Is Global AI Cooperation Even Possible?

0:00Today on the AI Daily Brief, China's new AI action plan and an exploration of whether AI collaboration is even possible. Before that, in the headlines, developer Doomsday as Anthropic is forced to throttle Claude. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:23Hello, friends. Quick announcements before we dive in. First, thank you to today's sponsors, KPMG, Blitzy, Vanta, and Superintelligent. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief. And if you're interested in sponsoring the show, shoot me a note at nlw at breakdown.network. And with that, let's get into the headlines. Welcome back to the AI Daily Brief headlines edition, all the daily AI news you need in around five minutes. We've talked frequently on this show about just how much developers are using Claude. Some of them, it turns out, are using it so much that Anthropic has actually had to introduce some new limits.

1:00TLDR is that new weekly rate limits are being introduced at the end of next month for both the$20 a month pro plan as well as the premium$100 and$200 a month max plans. They wrote, Claude Code has seen unprecedented demand, especially as part of our max plans. Some of the biggest Claude Code fans are running it continuously in the background 24-7. These uses are remarkable and we want to enable them, but a few outlying cases are very costly to support. For example, one user consumed tens of thousands in model usage on a$200 plan. McKay Wrigley of Takeoff AI, who has been running CloudCode 24-7 as a computer use experiment, responded, Sorry about that.

1:39Now, Anthropic claimed that less than 5 % of users would be impacted, and that those that absolutely have to run CloudCode all day can buy more usage at API rates. The policy change also comes a few weeks after Anthropic introduced rate limits without informing customers in a really clear way. Power users flocked to CloudCode's GitHub page to complain, and while most acknowledged that they were stretching the deal, churning through thousands of dollars of inference a day on the$200 a month plan, in general they wanted clear communication rather than a random rate limit message. Now, Anthropic's status page has showed seven outages for CloudCode over the past week, so resource constraints are clearly becoming a problem again.

2:14More broadly, these rate limits call into question just what the right price for powerful agentic AI coding tool should actually be. Cursor and Replit have also made pricing and usage adjustments in recent months as coding agents became more capable. And the issue is, this is not some example of people abusing the system. It is increasingly what power usage looks like. These tools now allow you to fire off multiple agents at once and run tasks continuously in the background, which might be expensive but is what power user workflows are increasingly about. Now the good news from a long-term perspective is that the cost of these models has come down precipitously, much, much faster, I think, than anyone anticipated.

2:52But at the moment, power users are going to face some new constraints. The response has been pretty varied. On the one hand, you have folks like Oscar Lee who writes, Anthropic are driving your most power users away, who are also your most avid brand ambassadors. They will go to another provider and that may very well trigger the downfall of Anthropic. On the other hand are people like Ian Nuttall who writes, even with these new weekly limits, there is nowhere else you can get this level of usage for$200. Walker Reynolds writes, in case you're wondering whether or not AI is changing the game, just a little insight into how much power users are leveraging agents for development.

3:24And indeed, that's the most interesting thing here to me. Fundamentally and already, our appetite for intelligence is becoming unlimited. Next up, we have a new player in the AI browser wars, and that is Microsoft. The company has introduced a big overhaul to their edge browser, now including a feature called co-pilot mode, which brings the AI assistant to the web user experience. The feature is sort of halfway between Google's limited Gemini assistant in Chrome and Perplexity's fully agentic Comet browser. Copilot has some agentic capabilities, able to browse, for example, through different tabs to compare pricing or extract the information from a crowded recipe blog, for example.

4:00It doesn't, however, go all the way to being able to fill a shopping cart or book flights. Copilot is able to analyze multiple open tabs, and voice controls are also included in this iteration of the feature. Microsoft says they plan to bring fully agentic capabilities to the new browser update soon. Sean Linder, the VP of Product for Microsoft Edge, wrote, As AI begins to reshape nearly every facet of digital life, we're witnessing a turning point in how we interact with the web. Now it's worth asking, is your browser working for you as much as it should? He added, This is just the beginning of our journey in introducing new AI innovation into your everyday browsing.

