Elon Musk Podcast

OpenAI alerts parents to violent teen threats

11 min
Jul 20, 2026about 1 month ago
Listen to Episode
Summary

The episode explores Meta's potential $10 billion data center lease to Anthropic and Anthropic's $35 billion compute deal with xAI, analyzing how AI companies are forced to monetize infrastructure to offset massive capital expenditures and how these logistics agreements directly impact user experience through rate limits and feature availability.

Insights
  • AI companies are pivoting from pure software businesses to infrastructure landlords out of financial necessity, not strategic choice, driven by unsustainable $100B+ annual hardware spending
  • Compute capacity is the true bottleneck limiting AI advancement, not chip availability—power grid connections and facility buildout create multi-year delays that force companies to lease rather than build
  • User-facing features like rate limits and usage caps are direct functions of backend hardware contracts, not software design choices—more infrastructure leases directly translate to fewer restrictions for end users
  • Competitors are willing to arm each other with computing resources because financial survival (offsetting depreciation and operating costs) outweighs competitive advantage in the near term
  • Anthropic's pre-IPO timing of the xAI deal was strategically critical to lock in pricing and capacity before public market shareholders demand profit margins over growth
Trends
AI infrastructure becoming a utility business model with recurring revenue streams replacing volatile ad-dependent business modelsCompute leasing emerging as a $50B+ annual market with multi-year contracts becoming standard for AI companiesPhysical infrastructure constraints (power, cooling, real estate) becoming the primary bottleneck for AI capability advancement, not algorithmic innovationVertical integration of compute resources becoming critical competitive moat as companies secure long-term capacity before market saturationUser experience and product capabilities becoming directly tied to backend infrastructure procurement rather than software engineeringDepreciation curves on specialized silicon creating urgency to monetize idle capacity within months of acquisitionPre-IPO strategic contracting becoming standard practice to lock in favorable terms before public market pressures alter deal structuresCross-competitor infrastructure sharing becoming normalized despite direct market competition due to financial gravity
Companies
Meta
In early talks to lease data centers to Anthropic for up to $10B over two years; considering infrastructure rental as...
Anthropic
Negotiating $10B lease with Meta and $35B deal with xAI to secure compute capacity; spending $15B annually on hardwar...
xAI
Signed $35B three-year compute agreement with Anthropic; going public and locking in infrastructure contracts pre-IPO
OpenAI
Referenced in episode title regarding violent teen threat alerts to parents; context not elaborated in transcript
People
Mark Zuckerberg
Mentioned fielding weekly requests to buy data center space and considering hardware leasing as viable business option
Elon Musk
Referenced as owner of xAI, which signed $35B compute deal with Anthropic; podcast show name references him
Quotes
"You don't usually see two companies racing for the same exact market share agree to share their core logistics."
HostEarly in episode
"If your capital expenditures outpace your ability to generate revenue, you won't have the resources to compete in the long run anyway."
HostMid-episode
"You are entirely willing to pay a premium to guarantee their engineering teams will always have the physical hardware they need."
HostMid-episode
"The functionality of the entire software industry is currently subordinate to physical logistics."
HostLate episode
"You aren't constrained by the code. You are constrained by the physical hardware contracts these companies negotiate in the background."
