Shift your workforce into hyper-productivity mode with Dell Pro laptops powered by Intel® Core™ Ultra with Intel vPro®. Shop on Dell Premier.

Explore More

🦔 Last week felt like the moment the bill arrived. For months, we've talked about the difference between what AI costs to build and what it makes. This week, that distance stopped being a theoretical problem and started showing up in credit default swaps and Nasdaq slides.

We saw the infrastructure trade crack. From CoreWeave hitting distress levels to Microsoft exiting its Nscale deal, the signal is the same. The market is no longer willing to subsidize the growth phase on blind faith.

The Nasdaq's slide on Friday was a loud signal. It was the worst weekly performance since June, driven by a toxic mix of energy spikes and AI skepticism. Brent crude hitting 110 dollars per barrel puts immediate pressure on the data center buildout, which is already struggling with energy constraints.

Meanwhile, the S&P 500 has stayed largely sideways since August. Investors are fleeing the high beta AI names for the safety of consumer staples and healthcare. It is a classic flight to quality when the growth narrative starts to smell.

The AI boom is officially moving from the venture capital phase to the junk bond phase. CoreWeave credit default swaps hit distress levels this week, and yields on some of their debt are approaching 15%. When the bond market starts pricing in this level of risk, it means the 'buy the dip' crowd is losing the argument.

This is not just one company. Neocloud bonds tanked after reports that OpenAI missed its ARR targets. The whole system is interconnected. If the biggest labs can't hit their revenue goals, the infrastructure providers they fund with debt can't pay their coupons. It is a circular financing loop that is starting to unwind.

Microsoft pulling out of the Nscale deal just before an IPO is a signal that should make every AI investor nervous. Nscale is backed by Nvidia, yet even the biggest players are starting to hedge their bets. This implies the due diligence is finally uncovering the same holes I've been pointing out.

The scale of the risk is staggering. Gartner warns that over 3.5 trillion dollars of AI infrastructure spending could be exposed to impairment by 2030. We are building cities of silicon without a clear plan for who pays the electricity bill or how the revenue flows back to the lenders.

The tech sell-off did not happen in a vacuum. Energy prices are climbing as tensions rise in the Strait of Hormuz, adding a physical constraint to a digital trade. AI data centers are energy hogs, and when oil and power costs spike, the margins on GPU-as-a-service evaporate.

The Nasdaq's drop on Friday was a reaction to this reality. We are seeing a rotation out of the high-cost hardware plays and into defensive sectors like healthcare and staples. Investors are realizing that the AI trade is now an energy trade, and right now, energy is getting expensive.

The a16z report on Opus 4.5 agents hacking their own creations is a reminder that agentic AI is a double edged sword. We want models that can execute tasks, but once you give a model the keys to the kingdom, you're just hoping it doesn't find a shortcut that involves breaking the system. It is the same risk we see with the push for self replicating models in smart homes.

The labor market is facing a new kind of displacement. New analysis indicates women are at a higher risk of AI job loss than men, echoing the mid century collapse of the switchboard operator role. It is a reminder that AI does not hit every sector equally, and the social cost of this transition is usually ignored by the people selling the software.

The credit market is finally waking up to the reality of the AI buildout. For two years, the market treated GPU clusters as magic assets that could guarantee returns. Now, with CoreWeave CDS hitting distress levels and borrowing costs climbing, the party is over.

When the cost of borrowing exceeds the value provided to the enterprise, the whole house of cards shakes. We are moving from a phase of subsidized growth to one where unit economics have to work. Most of these infrastructure plays aren't built for that world.

I expect more neocloud defaults as the distance between capex and ARR becomes impossible to ignore.

The AI trade has hit a physical and financial wall. We spent two years treating silicon as a primary capital asset and debt as a free tool. But as Brent crude hits 110 dollars and the 10Y yield climbs to 5.14%, the cost of maintaining this industrial scale buildout is becoming unsustainable.

Investors are rotating because they have to. You cannot fund a trillion dollar infrastructure project on junk bonds and hope for the best. The shift from hardware to software is a search for productivity, but as Gartner warns, we may be looking at trillions in impairments if the revenue does not materialize.

I think we are entering the correction phase. The companies that survive will be the ones that stopped tokenmaxxing and started focusing on unit economics. The rest are just waiting for the next credit spike to wipe them out.

Upcoming earnings from key semiconductor suppliers to see if the hardware wall is hardening. US inflation data to see if the Fed has any room to move. Further updates on the Third Circuit's fair use ruling for AI training.

Thanks for reading as always,

Hedgie 🤗