A Memo from Our CEO: On Sandberg and the Machine Economy.

$295,000,000. $195,000,000. These are the 2025 annual revenues from Google and Meta, respectively, from advertising. One half a trillion dollars exchanged. This essay discusses where that money comes from, where that half-trillion is going, and what’s necessary to get it there. 

Since 2005, these two advertising firms have earned more than $3 trillion,  or about $150 billion annually, with a growth rate of approximately 20% year over year.  Google’s annual revenue from advertising exceeds all revenue from Microsoft. The combined revenue from these advertising firms exceeds all but thirty countries. All based on the input of human attention. 

The mastheads of these firms show Brin, Page, and Zuckerberg. The engine of this revenue is Sheryl Sandberg. 

Sandberg is the single most important businessperson of the 21st century to date (I’d reserve Xi Jinping as the most important person, economic or otherwise, of this century). In 2001, Sandberg took Stanford R&D from Brin and Page and turned it into the largest recurring cash machine ever created. Incredibly, in 2008, she did it again at Meta under Zuckerberg. This cash machine funded the internet we know and use today, largely for free; maps, texts, email, file storage, and now large language models. Our digital life from 2002-2025 has a single origin, and that’s Sandberg. 

Adwords existed before Sandberg, as did its auction mechanism, scoring, and ranking. Zuckerberg and his team built the product, graph, feed. These mechanisms, however, were not creating durable economic moats and cash flow at either firm prior to Sandberg. She invented this cash flow engine and the factory that supports it, and built it in a manner that has led these firms to trillion dollar valuations over a decade following her work. 

This essay is about who she is, what she built, how she built it, and the economy and culture that followed. It’s also a thesis that we at Radius are building our entire company around. That the Sandberg digital world, and its economy, is sunsetting. What’s emerging is better for machines, and for people. And we hope to be a critical part of that new economy. 

The Sandberg economy, defined.

Take away the product and technology and the Sandberg economy can be defined simply as the offering of services for free in exchange for the capture, and sale, of a user's attention. 

This is the brilliance, and simplicity, of Sandberg’s system. Search, social, text, email, storage, maps, large language models. All for free. The users of these systems provide the inventory; their attention (which, while vast, is not infinite). The customer is the advertiser. The product is a specific point-in-time of a human’s attention, auctioned a quarter of a second at a time. Everything else Google and Meta do as advertising firms serves that quarter second. A digital Fast and the Furious if you will. 

The internet, as most humans overwhelmingly experience it, is through Sandberg’s tollgate. A search, query, or, in the case of mobile social media, a scroll, pause, or click, reveals what you may want, and what you may be induced to want with the right advertising. A toll on the act of wanting.. 

Sandberg’s Design

A critical feature of the Sandberg economy is the auction, and the self serving nature of that auction. At the launch of her career at Google, advertising sales remained limited by human connectivity. Salesman, rate cards, and contracts. Executed across AdWords and AdSense, Sandberg removed the human from the sales side of the equation. Any advertiser needing the necessary information on potential customers could self-service their account, and bid against the world. Importantly, Sandberg made two critical decisions in this design. First, she let the auction be the determinant of the price, not Google. Every quarter-second the market determined the price of your attention, and the quality score system fine tuned that model over decades. Second, due to the removal of humans from the loop, the auction system linearly scaled, like the internet upon which it was built. 

At Meta, Sandberg made one critical change. Rather than focusing on the query, or the user’s stated intent, Sandberg built the auction around the implied intent. Sandberg’s and Meta’s precognition system guessed where you wanted to go, and built your experience around that path, and sold access to that pathway to advertisers. 

Sandberg’s Operation

At Google, a query triggers the auction. Advertisers bid a cost per click. Click through rate times bid sets the rank. The advertiser cares only about return on ad spend. More queries, more auctions, higher bids, higher prices, more revenue for Google, on loop for a quarter century. Funding Gmail, Maps, Drive, and now Gemini. 

At Meta, the impression triggers the auction. The feed scrolls ad infinitum (or based on your phone’s battery life), sells ads by “cost-per-mile” or CPM, targeted by the social graph, refined by pixel that determines whether the advertisement led to sale. More of your time on Meta product, more impressions, more data for the graph, improved targeting, higher prices, more revenue. 