4:32Copilot mode is experimental and will evolve over time, and we're just getting started. Based on the way the team is framing it, at least initially the ambition seems to be much more about simple improvements rather than radical overhauls. Microsoft's Mustafa Suleiman said, No more flipping between a million tabs. Copilot can cross-reference for you instead. CEO Satya Nadella wrote, My favorite feature is multi-tab rag. You can use Copilot to analyze your open tabs, like I do here with papers our team has published in Nature Journals over the last year. Still, it's very clear that this is just step one into this new agentic browser era, and that more is on the way.

5:05An update in the one AI wearable that has any sort of real traction, Meta's Ray-Bans are apparently still flying off the shelves as AI features come to the fore. Ray-Ban maker Esselor Luxottica reported earnings last night, disclosing that sales of Meta's smart glasses tripled in the first half of the year. Now, they didn't disclose the actual numbers, so we can probably assume that the tripling was from a fairly modest base. Still, the sales success was important enough that it was featured prominently in an earnings call. Luxottica and Meta have already expanded their partnership to include an Oakley-branded pair of smart glasses, with a Prada pair to be introduced soon.

5:39Meta has also purchased a 3 % stake at the eyewear manufacturer, looking to cement a decade-long partnership. Now, Meta's smart glasses are at this point undeniably the most successful AI device play so far. They introduced the Ray-Bans in late 2023, well ahead of the company having any sort of functional AI interface. Now, of course, it is not 100 % clear where smart glasses will fit in the overall AI wearable landscape. But for now, it's meta and meta alone playing in that space. Lastly today in the headlines, Tesla has signed a$16.5 billion deal with Samsung to produce their AI chips in the US.

6:14The chipmaker announced the deal on Monday, which will secure Tesla as an anchor customer for their upcoming plant in Texas. The agreement runs until 2033 and will see Samsung produce Tesla's next generation AI6 chip, which will power both vehicle self-driving and the Optimus robots. Elon Musk posted, Samsung's giant new Texas fab will be dedicated to making Tesla's next-generation AI6 chip. The strategic importance of this is hard to overstate. Samsung currently makes AI4. TSMC will make AI5, which just finished design, initially in Taiwan and then Arizona. Samsung agreed to allow Tesla to assist in maximizing manufacturing efficiency.

6:48This is a critical point, as I will walk the line personally to accelerate the pace of progress. And the fab is conveniently located not far from my house. He also added that the$16.5 billion contract value was, quote, just the bare minimum. Actual output is likely to be several times higher. Now, Tesla started producing their own chips in 2019, switching away from NVIDIA's drive platform. The goal was to create a custom chipset that featured the redundancies required for full self-driving, not just advanced driver assist. For Samsung, this is a huge deal that could keep them in the AI race. The company's high bandwidth memory has been highly successful, becoming the standard for NVIDIA's leading AI chips.

7:23At the same time, their chip manufacturing has been dogged by quality control issues that drove customers to TSMC. Said one market commentator, their foundry business has been loss-making and struggling with underutilization, so this will help a lot. Tesla's business may also help them to attract other customers. Now, obviously, the undertones of this are all about the global competition for AI infrastructure, which is also, consequently, the focus of our main episode. So with that, let's wrap the headlines and move on over into the main. Today's episode is brought to you by KPMG. In today's fiercely competitive market, unlocking AI's potential could help give you a competitive edge, foster growth, and drive new value.

8:00But here's the key. You don't need an AI strategy. You need to embed AI into your overall business strategy to truly power it up. KPMG can show you how to integrate AI and AI agents into your business strategy in a way that truly works and is built on trusted AI principles and platforms. Check out real stories from KPMG to hear how AI is driving success with its clients at www.kpmg.us.ai. Again, that's www.kpmg.us.ai. This episode is brought to you by Blitzy, the enterprise autonomous software development platform with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale code bases with millions of lines of code.

8:45Enterprise engineering leaders start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and precompiles code for each task. Blitzy delivers 80 % plus of the development work autonomously while providing a guide for the final 20 % of human development work required to complete the sprint. Public companies are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with their coding co-pilot of choice to bring an AI native STLC into their org.

