Big Tech Sued for Coordinated AI Slowdowns

0:00When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications, and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsor job credit at Indeed.com slash podcast. That's Indeed.com slash podcast, terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs. Imagine if this stability piece and the innovation piece were each other's missing piece. For Citizens Bank and Silicon Valley Bank are combining financial security and innovation expertise to create a full suite of solutions for all your business needs. With the scale to back every stage of growth, from pre-C to IPO and beyond, you have everything you need to make your big ideas click into place. For Citizens Bank and Silicon Valley Bank, it's the best of bank worlds.

1:15Anthropic, open AI, SpaceX AI, and Google have been hit with a federal antitrust lawsuit accusing them of an illegal coordinated effort to intentionally decelerate artificial intelligence development. That completely flips the standard tech narrative backward. For years, the baseline assumption was just that we were watching this cutthroat race force supremacy. Right. The expectation was that these companies would just burn through capital as fast as physics and hardware supply chains allowed. Exactly. Constantly trying to outbuild each other. But now, the most powerful tech labs in the world are literally being accused of forming a cartel to pump the brakes. Which really brings up the core issue that everyone is looking at right now. Right. If these dominant players are coordinating a slowdown, is it actually to protect humanity from existential risk? Or is it just a calculated move to freeze the market, lockout competitors, and artificially inflate prices? Well, the chronological origin point alleged in the lawsuit really anchors this whole situation. The coordination between these rival tech labs supposedly began two months before Anthropic

2:21CEO Dario Amadeh published a highly visible essay. That's the one that explicitly called for a global slowdown to quote, pace the frontier. Right. Yeah, he didn't just suggest taking a breath. He outlined a framework for collectively slowing the rate at which these models gain new capabilities. And within hours of an M.O.D.s call for industry-wide coordination, you have Sam Altman, Elon Musk, Demis Asabis, and Satin Nadella, all publicly voicing their agreement. That speed is just so unusual. You have the leaders of the most heavily funded competitors in a single sector, companies that normally fight bitterly over talent and market share, all spontaneously agreeing to slow down simultaneously. It creates a completely unified front from the very entities driving the technology. If you're not subscribed yet, take a second and hit follow on whatever podcast app you're using. It helps us keep making this. We appreciate you being here. And the lawsuit focuses heavily on the consumer impact of that unified front. It claims this deliberate deceleration actually shrank the value of consumers paid AI subscriptions.

3:25Because the companies took money for a certain trajectory of progress and allegedly agreed to deliver less. Yeah, when you pay a monthly fee for access to a leading model, you are funding an ecosystem with the expectation of continuous rapid capability updates. It looks a lot like traditional price fixing. Like in standard industries, firms collude by limiting supply to keep prices high, right? OPEC restricts oil barrels to maintain the price of crude. Airlines limit flight capacity so they don't have to compete on ticket prices. Here they are allegedly limiting capability growth to keep compute prices, model access prices, and enterprise licensing margins artificially elevated. But lawsuit alleges these companies weren't just discussing philosophy regarding existential threats. They actively discussed seeking government waivers. Wait, government waivers? Yeah, to allow coordination that would otherwise blatantly violate antitrust laws. Oh, so the argument is that the safety conversations might just be acting as a convenient cover. Right. Seeking a waiver indicates an acute awareness that their actions without government protection

4:28look exactly like illegal market manipulation. I mean, antitrust law doesn't require a secret smoke-filled room to prove collusion. Competitors publicly signaling and aligning on market shaping behavior is entirely enough to constitute an illegal agreement. It happens out in the open. Exactly. When executives use public essays, social media posts, and coordinated media appearances to negotiate the terms of a slowdown, regulators just view that as tacit collusion occurring in plain sight. That behavior forms what is basically known as a capability cartel. A capability cartel. Yeah. The top four frontier labs agree to pace development. They are effectively agreeing to substitute collective restraint for individual accountability. So by collectively deciding not to push the boundary, they completely freeze the competitive environment. Right. No one has to risk burning another $50 billion to stay ahead if everyone just agrees to stop running. In a standard cartel, participants agree to set a floor on prices so nobody undercuts