Sandberg’s Economics 

Sandberg’s system has near-zero marginal cost, pricing power, and almost zero capital investment. Due to the near infinite supply of queries and impressions, there is a near infinite supply of inventory, none of which Google or Meta needs to carry. It just vanishes if not purchased. Gross margins at Meta regularly exceeded 80% over the last decade. Google and Meta created scale, data, and distribution that made their incumbency almost impenetrable. That fortress earned them hundreds of billions a year in cash. In short, the Sandberg system is likely the single most profitable business model ever created, importantly because the actors in the system, namely the users, have a near 0 cost of supplying the inventory to the advertiser. 

Sandberg’s Impact

The revenues rolled in for 25 years. While valuations changed, the cash was real. The auction proceeds funded a newly expansive “free” internet, maps, entertainment, social media, communication, large language models. This free internet also leveled the world. Meta employees used the same version of WhatsApp as the farmer in Nepal. A critical benefit of Sandberg’s system was the democratization of the digital world. 

There’s also a dark side to this system. As Charlie Munger often reminded us, “show me the incentives, and I’ll show you the outcome.” If queries and impressions were the source of revenue for these firms, then it’s in their interest to maximize the quantity of those queries and impressions through increased volume by the user. User volume is a function of user time on those platforms. Human time, unlike the internet’s scale, is definitively finite. These firms under Sandberg learned how to increasingly optimize for engagement, and learned how to reward the user with dopamine (the bounce at the bottom of an instagram reel or the placement of rage inducing content on a news feed). Later, ecosystem participants learned to do the same, with encouragement from these platforms (think of your X feed bot %s). The machine, built by these firms, with both perfect efficiency, knowledge of our finiteness, and also completely 0 accountability, learns to capture that finiteness to sell us a version of the world we can’t look away from - that may not even be real. The same model that created maps and whatsapp just made you miss your child’s soccer goal. In short, human’s finite time on this earth is the product.

The Transition

Every dominant incumbent looks permanent until its resources stop being scarce. The entire model of the Sandberg economy relies on our finiteness. If clicks, scrolls, queries and impressions become infinite, the system breaks. Large language models, in part funded by these impression based cash flows, may lead to their destruction. 

A critical problem with the Sandberg economy is that it fails to distinguish value from traffic. Impressions based on bot based false-reality are poor signal. Value, providing a user something they need that improves their condition, is a depreciating asset. Once you find the perfect chocolate chip cookie recipe (and it's unquestionably this one  https://www.seriouseats.com/the-food-lab-best-chocolate-chip-cookie-recipe) your scrolling journey is over, because you’re in the kitchen making the dough. But what if you’re shown an impression that claims certain races, ethnicities, or religions make better cookies, or worse? Or aligns a certain type of cookie recipe with derogatory stereotypes?  And maybe wrap that content in a viral 18 second video? Now you’re getting impressions and beating the street’s quarterly revenue expectations. The path to profitability is arising from the seven deadly sins in the hearts and minds of a user with a full battery and a red stoplight. And profitable they have been. 

Large language models break this model. LLMs crawl content directly, send it to the user, who never sees the underlying page or scroll. The web is being scraped by near infinite resources that aren’t limited by time, sleep, diet, battery life, and green stoplights. None of whom pay the tollbooth. 

Importantly, as these LLMs get smarter, they learn to avoid the low-value content. Like cheap motels on Route 66 after the creation of the Interstate, these rage-baiting and low value content providers get left behind. 

The user is no longer the human with the finite lifespan. The user is now the machine. Agents shopping have no attention to sell, no bias to inflame, no sexual desires to arise. The entire foundation of the Sandberg economy, that a human’s attention can be captured and auctioned, fails when the human leaves the loop. You cannot run an attention business when the customer has infinite attention. 

The Machine Economy (or the Prince Economy)

If this has been the Sandberg quarter century, what’s next? It's clear that machines are already the primary actor in the digital world. Impressions by machines for the first time exceeded those of humans this week (as measured by Cloudflare). Machines will be the driver of digital interactions, and machine-to-machine interactions will be the default engagement model.

The leading voice of this new economy is Cloudflare’s Matthew Prince. Prince is a likely candidate as he, and Cloudflare, sit at the pain point between the old internet economy and the new. Cloudflare acts as a protective tollgate between websites and their users. The volumes of user requests to those websites have dramatically increased since the launch of large language models, and Cloudflare sees every uptick. 

Above we defined Sandberg’s model as  the offering of services for free in exchange for the capture, and sale, of the attention it captures. Prince’s model can be summarized as the direct charging of services to machines who operate on the internet for the content they consume. Prince’s unit is the request, rather than the query or the impression. Where Sandberg removed the humans from the sales side of the translation, Prince removes both the advertiser and the human from the user side of the transaction. Importantly for Cloudflare shareholders, the toll, and its half-trillion in annual revenue growing at 20% annually, remains. It simply moves from the advertiser to the machine. 

