Roko's Basilisk
Posts
When Cash Isn't Enough

When Cash Isn't Enough

Plus: Poke lands on Apple, Microsoft revamps Notepad, Wall Street rethinks Tesla.

Roko's Basilisk
June 08, 2026

Here’s what’s on our plate today:

🧪 Why Alphabet decided $174B in cash flow wasn't enough.
📰 Poke becomes Apple's first agent, Microsoft's love letter to Notepad, JPMorgan warms to Tesla.
💡 Roko's Pro Tip: When the richest company borrows to build, check your own runway.
🗳️ Poll: What does Alphabet's $85B raise really signal?

Let’s dive in. No floaties needed…

Launch fast. Design beautifully. Build your company's website on Framer.

Framer helps teams design, build, and launch their marketing sites lightning fast. With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster.

Trusted by companies like Miro, Bilt, and Perplexity.

_{*This is sponsored content}

The Laboratory

TL;DR

Cash flow isn’t enough anymore: Alphabet’s 2026 capex guidance of $180B to $190B exceeds its entire operating cash flow, forcing a company worth $4T to raise $84.75B from outside investors.
The whole sector is there, too: Microsoft, Amazon, Meta, and Alphabet plan to spend roughly $700B on capital projects this year, about double last year's total, with Amazon's free cash flow projected to turn negative.
Circular financing masks real demand: Chipmakers invest in AI companies that rent data centers that buy chips back, meaning much of the apparent demand may be the same dollars lapping the track.
The accounting risk: Michael Burry argues hyperscalers are overstating chip lifespans to suppress depreciation, pointing to a potential $176B earnings gap across the industry through 2028.
The stakes: If AI demand matures more slowly than the build-out, this becomes the fiber boom, and the companies that define asset-light software become the cautionary tale.

Why Alphabet decided $174B in cash flow wasn’t enough

Every few generations, people start building the future faster than the future can afford to arrive. In the 1840s, British investors financed railway lines years before there were enough passengers to justify them. A century and a half later, telecom firms buried fiber-optic cable for an internet that had not yet arrived. In both cases, the infrastructure eventually proved indispensable, even as many of the companies that built it suffered enormous losses. The wager never changes: build first and trust that demand will catch up later.

AI infrastructure costs are rising so quickly that even Alphabet’s vast cash generation is no longer enough to fund expansion entirely from within. Photo Credit: Reuters.

For about two decades, the largest technology companies seemed exempt from that pattern. They ran what economists call an asset-light model, making enormous profits without owning much physical stuff, selling search results, advertising, and software that cost almost nothing to copy. However, in 2026, artificial intelligence is quietly dismantling that arrangement, because the machines that run modern AI are expensive, power-hungry, and quick to wear out, and the companies building them have begun to resemble the railway and fiber barons who came before.

That resemblance became impossible to ignore in the first week of June 2026, when Alphabet, the parent company of Google, told the market it would sell $80B in stock to help pay for AI infrastructure, and within two days, investor appetite was strong enough that it lifted the figure to $84.75B. The package combines public share sales, a private investment from Warren Buffett’s Berkshire Hathaway, and other financing mechanisms. An equity raise of this kind was once close to unthinkable for a company this profitable.

The significance of the raise lies less in its size than in what it reveals. Although only about $44.75B is earmarked for AI infrastructure and related spending, the larger question is why Alphabet, one of the richest companies ever created, felt the need to raise external capital at all. For most of the internet era, successful technology firms financed expansion from their own cash flows. Alphabet’s decision suggests that AI may be reshaping that model, as the cost and pace of infrastructure investment begin to exceed even the resources of the industry’s largest players.

Why Alphabet needed outside capital

The puzzle is that Alphabet did not appear to be a company in need of money. It is worth more than $4T, and over the year ending in March, it produced $174B in operating cash flow, the cash a business throws off from day-to-day operations. The trouble lies in what it intends to spend, because Alphabet has guided its 2026 capital expenditure, the money sunk into long-lived assets like data centers and chips, to between $180B and $190B, with a further increase promised for 2027. The arithmetic is blunt: the yearly cost of building has grown larger than all the cash the business generates, before a dollar goes to dividends, buybacks, or taxes.

The balance sheet looks even heavier once $75.6B in future data-center lease obligations are included alongside a debt load that surpassed $100B within roughly a year.

A company that once funded its ambitions out of pocket now reaches for every source of capital at once, and the unsettling part is that Alphabet is the healthy one in this story. If the healthiest balance sheet in the sector is beginning to show strain, the question naturally arises: is this an Alphabet problem or an industry-wide phenomenon? The evidence increasingly points toward the latter.