9:15Blitzy is providing a limited time, 30-day free proof of concept for qualifying enterprises. The team will provide a 5x velocity increase on a real development project in your org. Visit blitzy.com and press book demo to learn how Blitzy transforms your STLC from AI-assisted to AI-native. That's blitzy.com. As a founder, you're moving fast towards product market fit, your next round, or your first big enterprise deal. But with AI accelerating how quickly startups build and ship, security expectations are higher earlier than ever. Getting security and compliance right can unlock growth or stall it if you wait too long.

9:51With deep integrations and automated workflows built for fast-moving teams, Vanta gets you audit-ready fast and keeps you secure with continuous monitoring as your models, infra, and customers evolve. Fast-growing customers like Langchain, Writer, and Cursor trusted Vanta to build a scalable foundation from the start. And look, as someone who lives in the world of enterprise procurement, I love how Vanta makes it easy to get compliance right. The last thing you need when you're trying to win that big deal is to have it scuttled by something that Vanta has solved for over 10 ,000 companies. Go to vanta.com slash NLW to save$1 ,000 today through the Vanta for Startups program and join over 10 ,000 ambitious companies already scaling with Vanta.

10:29That's V-A-N-T-A dot com slash NLW to save$1 ,000 for a limited time. If you are a regular listener, you will have heard about Superintelligent's agent readiness audits at this point. But I wanted to tell you today about the full suite of agent readiness products that go beyond just the initial readiness report. Over the last six months, Superintelligent has built out an entire agent planning suite. We help you move from discovery to planning to implementation. In addition to the agent readiness audit, after you've completed your agent readiness audits, we help you double-click on your most important use cases with what we call our use case planning reports.

11:08These reports are going to help you understand what sort of technical preparation you need to do to be ready for a use case, what challenges you might face in implementation, and whether you should be thinking about building, buying, partnering, or some combination. After that, you can even get a spec document in what we call our technical blueprint that gives either your developers or the developers of the partner you work with what they need to build exactly the agent that you're looking for. If you want to learn more about Superintelligence Agent Planning Suite, we've built a custom GPT to answer your questions.

11:38Just go to bit.ly slash super agent. That's B-I-T dot L-Y slash super agent, all one word. And if you have any questions, the agent can even help you book an appointment with our team. Welcome back to the AI Daily Brief. Today, we are building off of our coverage from the AI action plan last week. Obviously, a lot of that plan focused on China. And subsequently, China released their own plan. And so today, we are looking into the follow-up and asking more broadly, is cooperation or anything less than outright competition even possible? Now, by way of reminder, last week we got the White House's AI action plan.

12:17As described by former VC and action plan leader Sriram Krishnan, the three core themes of the document were to accelerate AI innovation, build American AI infrastructure, and lead an international AI diplomacy and security. There was a lot that I found very interesting about this report. Most specifically the fact that in many ways, it kind of runs counter to the general approach to foreign policy, at least the stated approach to foreign policy of this White House. Politica's Daniela Cheslow, who is actually an old classmate of mine from Northwestern, captured this pretty succinctly when she wrote, Is AI the exception to America first?

12:52She writes, The AI action plan that President Donald Trump rolled out last week contained quite an Easter egg for globalists. She says that while in many ways the tone and posture was in line with Trump's previous foreign policy, that in point of fact, or as she wrote, packed into the AI plan's pages, was a pivot, a call to forge an enduring global alliance on the technology. She continues, the plan gets pretty specific. It calls for the state and commerce departments to leverage the U.S. position in international bodies. It names the United Nations, the OECD, G7, G20, and the International Telecommunication Union, among others, to advocate standards and governance approaches that reflect American values.

13:29So wait, did AI just carve out its own exception to America first? And what does that mean for America's increasingly fragile relationship with Western allies? Now another place where I noticed this was in the section about open source and open weight AI. The action plan writes, we need to ensure America has leading open models founded on American values. Open source and open weight models could become global standards in some areas of business and in academic research worldwide. For that reason, they have geostrategic value, while the decision of whether and how to release an open or closed model is fundamentally up to the developer, the federal government should create a supportive environment for open models.