HostLate episode
Full Transcript
Meta is in early talks to lease some of its data centers to Anthropic in a deal worth up to $10 billion over two years. Which you look at that and you have two organizations that are direct competitors. They're both actively trying to build the most capable artificial intelligence models available. And they are sitting across the table negotiating a hardware lease. There's just an inherent tension there. You don't usually see two companies racing for the same exact market share agree to share their core logistics. Right. So why would a company that generates its revenue primarily through targeted advertising hand over the keys to its custom built computing infrastructure to Arrival? Well, providing raw computing resources to external clients, I mean, that'd be an entirely new operational lane for Meta. They make the vast majority of their money selling ad space on screens. That is a software optimization problem. You keep users engaged with the feed. You serve them targeted content. You collect revenue. It's, you know, highly scalable and relatively low friction. Yeah. But renting out physical server racks is heavy industry. You are acquiring massive tracts of real estate. You're negotiating with local governments for dedicated power substations, building out liquid cooling infrastructure. It's a completely different business model. You're moving from selling digital attention to managing physical supply chains. Exactly. And managing service level agreements for external clients who need like 99.9% uptime requires a totally different corporate infrastructure than optimizing an ad algorithm. Because if a social feed goes down for 10 minutes, people just complain on another app. But if a client's training cluster goes down for 10 minutes, it can corrupt a multimillion dollar training run. The stakes of the infrastructure are physical and immediate. Mark Zuckerberg actually brought this up directly on an earnings call. He told investors he fields requests to buy data center space almost every week, so he considers leasing out this hardware a viable option. Which functionally turns Meta into a cloud services provider. That opens up a multi-billion dollar business line that operates independently of ad clicks. Because when your balance sheet is strictly tied to advertising, you are entirely exposed to macroeconomic cycles. Right, and compute leasing is a long-term contract business. The entities renting these servers need guarantees they will be available for years. That provides Meta with a recurring, predictable cash flow insulated from the volatility of the digital marketing sector. You trade the high margins of software for the stability of a utility provider. The catalyst driving this shift, though, is the upfront price tag. Meta expects to spend between billion and billion a single year on these data centers That is an astonishing amount of capital to convert into physical infrastructure And well it makes investors very nervous They don have an infinite tolerance for capital expenditure that doesn immediately generate revenue Think of it like building a manufacturing plant so large you're forced to rent out the spare conveyor belts just to cover your baseline utility bills. You construct this massive facility because your own internal production lines demand an enormous footprint. but the operating costs are punishing. Leaving any section of that building idle actively damages your balance sheet. Right, but I have to ask, isn't it dangerous to provide the actual computing fuel for Anthropic, a company that is actively trying to beat you in the market? It sounds incredibly dangerous to arm a direct competitor. 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. So like I was saying, you are handing them the exact logistical tools they need to outpace your own research. It does seem counterintuitive, sure. But financial gravity forces their hand here. The need to offset $145 billion in hardware costs simply outweighs the rivalry. Because financial sustainability takes priority over guarding the hardware. Right. If your capital expenditures outpace your ability to generate revenue, you won't have the resources to compete in the long run anyway. And idle time is uniquely punishing with this specific hardware. The depreciation curve on these processors is just brutal. Oh, absolutely. If you have thousands of chips sitting in a standby state because your internal engineering teams are prepping data or adjusting their algorithms, you are still paying to power and cool that facility. Right. And more importantly, those chips are losing value every single day the next generation of silicon gets closer to release. An idle server isn't just pausing. It's a depreciating liability. Exactly. So taking Anthropix money transforms these data centers from a pure cash strain into a revenue generating asset. You could try to starve a rival by hoarding your servers, but if the cost of hoarding those servers destroys your own balance sheet, well, you've lost the race before the models are even finished training. Which brings us to Anthropix's side of the equation. I mean, they have an appetite for compute resources that is difficult to fully quantify, and they just don't have the internal infrastructure that Meta has spent the last decade building. The volume of processing power required to build these systems isn't just about storing data or running simple database queries. The training process requires continuous synchronized calculations across tens of thousands of chips simultaneously. And that can take months at a time without interruption Yeah And if the compute pipeline stutters or if the networking hardware connecting those chips drops a single packet the entire training run can be