Microsoft's $30 Billion Round Trip

Programming note: We've just released our new thematic report: Enterprise AI Monetization: 2025 Scorecard. It's a 24-page analysis deconstructing how software giants are winning the AI revenue race—and who is falling behind. Access a free copy here.

The Infinite Cash Glitch

If you loan a friend ten dollars to buy a cup of lemonade from you, you can technically report ten dollars in revenue. The cash has returned to your pocket, and you now also have a ten-dollar IOU on your books. On paper, you look twice as rich as you did five minutes ago. The economy has grown. Everyone is happy. The only problem arises if you actually need new cash to pay your rent, and your friend has no ability to ever pay back the loan.

Ray Dalio, the founder of Bridgewater and a man who thinks about economic cycles the way mechanics think about gear ratios, went on television last week to explain bubbles. His thesis is elegant and boring: bubbles don't pop because valuations get too high, or because the technology is fake. They pop because of a "need for cash."

Here is Dalio:

Bubbles don't happen because of good estimates of what's in the future. It happens because of the need for cash. Do you sell that asset? Do you have to sell that asset for cash for some reason? ... Wealth can't be spent. In order to get money, you have to sell wealth.

The argument is that you can have a trillion dollars of Nvidia stock ("wealth"), but you cannot use it to pay your taxes or service your debt until you sell it. The "prick" of the bubble comes when liquidity dries up, interest rates rise, or a wealth tax hits, and everyone rushes to turn their "wealth" into "cash" at the exact same moment.

But in the AI sector, the industry has found a clever workaround to delay this reckoning. They aren't waiting for the customer to find money for the lemonade; the vendors are simply handing them the cash to buy it.

We are seeing the industrialization of "Circular Financing." The big tech companies are effectively printing revenue not by lending money, but by buying equity in the startups that buy their products. It is vendor financing disguised as venture capital.

Look at the headlines from just last week. Microsoft and Nvidia are preparing to invest up to $15 billion into Anthropic (with Microsoft committing up to $5 billion). That sounds like cash going out the door. But wait: Anthropic has simultaneously committed to spending $30 billion on Microsoft's Azure compute.

So the flow of funds looks roughly like this:

1. Microsoft wires $5 billion to Anthropic.
2. Microsoft books the equity investment as an asset on its balance sheet.
3. Anthropic wires that money back to pay for Azure credits (and promises to find another $25 billion to pay for the rest).
4. To fulfill that compute demand, Microsoft presumably takes the money and wires it to Nvidia to buy more GPUs.
5. And who else invested in this Anthropic round? Nvidia. Which suggests that Nvidia is likely getting its own investment dollars back as revenue via Microsoft's hardware purchases.

The money leaves the building, takes a scenic tour of a startup's balance sheet, and arrives back home as revenue growth. It is a perpetual motion machine of capital, but with a multiplier. Microsoft is effectively using a $5 billion down payment to secure a $30 billion revenue stream from a company that it partially owns, while Nvidia funds the startup that creates the demand for the chips that Microsoft likely has to buy from Nvidia. Everyone records massive growth, but the same dollars are just doing laps around the track.

SoftBank, a company that has never met a capital-intensive hype cycle it didn't like, is executing a similar maneuver. Having sold its Nvidia stake to bankroll OpenAI, SoftBank is now reportedly investing $3 billion to overhaul a factory in Ohio to build data centers... for OpenAI.

Andrew Ross Sorkin asked Dalio about this specific dynamic—referencing the circular transactions where Nvidia takes a stake in a company like CoreWeave, which then commits to buying Nvidia chips. Dalio dismissed it as "an issue" but not "the main issue."

You can perhaps squint at that dismissal. The "main issue," according to Dalio, is whether the stock is in "strong hands" (insiders/institutional players) or "weak hands" (the leveraged public).

But is it? The twist is that circular financing embeds leverage into the system. Two things can be true: the AI revolution is real, and the financing supporting it may resemble a hall of mirrors. It creates an illusion of demand that isn't exactly backed by organic end-user cash flow. As Reuters recently reported, there is growing anxiety that the "AI boom has outrun fundamentals," with business leaders noting that these "circular deals... add to the bubble risk." Michael Burry is betting against it. The market is punishing Meta for spending capex without clear ROI.

The financial engineering is impressive, but it runs into the laws of physics.

You can financialize the deal, but you can't financialize the silicon. While dollars loop endlessly in the cloud, the physical supply chain is seizing up. Standard server-memory fulfillment has already crashed to ~35% for some buyers. Nvidia’s fix? Raid the smartphone supply chain for LPDDR5X memory originally meant for phones and tablets. Counterpoint now forecasts this cannibalization will double memory prices by 2026.

In the end, financial loops move fast; atoms don't. Silicon Valley appears to have built a sleek capital machine, but energy, wafers, and chips remain finite. If liquidity tightens, as Dalio warns, we'll see if the hands holding AI wealth are truly strong.

More on Circular Investing:

• Nvidia CEO hits back at AI bubble, circular investing fears: 'We see something very different' (Yahoo Finance)
• Here's why concerns about an AI bubble are bigger than ever (NPR)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

Lenovo's Stockpiling Strategy

• The Headline: PC giant Lenovo is stockpiling memory and other components, holding 50% more inventory than usual, to navigate a supply crunch and rising prices caused by the AI hardware boom. (Bloomberg)
• ARPU's Take: This is a clear case of an industry leader using its massive scale to turn a market-wide crisis into a competitive advantage. Lenovo isn't just reacting to a shortage; it is making a deliberate, aggressive bet to build a fortress of components, effectively weaponizing its supply chain against smaller, less capitalized rivals.
• The Operations Question: The AI boom is creating significant collateral damage in adjacent hardware markets, turning basic components like memory into scarce assets. For PC makers like Dell and HP, Lenovo's move forces a painful strategic re-evaluation of their supply chain philosophy. This shifts the definition of operational excellence from lean "just-in-time" efficiency to resilient "just-in-case" preparedness, creating an imperative to decide how they will compete when their largest rival has already cornered the market on critical supply.

The Low-Tech Chip Chokepoint

• The Headline: A geopolitical dispute over Dutch chipmaker Nexperia led to a Chinese-led export halt of critical, low-tech automotive chips, causing production disruptions for major carmakers and exposing a deep vulnerability in the industry's supply chain. (Reuters)
• ARPU's Take: The auto industry fixed the wrong problem after the last chip crisis. While it focused on securing high-end processors, this crisis proves that geopolitical risk extends to the most mundane, fractions-of-a-penny components. China's ability to weaponize these seemingly insignificant parts created a global chokepoint, revealing a massive blind spot in the industry's risk assessment.
• The Operations Question: This crisis expands the definition of supply chain risk beyond natural disasters to include geopolitics, even for low-cost parts. For supply chain leaders, this creates an imperative to shift from "just-in-time" inventory management to building costly but necessary "just-in-case" buffers for any component sourced from a geopolitically sensitive region, fundamentally challenging decades of cost-optimization practices.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.