5:29the group. And in a capability cartel, they are agreeing to set a ceiling on product development, which has the exact same effect on the market. It's functionally identical because it prevents disruption. It neutralizes the primary weapon of any challenger. Yeah, which is creating a vastly superior product. By neutralizing that threat, the incumbents ensure they remain the market leaders indefinitely. They can transition from a wildly expensive research and development phase into a pure, commercial harvesting phase. And extracting maximum profit from the models they've already built. Without the constant looming threat of arrival, making their technology obsolete overnight. The transition from R&D into commercial harvesting really requires us to look at exactly how these companies spend money. Training a new Steady Art AI model is an exercise in extreme capital destruction. It's not just paying software engineers. No, not at all. When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored

6:30Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications, and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsor job credit at Indeed.com slash podcast. That's Indeed.com slash podcast, terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs. Imagine if this stability piece and the innovation piece were each other's missing piece. First Citizens Bank and Silicon Valley Bank are combining financial security and innovation expertise to create a full suite of solutions for all your business needs. With the scale to back every stage of growth, from pre-C to IPO and beyond, you have everything you need to make your big ideas click into place. First Citizens Bank and Silicon Valley Bank. It's the best of bank worlds. For a limited time, you can get a big Mac meal for just $8. That's a burger, fries, and a drink.

7:31They don't call it an extra-value meal for nothing. Get a big Mac meal only at McDonald's. Price and participation may vary. Promotion pricing may be lower than meal pricing. We are not talking about hiring a few dozen developers to write code in an office building. We are talking about procuring tens of thousands of specialized processors. Building massive data centers to house them. Right. And securing enough energy to power a small city just to run a single training cluster for six months. The capital expenditures required to push the frontier forward are staggering. So if OpenAI decides they're going to push the frontier and train a next-generation model, the market dictates that Google must immediately match that investment just to maintain parity. And Thropic is forced to match it. SpaceX AI is forced to match it. It becomes a compulsory capital drain for all the major players. They are trapped in a spending cycle dictated by their fiercest competitor. If one runs, they all have to run. And that competitive dynamic is the exact mechanism that antitrust laws are actually designed

8:31to protect. Because that intense rivalry is what produces better, cheaper products for the consumer. When companies are forced to compete aggressively, they can't just hoard their capital. They have to spend it on innovation. But if they can quietly signal to each other, that the current generation of models is sufficiently capable. And they all collectively agree to stop pushing. That compulsory capital drain just vanishes. Exactly. They suddenly save themselves tens of billions of dollars. The profit margins explode the moment the research race stops. Instead of constantly buying new hardware to train the next model, they can simply rent out access to the models they already possess. They harvest the subscriptions, the enterprise licensing fees, and the API usage charges. Without the crushing overhead of building the replacement product, I mean, the financial incentive to form a capability cartel is overwhelming. Which connects directly back to how tastic collusion operates in plain sight. You brought up the airline industry earlier. When airline executives go on earnings calls and publicly state that the industry needs

9:34to show, quote, capacity discipline, they are sending a signal to their competitors. They are suggesting that no one should buy new planes or schedule new routes. Yeah. And if all the competing airlines hear that signal and decide to freeze their capacity, ticket prices stay high, and nobody has to spend money on new aircraft. The consumer loses, but the airlines thrive. The digital equivalent of that capacity discipline is exactly what prosecutors are targeting in this lawsuit. When Dario Amadea writes about pacing in the frontier, and the other CEOs immediately validate that concept on social media and in interviews, the legal argument is that they are negotiating the terms of a ceasefire in public. They are communicating their willingness to cap their own development if the others do the same. But wait, are we sure this isn't just genuine panic? We have all heard the warnings from researchers about the theoretical dangers of artificial general intelligence. Right. The existential risk argument. Yeah. Maybe Dario Amadea legitimately looked at his latest internal model, saw capabilities

