This is exactly what I predicted would happen.
I said Chinese models would have advanced cyber capabilities within a matter of months, and the only thing to do about it was to use AI-powered cyber defense to protect our systems.
Trying to gatekeep models doesn't work.
Box's Aaron Levy agrees, saying it's fairly obvious that gatekeeping models will not work at scale.
Competing in AI is too economically and strategically important for China at this point, and we've now crossed the Rubicon where it's clear that they can compete at near-frontier levels.
The solution to this isn't to get more locked down and slow your own ecosystem.
If that happens, you can guarantee that America loses the global battle.
The solution is to safely ensure that you keep a high rate of progress and drive diffusion of the technology, build out infrastructure, enable US open source software, and more.
But given that almost everyone seemed to be against Dean, is there any merit there?
A few people tried to look dispassionately at what he was saying and give him the benefit of the doubt.
GrowingDaniel on Twitter wrote, I'll take a hack at Dean's argument without his conflict of interest.
What China is doing in AI is called dumping.
They do it in literally every industry they enter.
The goal is to kill all local competition by subsidizing their own industry so they can produce at a loss, then once all competitors are dead, they can charge profitable prices and control the market.
In steel and automobiles, this is just bad.
In AI, it's potentially fatal to our country.
Unfortunately, dumping is a very common argument for rent-seeking domestic firms who want protection, and Dean's position here seems to be a Jones act of sorts for AI.
This has famously not saved our shipbuilding industry, and I suspect it won't save our AI labs.
Stopping open waits is virtually impossible, so our only other option is governments taking stakes in labs and subsidizing our own industry.
Investor Hasib Qureshi also argued that this is what Dean was trying to say, that quote, releasing the waits for a frontier-level model is effectively dumping.
Former meta-chief AI scientist Jan LeCun pounced on Hasib, writing, so releasing Linux was dumping?
Apache?
MySQL?
PHP?
The open-source stack of the mobile communication network?
Signal?
PyTorch?
Lama?
To which Hasib responded, To be clear, I don't agree with Dean and I oppose his call for state intervention, but I'm explaining his argument because most people refuse to actually engage with it.
He has a point that if Chinese frontier labs are now being encouraged to be totally open, China now sees this as explicitly part of their strategy.
I would not assume that this is altruistic but calculated, unlike the traditional OSSU layout here.
If all of the Chinese labs are extremely unprofitable, and they are, and they are encouraged at a state level to remain unprofitable, it is likely to have large and reverberating economic consequences on USAI as well.
That's Dean's point, and I think it's worth taking seriously.
I don't think China is encouraging this strategy with the same spirit of the people who built Linux.
Now, Philippe Lemoine points out, and this is exactly why we're having this conversation on this show as well, that this is not just a Twitter debate, but it's a conversation that is happening in the halls of power right now.
Philippe writes, It's now clear that Dean Ball's post wasn't random, but was a public manifestation of a debate that is currently taking place within the Trump administration about how to deal with Chinese open-weights models, pitting advocates of competition, who probably have their own ulterior motives but still defend the U.S.
consumer in this case, against an unholy coalition of industry lobbyists, geopolitics-pilled people, and AGI-pilled people who are trying to, quote-unquote, protect U.S.
consumers from the benefits of competition, by variously arguing that not doing something to hinder the deployment of Chinese open-weights models in the U.S.
would destroy American AI companies, empower the CCP to harm Americans, or push back the advent of the machine god.
Now, to the extent that we are trying to take the conversation forward, In a weekend piece for the American Enterprise Institute, Ryan Fedasek added an important aspect to the conversation.
He noted that, yes, the conventional wisdom around how far behind the U.S.
Chinese frontier models were has changed, but that there is a very important part of this competition that goes beyond model benchmarks that we need to consider.
He noted that while China may have frontier training capabilities, they don't have anywhere near close enough compute to serve their models to the world.
Who, by the way, spent two years as the State Department's main contact with the Chinese embassy, quote, He concluded, Kimi K3 is an important milestone in the U.S.-China AI competition, and Americans should treat it as one.
Not as a Sputnik moment demanding panic, but as the formal close of the era in which model capability alone conferred lasting advantage.