The same move, across the industry

The same shift is underway across the hyperscalers. Microsoft, Amazon, Meta, and Alphabet together plan to spend roughly $700B on capital projects in 2026, about double the amount spent the year before. For investors who prized these firms for the spare cash they generated, the warning sign is that the spare cash is thinning, with Amazon’s free cash flow, the money left after building and equipment are paid for, projected to turn negative.

What makes the change look permanent is the source of the money. For years, executives assured shareholders that the AI build would be paid for with internal cash, which kept it walled off from the people who lend money, and that assurance has quietly broken down. Oracle raised roughly $18B in a single bond sale, one of the largest on record, and Alphabet has gone so far as to issue a 100-year sterling bond, a loan that does not come due until the next century. The companies that defined the asset-light era are taking on debt like the heavy industries they once rendered obsolete, and the question of who ultimately bears the risk has shifted from shareholders to bondholders and the public.

As capital requirements rise, another question emerges. If the industry’s biggest companies are increasingly borrowing to build AI infrastructure, what exactly is giving lenders and investors the confidence to keep supplying the money? Part of the answer lies in the AI ecosystem’s unusually self-reinforcing structure.

The machine that funds itself

Beneath the borrowing sits an arrangement stranger than either debt or equity, and it explains why the demand looks so dependable. According to reporting by Bloomberg, a web of deals at the center of the AI economy runs on what observers call circular financing, a loop in which a chipmaker invests in an AI company, the AI company spends that money renting data centers, and the data centers buy the chipmaker’s hardware, so the same dollars travel in a circle and can return looking like fresh revenue. The clearest example is NVIDIA’s plan to invest up to $100B in OpenAI, the maker of ChatGPT, which intends to use the money to build data centers filled with NVIDIA chips, sending much of the spending back where it started.

The structure is not new, and the precedent is uncomfortable, since the fiber boom of the late 1990s ran on nearly the same mechanism, with equipment makers lending carriers the money to buy their equipment and carriers trading capacity to make demand look larger than it was, until the traffic failed to appear and the loop snapped. Circular deals can be perfectly rational, letting companies lock in customers and finance huge projects quickly, though they also make it hard to tell how much of the demand is real and how much is the same money lapping the track.

Yet the existence of financing does not resolve the underlying uncertainty; it merely postpones it. Every infrastructure boom eventually confronts the same question: is the demand genuinely large enough to justify the build-out, or has financing itself begun creating the appearance of demand?

Boom, or glut

This is where the sharpest disagreement lives, and it turns on an accounting choice most people never consider. Investor Michael Burry, who profited famously from betting against the 2008 housing bubble, has argued that hyperscalers inflate their profits by assuming AI chips will last five or six years, when their working life is closer to two or three. Stretching that assumed lifespan lowers depreciation, which lifts reported earnings today and tucks a larger replacement bill into the future. Burry put the gap at $176B across the industry between 2026 and 2028 and called the maneuver "one of the more common frauds of the modern era," though the same reporting notes the claim is hard to prove, because firms are granted wide latitude in these estimates.

The debate ultimately revolves around timing rather than technology. Few serious observers doubt that there is a demand for AI. The disagreement is whether demand is arriving quickly enough to justify the pace of investment happening today. That distinction matters because infrastructure cycles rarely fail due to demand never materializing. They fail because demand arrives years later than investors expected.

The opposing case rests on something more tangible than accounting. Alphabet’s cloud backlog of signed but undelivered orders nearly doubled in a single quarter to more than $460B, the optimists’ strongest evidence that real customers are paying real money. Further out, McKinsey has projected that data centers worldwide could require $6.7T in investment by 2030 to keep pace with demand for computing power, a figure that makes sense only if demand proves as durable as its advocates believe.

The railways and the fiber teach the same ambiguous lesson, that a build-out can be a bubble and a foundation at once. The track laid in the 1840s ruined a generation of investors and still carries trains into the next century, and the cable buried in the 1990s bankrupted its diggers and still carries this sentence to the screen. Alphabet, with its $4T valuation and its $174B of annual cash, has concluded that the safest course is to take money from everyone willing to offer it, which is an odd verdict for the richest kind of company in history to reach. Whether it means the demand has grown so vast that even a trillion dollars cannot keep up, or that the spending has outrun the logic still holding it together, is something the depreciation clock will settle on its own schedule, once the chips bought this year have either paid for themselves or quietly expired.

Roko Pro Tip

💡

When even the richest company on earth raises outside capital to build, read it as a signal about your own runway. If your AI strategy assumes cheap, abundant compute forever, model what happens when capital tightens, lock in pricing, and don’t build a business that only works while the money keeps lapping the track.

Be the future of AI

Hunting for AI roles eats hours. Careers pages, group chats, internal referrals, a dozen job boards.

The Athyna AI Job Board does the watching for you. We track openings at frontier AI companies, match them to your profile, and notify you when a role hits a 75% matching index.

Set up a profile and let the matches come to you.