10:35that frightened him, and thought the industry needed to stop before something catastrophic happened. Is it fair to immediately assume it is a cynical cash grab rather than a genuine safety concern? Well, that is exactly the defense these companies will mount in court. They will argue that artificial intelligence is uniquely dangerous, categorically different from airline seats or barrels of crude oil. They will position their coordination as a necessary responsible action to prevent disastrous outcomes for humanity. But the lawsuit pre-empses this defense by pointing directly to the discussions about seeking government waivers. The pursuit of an antitrust waiver is the crucial detail here. Because if you are entirely confident that your actions are purely philanthropic. Right. If you genuinely believe you are acting solely to ensure human survival, you do not typically hold quiet discussions with your competitors to figure out how to shield yourselves from federal antitrust liability. The act of seeking the waiver indicates a very clear, sophisticated awareness that their

11:35market behavior was structurally anticonpetitive. They understood that from a regulatory perspective, agreeing to stop making better products is illegal, regardless of the state and motivation. And this leads us to map out the downstream damage of a frozen market. If you start categorizing who actually wins and loses when the frontier stops moving, the list of losers is pretty extensive. The incumbents benefit immensely by securing their market dominance. But startups take the hardest hit. Startups rely almost entirely on the ability to push boundaries. The new company entering this sector needs to offer a capability that the big four do not possess or they need to offer it more efficiently. If the industry standard becomes this artificial ceiling dictated by the incumbents, Vigia Capital stops funding startups aimed at breaking that ceiling. The venture model requires the possibility of absolute disruption. Yeah, if the big players have collectively agreed that disruption is no longer allowed, the entire startup ecosystem loses its oxygen. It functions as an incumbent tax on innovation.

12:37The market expectations are said by the companies that control the most compute and hold the largest market share. If they stop progressing, the demand for competing advanced systems flattens. But it is not just the startups that suffer. Researchers in the open source community lose heavily in a capability cartel. The academic world and the open source community have historically relied on a trickle-down effect, right? Here labs, the ones with the massive budgets, push the boundaries by discovering new training techniques or inventing new neural network architectures. Eventually those techniques filter down. Sometimes the company's published papers detailing their methods and sometimes they release the actual model weights to the public. Hold on, let's clarify that for a moment. When you say they release the model weights, you're talking about the actual mathematical brain of the AI, correct? Exactly. The weights are the core value of the model. When a company trains an AI, it feeds enormous amounts of data through an algorithm. The algorithm adjusts billions of internal numbers, the weights, until it accurately predicts

13:39patterns. And those weights cost millions of dollars in electricity and hardware to calculate. Right. When a company releases an open rate model, they are essentially giving away the finished product of all that expensive computation. Independent researchers can take those weights, install them on their own local machines, and modify them without having to pay for the initial training. So if the frontier labs freeze their development to pace the frontier, that pipeline of knowledge and resources to the academic world dries up entirely. There are no new architectures being discovered, no new papers being published on advanced techniques, and certainly no new open weight models filtering down to independent developers. Open access to frontier capabilities becomes a direct casualty of this coordinated pacing. The knowledge becomes centralized and locked within the cartel. Then you have the direct impact on the consumers, which the lawsuit specifically highlights. If you are listening to this and you're paying $20 or $30 a month for access to a leading AI model, you are the actual victim mentioned in this legal filing.

14:42The value proposition of that subscription is based heavily on the assumption that the model will get progressively smarter, faster, and more capable over time. And the companies agree to pace the frontier. They are effectively degrading the future value of your subscription. You bought into the ecosystem expecting a rocket ship. You expected your $20 to fund the research that would deliver a drastically improved assistant next year. If the company's cap their development, they are essentially giving you a golf cart to save on gas while continuing to charge you the rocket ship premium. You face higher prices for stagnant technology and you have fewer alternatives because the competitive startups have been starved of funding. There is an even broader perspective regarding public accountability that factors into this dynamic. We are talking about technologies that the creators themselves claim could possess society altering power. Decisions surrounding existential risk, global safety standards, and the acceptable pace of technological integration carry weight that impacts society at a fundamental level.

