Dec. 14 at 6:13 PM
Sharing my latest thoughts on the AI trade -- long-form, structural, and risk-focused.
TLDR: The AI Trade’s Hidden Fragility: Circular Capital, Power, and the Illusion of Certainty
The debate around AI has become oddly binary, dividing critics into those who are accused of "not getting it" and those who are lumped in with people calling for a dot-com style collapse. Both miss the point.
The real risk in today’s AI trade isn’t whether the technology works. It clearly does. The risk is structural -- embedded in how capital is being deployed, financed, recycled, and justified across a remarkably narrow group of companies, all operating under assumptions that require near-perfect execution.
What we’re watching isn’t just an AI boom. It’s a circular financing ecosystem, and those systems tend to look strongest right before their fragility is exposed.
Circular capital is not the same as diversified demand
At the center of this cycle sits OpenAI -- not as a public company, but as a gravitational force pulling capital, compute, power, and infrastructure commitments from nearly every major hyperscaler. Microsoft (
$MSFT ) funds OpenAI. OpenAI drives demand for compute. That compute is supplied by Nvidia (
$NVDA) hardware, financed and hosted by companies like Oracle, built on power-intensive data centers, often backed by long-dated contracts with… the same hyperscalers.
At the same time, many of these firms are investing directly in each other, forming partnerships with competitors, and co-funding infrastructure that assumes sustained, exponential demand growth across the entire AI stack.
This isn’t proof of strength. It’s correlation.
When capital, counterparties, and assumptions all concentrate in the same handful of balance sheets, the system becomes less resilient -- not more. The market treats "blue-chip counterparties" as a substitute for risk analysis, but strong counterparties don’t eliminate risk; they synchronize it.
Why Oracle matters and why credit noticed first
This is why the Oracle (
$ORCL) discussion matters far more than people want to admit.
The concern isn’t whether Oracle can meet its obligations today. That’s a distraction. Credit markets don’t price certainty; they price duration, leverage, execution risk, and sensitivity to change.
Oracle is leaning into a capital-intensive buildout tied to long-dated assumptions about utilization, pricing power, and customer behavior. Those assumptions may prove correct -- but they also require a clean runway: no delays, no renegotiations, no policy shocks, no demand hiccups.
Credit spreads tend to widen not when something breaks, but when the margin for error disappears. That’s what people are reacting to -- not imminent default, but the recognition that this trade has become unforgiving.
You don’t need a collapse to get damage. You just need repricing.
The macro backdrop is not neutral anymore
What makes this cycle more fragile than many appreciate is that it’s unfolding against a backdrop that is anything but stable.
We’re heading into a politically charged period with rising scrutiny over trade policy, industrial policy, and government involvement in public companies -- from semiconductors (
$INTC ) to defense to energy infrastructure. Tariff uncertainty isn’t going away; it’s morphing. If courts limit one tool, administrations will reach for others.
At the same time, the U.S. is selectively allowing advanced chips to be sold into China, while China responds with approval gates, domestic steering, and quiet retaliation. This isn’t de-escalation. It’s a more bureaucratic phase of the same contest.
Layer on Taiwan risk, geopolitical fragmentation, and the reality that China will not sit still while the U.S. dominates advanced compute, and the idea of a frictionless global AI rollout starts to look naïve.
Monetization remains the least examined assumption
For all the focus on infrastructure and demand, the end-market economics remain strangely under-discussed.
On the consumer side, AI is already being bundled, subsidized, or given away to protect ecosystems. Google (
$GOOGL ) offering Gemini broadly for free isn’t generosity -- it’s competitive defense. That dynamic pushes consumer pricing toward zero, not toward sustained margin expansion.
On the enterprise side, AI may drive productivity, but productivity gains don’t automatically translate into proportional revenue growth for vendors. Cost savings accrue unevenly. Pricing power erodes faster than most models assume. And when everyone offers "AI-enabled" solutions, differentiation compresses.
This doesn’t mean AI won’t be monetized. It means the payback period is longer, messier, and more contested than the market is currently discounting.
The longer-term risk people aren’t ready to price
Beyond earnings models and capex plans lies a risk few want to touch: social and political backlash.
AI is not just another software upgrade. If it meaningfully displaces jobs -- and it likely will -- public pressure will follow. That pressure will find its way into regulation, permitting delays, labor protections, and local resistance to massive data-center projects that strain power grids, water supplies, and utility costs.
Communities don’t need to understand AI to oppose it. They just need to feel threatened by it.
That friction will slow timelines, raise costs, and challenge the assumption that this infrastructure can be built and scaled without resistance.
Power, Permits, and the Physical Constraint Beneath the AI Trade
There is also a more basic constraint that remains widely underappreciated: power. At the expansion rates implied by today’s AI investment assumptions, the electrical grid is not positioned to deliver the required capacity at scale. Expanding generation, transmission, and interconnect infrastructure is not a software problem; it is a physical, regulatory, and political one.
This has already introduced a new layer of speculation around energy, particularly nuclear, whether through the reconstitution of long-shuttered plants or the development of small-scale and modular reactors. But, in my view, none of this happens quickly. Permitting can take years, construction spans political cycles, and local or environmental resistance can halt projects before ground is ever broken.
Moreover, a change in administration could easily reset priorities and delay progress. The implication is straightforward: without power and permits, even the most advanced AI models encounter a hard ceiling. When core growth assumptions depend on infrastructure that does not yet exist, uncertainty ceases to be abstract -- it becomes financial risk.
Market complacency as an accelerant
This longer-term tension is unfolding inside a market that has grown deeply complacent.
A large portion of today’s retail participation has never experienced a prolonged, grinding drawdown -- only sharp selloffs followed by rapid recoveries. Volatility has been trained out of behavior. Risk has been conditioned away.
At the same time, leverage has become normalized. Margin, leveraged ETFs, options, and synthetic exposure are no longer viewed as amplifiers of risk, but as standard tools for participation. I’ve been around long enough to know: when leverage rises alongside narrow leadership and near-universal optimism, downside doesn’t unwind gently. It gaps.
Crypto fits squarely into this same mindset.
We’ve watched public companies incorporate bitcoin into their balance sheets and see equity prices surge on announcement alone. We’ve seen businesses effectively operate as crypto reserve vehicles while remaining publicly traded equities. And we’ve already lived through multiple crypto drawdowns of 50–75%, yet the belief persists that institutional adoption somehow neutralizes risk or guarantees upside.
That psychology matters because it reinforces a broader narrative: exposure itself becomes the strategy, and volatility is reframed as opportunity rather than risk.
This doesn’t end in zero, but it doesn’t end cleanly either
None of this argues that AI is a fraud or that the entire trade implodes tomorrow. That’s a straw man.
What it does argue is that the current setup -- circular capital flows, concentrated exposure, long-duration assumptions, leverage, and narratives priced for perfection -- leaves very little room for disappointment.
Markets don’t break because everyone is wrong. They break because everyone believes the same thing for the same reasons.
And right now, too much of the AI trade rests on the belief that scale eliminates risk, that partnerships guarantee returns, and that future demand will neatly validate today’s spending.
History suggests otherwise.
Happy holidays. Wishing everyone a healthy and successful New Year.