Replacing the auction is the pay-per-crawl. When a machine (I use machine over agent or bot, as they carry normative impressions) requests content, it pays, at whatever cost required by the content provider, in real time. Prince references Spotify as the model. Creators don’t get paid if they don’t get listened to. And agents aren’t clicking on rage baiting bot farms like your elderly parents. 

The standards are being assembled. X402 (https://www.x402.org/) built largely by Erik Reppel (@programmer) and the team from Coinbase Developer Platform, addresses the long running HTTP 402 “payment required” error from the early internet (Marc Andreesen was the first to popularize this in an August 2019 A16z podcast https://a16z.com/podcast/a16z-podcast-from-the-internets-past-to-the-future-of-crypto/). Reppel’s x402 finally heals that original sin, if paired with the right infrastructure. 

Prince’s economy turns Sandberg’s on its head. Rather than human attention as the market, content becomes the market. Humans continue to get content for free, as long as the machines are willing to pay for it. The value goes up, the noise goes down. We’re baking cookies rather than fighting each other. 

Importantly for Cloudflare and other actors moving into this space, the business model is just as attractive as Sandberg’s, maybe more so. The marginal cost remains near zero. The scale remains internet-scale. And the toll gates will continue to exert that leverage on the users with sufficient distribution. Cloudflare, sitting in front of ~25% of the internet, including approximately 80% of the LLM companies, is best positioned to take this toll. 

What’s missing, and where Radius fits

Prince makes one critical caveat to shift to the machine economy that Cloudflare is so well positioned to ignite. There’s no rail to pay machines with sufficient speed and cost. 

Cloudflare sees 500 million requests per second. Prince estimates that 1-10% (5-50 million) of those requests are monetizable, at whatever cost demanded by the content providers (likely fractions of a cent). Even at the lower bound, there’s no payment network that reaches 5 million transactions per second. The highest Cloudflare has seen is 2 million, and that network is Radius. The goal for Cloudflare, and the machine economy prepared to take over, is a linearly scalable network, growing with the demand put on the system by its machine users. It can’t be Visa or blockchains as currently constructed. Visa is too expensive, most blockchains are too burdened by their consensus and distribution models. 

Advocates for those systems claim that the 5-100 million transactions per second isn’t necessary. Financial systems have netted and batched for over half a century. Accounts can be pre-funded. We can just settle later. At 500 million requests per second, this logic breaks. Throughput isn’t solved by batching, netting, and prefunding, it relocates that transaction burden to new intermediaries, who themselves will demand a toll, increasing the price of the transaction, and the speed to which it ultimately finalizes. Wallets held by machines will have to be refilled due to the settlement lag, again clogging the system not with content transactions, but inter-account funding. What will occur is a transactional logjam that kills this economy before it starts. The TPS requirement named by Cloudflare isn’t ignorance of financial history, it's the price of keeping the machine economy moving and scaling. 

Radius is the network that will enable this machine economy. We are the only stablecoin network known to go well over 3M TPS. Our costs have been measured to be low enough for machine volume micropayments, up to 10M tx/USD. We’re not yet at 10M, but the entire focus of our team is on that number, and proving scalability to Prince’s requirements. Great engineers want the hardest problems, and Cloudflare’s is the most important problem set before the crypto community since Satoshi solved the byzantine general’s problem. 

So how does Radius accomplish this? Our thesis is that micropayments at extreme scale (into the 9-10 figures) must dispense with decentralization theatre. Token governance  models break, as their complexity causes additional weight on distributed systems. Just get the speed, and the new economy will emerge. No tokens, no decentralization. Just a better internet economy.

In this economy, Cloudflare stands best to win as the tollgate. X402 appears to be the protocol/handshake that will win. Stablecoins will be the medium of exchange, and while USDT and USDC dominate, there are thousands in the queue to compete. And Radius is the only rail with the speed to move the compensation for the content.  

Sandberg’s model was about finding the tollgate where her firms could maximize their revenue at the highest possible margin, with the lowest possible inputs (human time and attention). That tollgate is moving, from user-based ad auctions to machine based pay per crawl. Prince has called his shot. At Radius, we’re exclusively focused on building the rail that makes it all work. We look forward to enabling the machine economy for the next 25 years, and for the next half-trillion in annual revenue up for grabs.