15:44The argument is that decisions of that magnitude must be made by governments that are directly accountable to the public. They should not be made by private firms operating under the intense pressure of profit motives, venture capital valuations, and the desire for absolute market control. Safety is undeniably vital, but it cannot be privatized. We currently have a situation where the exact same companies that dominate the compute infrastructure, dominate model access, and dominate the lobbying efforts in Washington are now attempting to self-regulate the pace of human innovation. When you need to build up your team to handle the growing chaos at work, use Indeed Sponsor Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications, and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsor job credit at nnd.com slash podcast. That's nnd.com slash podcast. Terms and conditions apply.

16:45Need a hiring hero? This is a job for Indeed Sponsor Jobs. Imagine if this stability piece and the innovation piece were each other's missing piece. First Citizens Bank and Silicon Valley Bank are combining financial security and innovation expertise to create a full suite of solutions for all your business needs. With the scale to back every state of growth, from pre-C to IPO and beyond, you have everything you need to make your big ideas click into place. First Citizens Bank and Silicon Valley Bank. It's the best of bank worlds. For a limited time, you can get a big Mac meal for just $8. That's a burger, fries, and a drink. They don't call it an extra value meal for nothing. Get a big Mac meal only at McDonald's. Price and participation may vary. Promotion pricing may be lower than meal pricing. The concentration of power is the structural issue, the lawsuit attacks. In private entities assume the role of global regulators, their commercial incentives inevitably blend with their stated, altruistic goals.

17:46You cannot neatly separate a company's desire to protect its market share, from its desire to protect humanity, especially when the required action for both goals is exactly the same. Slowing down competitors and halting the release of new products. Bringing compute into the equation triggers a realization about the one major player who refused to play along with the slowdown. In Amade, Altman, Musk, and Hassabas were all publicly coordinating their messaging on pacing the frontier. Jensen Huang, the CEO of NVIDIA, conspicuously refused to endorse the slowdown. He stayed entirely out of the agreement. Yeah. That fracture in the industry alignment is incredibly revealing. If you look at the underlying business models, the reason for Huang's refusal becomes obvious. Everyone who endorsed the slowdown sells software models or the cloud infrastructure those models run on. They are the ones paying the massive capital costs to train the AI. Jensen Huang sells the underlying hardware chips that all of those software and cloud companies are forced to buy. NVIDIA is basically the arms dealer in this scenario.

18:48Their position in the ecosystem is unique, and their massive pricing power relies entirely on each lab being terrified that they will fall behind their rival. Google buys clusters of 100,000 NVIDIA GPUs because they are afraid open AI is buying them. And Thropic buys them because they are afraid Google is buying them. It is an economics of pure fear. NVIDIA can charge premium margins because the hardware is the absolute bottleneck for survival in the AI race. If you remove that competitive fear through a coordinated capability cartel, you instantly remove the premium on NVIDIA's chips. The moment the frontier labs agree to stop racing, the demand curve for cutting edge hardware flattens out entirely. NVIDIA requires the race to continue at breakneck speed to justify their hardware margins and to sustain their massive market capitalization. A truth among the labs is catastrophic for the hardware supplier. If open AI and Google decide they do not need to build the next $100 billion supercomputer because they have agreed to pace the frontier, NVIDIA's revenue projections just collapse.

19:50Huang Yu's refusal to participate in the slowdown signaling really highlights the purely commercial undercurrent of the entire debate. It forces us to acknowledge a specific nuance regarding human motivation. Both realities exist simultaneously in this industry. The fear of AI risk is likely very real among the software leaders. Many of them have spent years genuinely warning about the potential dangers of a rogue system. But the commercial interest in freezing the market, saving capital and protecting their margins is equally real. Insisting that this situation must be only one or the other produces incredibly weak analysis. Human beings, and corporate entities by extension, are entirely capable of holding dual motivations. An executive can genuinely believe they are saving the world from an existential threat while simultaneously recognizing that their strategy perfectly ensures their permanent market dominance. They do not have to choose between altruism and greed if the action satisfies both. You see that dual motivation clearly when you expand the scope to the international and