14:05Put another way, the White House is here declaring that open-source AI models are a tool of diplomacy. Now, as I mentioned in my initial discussion of the AI action plan, this has the potential to run at odds with how some people have thought about open-source models, particularly vis-a-vis China. If you were looking at the hardline stance that many were taking in 2023 when it came to China, there was a concern that releasing open-source or open-weight models would allow China to catch up. Now, of course, subsequent to that, China has basically caught up. They've been nipping at our heels just a few months behind, and when it comes to open models, they have screamed ahead.

14:41Point being that now when we are discussing the potential of open-source models and open-weight models as a geopolitical tool, it's in the context of China leading that particular part of the AI race. Overall, the response to the AI action plan was pretty divided. Or rather, it mapped pretty exactly to the groups that were receiving it. The tech community by and large received it incredibly well. Box's Aaron Levy writes, America's AI action plan is quite strong. It has a clear mission to win the AI race and accelerate the development and use of AI by removing roadblocks or aiding adoption. Importantly, it focuses on the positive benefits of AI, which we're all seeing every day.

15:18On the flip side, New York Times reporting largely focused on what wasn't there, specifically questions of copyright. Professor Kevin Bryan writes, Come on, man, there's no way you can read that AI action plan and think that this is the first order thing New York Times readers should know about it. AI has huge implications for the national economy, defense, future of science, but half this article is about copyright. No, it wasn't all divisive. Shaquille Hashim, editor at Transformer, writes, Most coverage of the AI action plan has been very negative. In Transformer, I make the case that it's actually not bad.

15:48Shaquille writes, almost everyone I've spoken to this week has expressed some pleasant surprise with what the White House put together. Brad Carson of Americans for Responsible Innovation told me it was cautiously promising. Michael Kleinman of the Future of Life Institute said certain aspects were a, quote, step in the right direction. Brendan Steinhauser of the Alliance for Secure AI said he was pretty happy with a lot of it. White House AI czar David Sachs even pointed out the surprise endorsement of the Washington Post, who wrote an opinion piece, Trump is off to a good start with an AI action plan.

16:15So this was the context coming into China's release of their own version of this plan. The policy was published as part of the World AI Conference in Shanghai last weekend. The event is one of the most prominent tech conferences on the Chinese calendar. Chinese Premier Li Cheng gave a keynote address at the event, stating, Currently, key resources and capabilities are concentrated in a few countries and a few enterprises. If we engage in technological monopoly, controls, and restrictions, AI will become an exclusive game for a small number of countries and enterprises. He made the pledge that China is, quote, willing to share our development experience and technological products to help countries around the world, especially those in the global south, strengthen their capacity building, and bring the benefit of AI to the world.

16:56The central pillar of China's AI action plan is establishing the World AI Cooperation Organization. The international group is envisioned to deal with issues of AI governance, regulation, and deployment. Think United Nations for AI only headquartered in Shanghai instead of New York. And really, the rest of the action plan is similarly focused on global consensus. It calls on the world to follow through with existing UN agreements about digital technology, stating, Only by working together can we fully tap the potential of artificial intelligence while ensuring the security, reliability, controllability, and fairness of its development.

17:27The document features the word cooperation 13 times across 13 key priorities, just to give you a sense of the theme. The entire plan is also steeped in the language of AI safety and calls for global coordination on issues ranging from copyright to AI misuse. George Chen, partner at the Asia Group and co-chair of their digital practice, said, The two camps are now being formed. China clearly wants to stick to the multilateral approach while the U.S. wants to build its own camp, very much targeting the rise of China in the field of AI. I actually think that this is not necessarily correct. I think that China very much wants a unilateral approach, just with itself at the center of a multinational coalition all using China's technology.

18:05In fact, I think the Chinese plan has overtones of a digital Belt and Road initiative. The strategy is quite clearly to offer Chinese AI broadly and cheaply across the developing world. And this, of course, starts to inform how we can think about the indications that the White House might be thinking differently about this particular set of policies. Now, interestingly, in the wake of this, even around the very small amount of cooperation the U.S. is offering China on AI, there are some serious concerns. One of the big announcements from last week was that export controls on NVIDIA's H20 chips had been removed, and the chips could once again flood into China.