compromised So they need uninterrupted high access to massive clusters of specialized silicon that are physically wired together in the same facility. You really see how aggressive they are about securing those exact resources when you look at their corporate maneuvering? This potential $10 billion deal with Meta isn't an isolated strategy. Right. They recently signed an enormous agreement with Elon Musk's XAI. A deal reportedly worth $35 billion over a three-year period. Just to put that number in perspective, Ampropic is committing $15 billion a year purely to rent computing hardware from an external provider. That is an astronomical figure for a hardware lease. Yeah. It really makes you wonder why a company with $45 billion to spend over three years doesn't just buy their own servers. Or build their own facilities. But buying the hardware isn't the actual bottleneck. The waitlist for the chips from manufacturers is months long, sure. But the waitlist to get a dedicated gigawatt power connection from a local utility grid is years long. Exactly. So renting from XAI means you get access to a cluster that is already plugged into the grid, already cooled, and fully networked. You're buying time just as much as you're buying compute. Because without the hardware, their software engineering is completely terrorized. They can hire the best researchers on the planet, but if they don't have the servers to run the calculations, They can't advance the product. The intellectual property is entirely dependent on heavy industry logistics. And what makes that XAI agreement particularly interesting is the timing of the negotiation. That contract was inked just ahead of XAI's initial public offering. Yeah, locking in a contract of that scale right before a company goes public is a highly strategic financial move. It guarantees Anthropic priority access to those servers before public market forces can step in and change the rules. Once a company goes public, the new shareholders demand strict profit margins and rigorous quarterly growth. By signing the deal beforehand, Anthropic secured their capacity and locked in their pricing structure before those new public market expectations could alter the terms. They aren't waiting for the market to stabilize or for hardware prices to optimize. they are signing massive multi-year contracts across the entire industry to ensure their compute supply doesn't dry up. They are entirely willing to pay a premium to guarantee their engineering teams will always have the physical hardware they need. So what do these multi-billion dollar backroom logistics deals actually mean for you, the person sitting at your desk using these tools? It's easy to look at a $45 billion hardware lease and view it strictly as abstract corporate movering but it dictates the exact parameters of the actual product you interact with on your screen Just look at the immediate aftermath of that XAI agreement As soon as that deal was finalized Anthropic raised the rate limits for subscribers of Cloud Code. The moment they secured more computing power on the back end, they allowed their users to interact with their system more frequently on the front end. Those infrastructure leases literally define how much a user is permitted to do. When you type a prompt into an interface, A physical server in one of these data centers has to process that text, load the model's parameters into memory, and calculate the response. And that takes a measurable amount of physical computing power. Compute is a finite resource. If too many users submit queries simultaneously, the servers hit capacity. The system runs out of memory bandwidth, trying to generate tokens for everyone at the same time. That physical limitation is the exact reason companies enforce arbitrary caps on how many messages you can send. When you get a notification saying you have reached your usage limit and need to wait an hour, that isn't a software mechanism designed to annoy you. No, it is a literal load management system. It exists to prevent the physical servers from crashing under the weight of too many concurrent requests. It's essentially a queuing system for physical hardware. If Anthropic successfully secures this $10 billion deal with Meta, it directly alters your user experience. Securing more hardware means fewer restrictions and higher usage limits for you when you log in. They are purchasing the physical capacity required to allow their customers to do more work without hitting a wall. It entirely redefines the boundaries of the software. You aren't constrained by the code. You are constrained by the physical hardware contracts these companies negotiate in the background. The more server racks they can rent, the more freedom you have to utilize the tool continuously without being temporarily locked out of your account. The functionality of the entire software industry is currently subordinate to physical logistics. We tend to conceptualize these tools as weightless digital entities, but they're entirely dependent on heavy industry. The software cannot function without cooling towers, millions of gallons of water, specialized silicon and dedicated power plants. Yeah, it's a completely different reality when you strip away the interface. So the companies building the most advanced software in the world are realizing they might make just as much money acting as landlords for the hardware. Which leaves you wondering, if a company like Anthropic is spending $45 billion here and $10 billion there just to rent the computers, what does that mean for the ultimate price of an AI subscription down the line? 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. Also check out our YouTube channel for more business and tech updates. There's a link in the description.