20:52physical requirements of this technology. Dario Amadez's proposal did not just suggest slowing down domestic development. It explicitly contained provisions advocating for strict export controls on China. The framing presented to the public was that to pace the frontier safely, American models and American compute capabilities must be heavily restricted from flowing to geopolitical rivals. The flaw in that logic becomes apparent when you look at how capital and talent actually operate on a global scale. A pact that only binds American firms or relies on domestic export controls does not actually slow the technology down globally. It merely relocates it. Capital is fluid. If you artificially cap development in Silicon Valley, the investment capital simply seeks a different regulatory environment. The engineering talent migrates to jurisdictions that are not bound by the capability cartel. We are seeing sovereign wealth funds in the Middle East aggressively courting AI infrastructure. If the United States decides to enforce a ceiling on capabilities, those foreign funds

21:56will simply say come build here. You end up shifting the center of innovation to other regions without actually mitigating the global risk you claim to be solving. You just lose control of the oversight. Relocating the technology involves staggering amounts of money and physical resources. AI infrastructure investment is projected to reach $31.6 trillion through the year 2050. That figure is difficult to even conceptualize. It covers the physical build out required to sustain this industry. We are talking about endless miles of copper wiring, millions of tons of concrete, advanced cooling systems, and massive networking gear. But the deciding factor for where that $31.6 trillion lands is not just tax incentives or cheap land. The ultimate bottleneck is cheap, reliable, low carbon electricity. Power is the absolute physical constraint of artificial intelligence. You can write the most efficient software imaginable and you can print the most advanced silicon wafers in the world. But these data centers require gigawatts of continuous power.

22:58A single massive training cluster demands the same electrical output as a medium-sized city. And they cannot rely on intermittent power sources like wind or solar without massive baseline support. They need continuous, unbroken power to keep the training run stable. The regions that can supply that level of cheap, reliable electricity dictate where the infrastructure is built. This shifts the balance of power in ways we are only just beginning to see. This highlights an untapped point of leverage that local governments possess. The states and regions that hold the power grids, the physical land, and the water rights for cooling are currently being treated merely as building sites. The tech companies arrive, negotiate heavy tax breaks, plug into the local grid, and treat the local governments as passive hosts. But these states and regions hold massive leverage that they simply are not utilizing. They are active parties in the AI decision making process by virtue of controlling the physical resources the industry requires to exist. If a state governor controls the permitting for a nuclear reactor, or oversees the regulatory

24:00approval for a major hydro facility that a tech giant desperately needs for a new data center, that state has the power to demand serious concessions. They do not have to just accept the terms offered by the tech firms. A local government could mandate specific open access requirements for their local universities. They could enforce their own safety frameworks, or demand regular audits of the models being trained as a strict condition of connecting to the municipal grid. The physical constraints of power and land connect directly back to the corporate constraints these AI giants face. Those corporate constraints outline an impossible situation that the AI giants are currently trapped in. It is the core paradox of their existence. Their governing mandate demands constant relentless acceleration. This mandate is shaped by intense venture capital structures, massive company valuations, and intense competitive pressure. Investors poured tens of billions of dollars into these firms with the specific expectation that they would conquer the global market and achieve unprecedented technological breakthroughs.

25:03You cannot maintain an $80 billion valuation by promising your shareholders that you are going to hold still. The financial structure demands that they keep pushing the frontier. Just that financial mandate, with the safety reality they are running into on the engineering side, these labs are chasing the holy grail of generalized AI. The problem is that pursuing generalization inherently introduces massive uncertainty into the system. Let's clarify what we mean by generalization versus narrow AI. When you build a narrow model, like a system designed purely to predict weather patterns or program built strictly to play chess, you can map the absolute boundaries of its behavior. You have mathematical certainty about what it can and cannot do. A chess program will never suddenly attempt to execute a financial trade. Generalized models by definition operate across unbounded domains. They are designed to process any text, write any code, or analyze any image you give them. Scaling up those generalized models improves their estimation capabilities, but it never