18:39The administration had basically started to adopt Jensen Huang's view that China is going to have AI data centers one way or another, so it's in the US's interest to have them using NVIDIA rather than Huawei chips. And demand seems absolutely off the scales. Reuters sources said that NVIDIA had placed orders for 300 ,000 units with TSMC to deal with demand in excess of their existing stockpile of around 600 ,000 to 700 ,000 chips. NVIDIA sold around a million H20 chips last year, so they appear to be gearing up to do at least that much business again as quickly as possible. And yet, Jensen's take is not the only one that's going around Washington right now.

19:12A group of 20 national security experts and former government officials wrote to Commerce Secretary Howard Lutnick on Monday urging the administration to reverse this decision. They called the decision a strategic misstep that will have detrimental effects on America's AI edge in both civilian and military applications. The letter stated, The H-20 is a potent accelerator of China's frontier AI capabilities, not an outdated AI chip. Designed specifically to work around export control thresholds, the H-20 is optimized for inference, the process responsible for the dramatic capabilities gains made by the latest generation of frontier AI reasoning models.

19:44For inference tasks, the H-20 outperforms even the H-100, an AI chip this administration has restricted access to due to its advanced capabilities. The letter also suggested that AI chips are currently a zero-sum game, with a surge in H20 demand necessarily making U.S. chip shortages worse. It argued, The decision to ban H20 exports earlier this year was the right one. We ask you to stand by that principle and continue blocking the sale of advanced AI chips to China as America works to maintain its technological edge. This is not a question of trade. It's a question of national security. It's a fairly diverse group of signatories, from Brad Carson who leads Americans for Responsible Innovation, to former Assistant Secretary of Homeland Security, Stuart Baker, to perpetual China antagonist and financial manager, Kyle Bass.

20:28Now, trying to reflect on where this all suggests we are when it comes to this U.S.-China battle, I think that in some ways, it's a more fluid conversation now than it has been in the last couple of years. There are very clearly conflicting pushes and pulls, not only in the general political environment, but I would hazard a guess, even within White House officials themselves. There is an inherent tension between the foreign policy of withdrawal from the world and engaging full-throatedly in the global competition for AI that's laid out in this AI action plan. And like I said, I'm not sure that this represents a full 180 or a shift in the US's position as much as a growing recognition that the landscape in which this competition is happening has changed.

21:09Foreign Affairs recently ran a piece called China's Overlooked AI Strategy that's all about how the release of these open models, like DeepSeek and others, is a soft power strategy to help Beijing gain global dominance when it comes to AI infrastructure. There's also just the very different disposition China has towards AI from a societal standpoint. The MIT Technology Review, for example, recently ran a piece called Chinese Universities Want Students to Use More AI, Not Less. Unlike the West, where universities are still agonizing over how students use AI in their work, top universities in China are going all in.

21:41The Washington Times recently wrote, quote, jarring moment, China's artificial intelligence gains bewilder top American researchers. And I think what's most notable here is that this is a publication with a very Washington Beltway type of audience. One thing that so far hasn't really been up for question is the assumption that China and the U.S. are in a great power struggle for supremacy in AI. And yet even that is now more up for grabs than it has been in the past. Law professor Peter Salib writes, The White House's AI action plan has some good stuff, but it begins the US is in a race to achieve global dominance in AI.

22:16Like many, Simon Goldstein and I think that an AI arms race with China is a mistake. Our new paper lays out a novel game theoretical approach to avoiding the race. The paper is called Collaboration at the Brink, International Law for the AI Arms Race. In the excerpt of the paper, they write, The centerpiece of our proposal is the formation of a joint AI lab that would combine the best US and Chinese AI talent, supercharged by US and Chinese national investment. We argue that, compared to either an AI race or a nonproliferation equilibrium, the joint lab would be both a safer and a faster route to AI development.

22:46They say that, in fact, AI safety advocates and AI accelerationists should endorse the joint lab approach. Now, I think this might be an interesting Long Read Sunday type of piece at some point, but the point for our purposes here is that this type of thing is starting to get into the ether. When Samuel Hammond agreed with Anthropic's call for the administration to maintain export controls on the H20 chip, Based Beth Jezos wrote, It's not clear-cut. Maybe we should host a discussion on GPU protectionism. He continued, Open-source models are a loss leader for AI labs to achieve market penetration.