Let’s start with Accenture, the Schrödinger’s stock. Accenture just had one of the best quarters of its corporate life. Record bookings—$22.1 billion in Q2 alone. Forty-one clients with more than $100 million in quarterly bookings, also a record. One hundred and four large deals of $100 million or more year to date, up 13$%. Q3 revenue of $18.7 billion, up 6%. Free cash flow of $3.6 billion per quarter. Bing, bang, boom. They have 700,000 employees. And now they have partnerships with OpenAI and Anthropic. By any traditional metric, this is a company at the peak of its operational power.
The stock is down more than 50% this year.
Wall Street is selling Accenture for two reasons, and here’s where it gets genuinely confounding. Reason one: DOGE gutted federal contracts. Accenture Federal Services represented about 8% of global revenue and roughly 16% of Americas revenue. Thanks Elon! That’s gone now, or going. Painful, but manageable. But that’s not what’s driving a 50% haircut.
Reason two is that investors think agentic AI is going to make Accenture obsolete. That all the consulting work—the digital transformation projects, the enterprise software implementations, the IT managed services—is going to be automated away. The machines are coming for the people who deploy the machines.
Here’s what kills me. Wall Street is simultaneously betting that Accenture will successfully deploy AI for its enterprise clients—which is why those record bookings exist—and betting that this success will make Accenture itself worthless. The glass isn’t half full or half empty. It’s overflowing or bone dry.
If Accenture succeeds, it automates its own service delivery and shrinks the business. If it fails, it proves the whole thing doesn’t work, crashing the broader AI thesis. Either way, the stock gets punished, earnings be damned.
Then there’s crypto. A guy named Michael Saylor built a religion. Strategy Inc. (formerly MicroStrategy) reoriented its entire corporate identity around one non-negotiable conviction: buy Bitcoin, hold Bitcoin, never sell Bitcoin, and the rest will sort itself out. Maximalism as business strategy. Audacious, ridiculous, and for a while spectacularly effective.
Now he’s thinking about selling.
The company holds $51 billion in Bitcoin. But the blended book value of its holdings has dropped below parity. The financing advantage that made the whole scheme work—issuing debt and equity at a premium over the underlying asset, using the spread to buy more Bitcoin—has evaporated. Preferred stock is trading below its $100 par value. The company is now reportedly contemplating selling up to $1.25 billion in Bitcoin to preserve liquidity. Doing the thing that the company was built around never doing. Makes sense.
But there’s a wrinkle, and it’s probably why Saylor can’t believe this is even happening. In any previous bull market environment—stock market near all-time highs, Wall Street awash in cash, institutional money looking for yield—this is exactly the moment when speculative assets catch fire. FOMO drives crypto. That’s the pattern. That is not what’s happening here.
Instead of flowing into speculative assets, the money is flowing into AI and, apparently, space. Bitcoin’s FOMO moment should be right now, and it isn’t arriving. In other words, Wall Street has a new speculative darling. I’ll give you a hint. It’s South African by origin, hates democracy, is known to give Nazi salutes when it’s really excited and just went public with a company built on lies. Did you guess?
That’s right! SpaceX went public. $75 billion IPO—the largest in history. But while all the attention was on the stock price, 11 days after the IPO Elon did another thing. SpaceX issued $25 billion in senior unsecured notes—originally targeting $20 billion but upsized because demand came in at $89 billion. Brand new debt to mostly repay the $20 billion bridge loan SpaceX took out when it acquired xAI. And just like that, Elon paid for Twitter.
Now don’t think for one second the other tech giants are just going to sit back and let Elon hoover up all that debt and equity from Wall Street. No, no, no. Meta, Alphabet, Amazon and Microsoft have all gotten in on the bond market action. Yes, the most cash-generative companies in human history are going to the bond market to fund another round of eye-popping infrastructure investments. Collectively, they are planning to spend approximately $725 billion in capital expenditures through 2026.
All told, the AI-related investment-grade bond issuance this year will run in the neighborhood of $140 billion. And they’re going to need every penny of it, because analysts project that the five primary hyperscalers will add nearly $2 trillion in AI-related assets to their balance sheets by 2030.
You’d think someone, or everyone, would be asking a fairly simple question: If AI is going to generate the returns everyone claims, why are the most profitable companies on earth borrowing money to fund it?
Let’s work through it because there are some logical reasoning problems to tackle.
First: AI is not making companies more productive or profitable. A recent Reuters investigation found that businesses are actively switching to cheaper AI alternatives because the bills are eating them alive. There’s a phenomenon now called “tokenmaxxing”—treating AI consumption as a proxy for productivity, with predictable results. Gartner projects that AI coding costs will surpass the average developer’s salary by 2028, and three-quarters of executives report rising tech budgets, with nearly half projecting double-digit jumps. An AI officer quoted in the investigation said that open-source models are “90% as good at 10% of the price.” So the premium product isn’t ten times better. It’s just ten times more expensive.
Second: The data centers will be obsolete before they’re finished. Construction timelines for large data centers run two to three years. AI hardware and architecture are iterating faster than that. TechTarget has reported that data center designs could be obsolete by the time construction begins. Electricity demand from data centers is projected to double by 2030. AI-ready capacity is growing at a 33% compound annual growth rate—meaning whatever you build today is underpowered before the paint dries. The capital-intensive buildout phase doesn’t happen once. It happens again and again, every cycle.
Third: Energy is a hard ceiling, not a soft constraint. The AI buildout requires electricity on a scale that does not currently exist. The rest of the world is switching to renewables to manage costs and hedge supply chains. The current administration has effectively frozen large-scale domestic renewable development. While also managing the largest oil price disruption in modern history through tariff chaos and geopolitical destabilization. High energy costs are now structurally embedded in the American economy. This is not a short-term problem. It is a permanent ceiling on AI expansion in the United States.
Fourth: China is winning, and we’ve decided that means we can’t talk about it. The four most popular AI models on OpenRouter—which is the most widely used AI model routing platform—are all Chinese. DeepSeek is number one. Chinese models are priced at as little as 18¢ per million tokens, against an average of $4 for the top American models. China’s AI capabilities, which analysts once estimated were more than a year behind U.S. capabilities, are now roughly four months behind, and closing. If China decides to truly open its AI market globally, it doesn’t eat our lunch. It eats the whole cafeteria.
Circulating the Cash or Circling the Bowl?
Money is circling through Wall Street at velocity, with fees extracted at every rotation. Bond issuances. IPO commissions. Refinancing charges. Bridge loans. Secondary offerings. New debt issued to retire old debt. Infrastructure built to be replaced. Billions deployed into technology that isn’t yet delivering measurable productivity gains, priced at multiples that assume it eventually will, powered by an energy grid that structurally cannot support the buildout, and increasingly outcompeted by a country we’ve decided to treat as an existential enemy rather than a cautionary market signal.
The whole thing is a toilet bowl in slow motion. It looks like it’s moving with purpose. There’s a lot of momentum. But the direction of travel is fixed.
When Andrew Ross Sorkin gets his next book deal he’ll be writing the definitive updated account of this crash. The thesis will likely be that all of these signs were visible and readable, in plain text, in SEC filings, in earnings calls and Reuters investigations and Gartner reports. That nobody missed them through ignorance. They looked past them because the fees were too good.