26:04transforms estimation into total certainty. You can throw ten times more compute at a generalized model, and it will become much more articulate and seemingly much more capable, but you still cannot mathematically guarantee it will not produce an unexpected or harmful output in a novel situation. The fundamental architecture of these large language models relies on probabilities, not deterministic rules. When you ask a question, it is calculating the statistical probability of the next correct word based on billions of training examples. It does not possess an internal model of objective truth. It is estimating. And while it gets incredibly good at estimating, the chance of a severe hallucination or a dangerous misinterpretation of a command is baked into the fundamental design. You cannot simply patch out probability. This creates a brutal three-way trap for the major players. First, they cannot safely continue at full speed because the underlying technology becomes increasingly unpredictable and harder to control as it scales up. It's a safety risk is genuine.

27:04Second, they cannot afford to slow down independently. If Google hits the brakes due to safety concerns, while OpenAI keeps accelerating, Google instantly loses its market position and faces a revolt from its shareholders. The financial mechanics punish caution. And third, as this anti-trust lawsuit demonstrates, they face serious legal action if they try to solve the first two problems by slowing down collectively. Anti-trust law actively prohibits them from solving their competitive dilemma through coordinated cooperation. They are caught between the demands of their investors, the limitations of probabilistic technology, and the strictures of federal law. The proposed resolution to this paradox requires a fundamental shift in how the technology is managed and deployed. Safety cannot depend on a fragile, collective promise to simply pump the brakes. A capability cartel is not a durable solution. The safety has to be engineered directly into the authority delegated to the AI before it executes anything in the real world. Meaning that authority means moving away from trying to make the statistical model itself

28:07perfectly safe, which the pursuit of generalization makes impossible. You cannot make a probability engine perfectly reliable. Instead, you build robust deterministic systems around the model. You constrain what the AI is allowed to access and what actions it is permitted to take. You completely decouple the intelligence of the system from its operational authority. To put that into a concrete example, let's say you ask an AI to manage your stock portfolio. If the model generates a plan to sell a specific stock, that plan has to pass through a non-AI deterministic security layer before any transaction is authorized. That security layer is built on traditional, verifiable software engineering. It checks your account balance, it checks the trading limits, and it uses hard-coded rules to approve or deny the action. The AI never actually touches the exchange's API. You rely on traditional software logic to enforce the boundaries rather than hoping this statistical model decides to be well-behaved. If you construct the deployment environment so that the AI operates inside a secure,

29:09limited container, the risk profile drops dramatically. This approach removes the need for an illegal capability cartel because it manages the risk at the deployment level rather than the development level. It forces the entire industry to mature. The underlying assumption that you can just scale a single neural network until it safely and flawlessly manages the entire global economy is giving way to the reality of complex systems engineering. You need multiple layers of traditional security wrapped around these probability engines. The antitrust lawsuit serves as a massive forcing function for this maturity. It strips away the illusion that these powerful companies can self-regulate through informal gentleman's agreements or public signaling. They can no longer rely on a tacit truth to manage their technical debt or their safety concerns. It forces them to actually engineer solutions rather than just agreeing to pause the problem. The era of unchecked AI acceleration is over. Replace by a complex legal reality where antitrust law will dictate the pace of human innovation

30:10just as much as engineers do. If we allow just a handful of dominant firms to decide the pace of innovation, we might not be protecting humanity. We might just be surrendering the future to the only players rich enough to freeze it. If you're not subscribed yet, take a second and hit follow on whatever app you're using. It helps us keep making this. We appreciate you being here. So check out our YouTube channel for more business and tech updates. There's a link in the description. Imagine if having your cake and eating it too was about banking. Not baking. Silicon Valley Bank and First Citizens Bank have created a recipe that blends decades of innovation economy experience with more than a century of stability. Now, you don't have to choose between deep sector expertise and lasting financial security. Instead, you can enjoy them both. Silicon Valley Bank and First Citizens Bank. That's the best of bank worlds.