Frontier AI is quickly becoming a commons.
The race now is to build industrial systems that put the frontier to work.
Now, reinforcing the point that China is severely compute constrained, Moonshot pulled Kimi K3 over the weekend.
On Sunday, they posted, Kimi K3 has received far more love than we expected and our GPUs are feeling it.
Over the past 48 hours, demand has pushed close to the limits of our current capacity.
To protect the experience of existing subscribers, we're temporarily pausing new subscriptions and prioritizing compute for current members.
We're adding capacity as fast as we can, and we'll reopen new subscription spots and batches.
Now, running out of compute is pretty normal following a major model release in the West, but this is the first time we've seen it from a Chinese lab.
Moonshot made the curious decision to launch entirely on their own servers, while previous high-profile Chinese releases had day-one inference partners in the US.
Now, the interesting part is what this says about just how constrained the Chinese labs actually are on inference.
Only hardcore AI enthusiasts were clearing the weekend to test the latest Chinese model, and that was still enough to knock over their servers.
It suggests, in short, that Chinese labs don't have anywhere near enough capacity to serve AI models to the world.
Council on Foreign Relations Chris McGuire jumped all over this, writing, Kimi admits it is compute constrained and is struggling to serve K3.
The same thing happened to DeepSeq when it released V4.
When Chinese AI labs say their number one constraint is compute, they aren't lying.
They don't have enough chips to serve the model at scale to customers.
If we stop China from buying, smuggling, or remotely accessing AI chips, it will be harder for them to either make advanced AI models or serve them at scale.
But instead, we are selling them the compute capacity they need most and have loosened restrictions on smuggling and remote access.
We are making it easier for China to catch up and are acting surprised when they release good models.
The good news is if we start closing loopholes in our export control policies and enforcing them more vigorously, we can still constrain China's future AI capabilities.
But this is the consequence of our non-serious approach to export controls over the past 18 months.
Family office investor Ricky Ho wrote, The most important takeaway is not that Kimi K3 is good.
We already knew that.
The real signal is that demand for Frontier AI is now being constrained by compute rather than customers.
Moonshot is effectively saying that it has found product market fit faster than it can deploy GPUs.
Ironically, this also highlights the biggest misconceptions surrounding open-weight models.
While the model weights may be free, inference is not.
Serving millions of users still requires enormous investments in GPU, networking, power, memory, and data center infrastructure.
Open-weights eliminate software licensing costs.
They do not eliminate physics.
Now, when push comes to shove, I'm not sure that I think that this is really the moment where anything dramatic shifts.
I think we are still heading towards the crescendo rather than having reached it when it comes to policy regarding Chinese open-weight models.
I think that pretty soon we'll get an announcement of Fable 5.1 or GPT-6, and all the attention will shift once again.
Professor Ethan Malek put it this way.
Regardless of what you think the answer should be, the inherent tension between a growing U.S.
regulatory and approval regime for frontier closed models and the lack of one for open models is going to need to be resolved in some way or another in the near future with large consequences.
Which way do things go?
1.
Approval regime for all models, official and unofficial, make life difficult for AI labs that do not do it.
2.
No required approval or true voluntary.
3.
Approval for closed waits not open, that's today or it's reverse.
4.
Bless or ban individual labs.
Now, if you need any more evidence that this conversation is going to get louder, not quieter, CNBC's Jim Cramer waded into it saying, we must not let our companies use these Chinese models to save a few bucks.
OpenAI and Anthropic are correct.
This is vital national security.
One way or another, this is going to impact how we build and use AI.
So much of our conversation this year has been about how to deal with the rising costs of AI and how to design more complex architectures that can route tasks to different models.
Different policy decisions or lack thereof could point in entirely different directions and market incentives for different models with huge consequences to come.
Even if you are just using AI as a consumer, even if you are just thinking about it in terms of how you're going to help your business's AI strategy, like it or not, this conversation is going to impact you.
And so now is a good time to start paying attention.
Of course, for those of you who have made it this far, you are getting gold stars in paying attention, and I appreciate you.
Thanks as always for listening or watching, and until next time, peace.