Here's the thing about the AI buildout: the people writing the checks and the people pouring the concrete aren't the same bet.

This week handed us three clean signals on OpenAI, the biggest check-writer in the whole story. It's still losing money by the billions. It's stuck in a price war while cheaper models undercut it. And it just pushed its long-awaited IPO out to 2027.

So the funder is getting weaker. The buildout, though, is already contracted and paid for. That split is the entire trade.

Let me give you one name to own and one to leave alone as it plays out.

ONE TO LOVE: GE Vernova (GEV)

The simplest way to play the "contractually committed buildout with a weakening funder" thesis is to own the supplier whose order book is locked and whose customers can't substitute.

That's GE Vernova (GEV).

Gas turbines for data center power are effectively sold out through 2028 and pushing into 2029. Those orders are signed. The pricing is locked.

And the demand curve doesn't care whether OpenAI hits its revenue target or pushes its IPO to 2027. The hyperscalers (the giant cloud operators like Amazon, Microsoft, and Google) have signed take-or-pay deals, which means they pay for the power whether they use it or not. The power gets built either way.

What makes GEV the cleanest pick in the basket:

Backlog visibility runs multi-year (the orders are booked years out). You're not buying a story. You're buying a delivery schedule.

Pricing power is real and rising. GE Vernova is booking new turbine orders at prices well above its existing backlog, and all three credible suppliers in the world are sold out through 2028. Substitution isn't an option for a 2027 data center build.

Margin profile is improving as legacy contracts roll off and new contracts come on at current pricing.

And the extra upside (optionality) from grid, transmission, and onshore wind sits on top of that gas turbine core. That's upside the market is still under-pricing.

What to watch: any management commentary on the order-book conversion rate (how fast signed orders turn into delivered revenue), and any sign of hyperscalers trying to renegotiate those take-or-pay terms.

This is "buy scarcity, avoid abundance" in its purest form.

ONE TO LEAVE: Oracle (ORCL)

The Oracle story is the cleanest example I've seen this cycle of stock investors getting paid for somebody else's credit risk.

Oracle (ORCL) co-founder Larry Ellison's net worth jumped by roughly 100 billion dollars in a single afternoon last September on the Stargate disclosure (the giant data center buildout he's funding alongside OpenAI and SoftBank). The market has treated the stock like a hyperscaler peer ever since.

The problem isn't the rich valuation. It's the contracts underneath it.

Three things to understand:

First, the OpenAI commitment is roughly 300 billion dollars in compute over five years. That single contract is the difference between Oracle as the slow-growth database company it was in 2023 and Oracle as the AI infrastructure story it is today. Take that contract away and the whole re-rating (the market's decision to value it like a fast-growing AI name) goes with it.

Second, Oracle is carrying the financing on its own balance sheet. It's raising debt to build the data centers that OpenAI has contracted to fill. That means Oracle is taking on counterparty risk (the risk the other side can't pay) to a private company that lost roughly 38 billion dollars last year, is locked in a price war with Anthropic, is watching customers test Chinese models that cost a fraction as much, and just postponed the IPO that was supposed to fund the next leg of the buildout.

The credit rating agencies have already started asking questions. The bond market is starting to price it. Watch the spreads (the extra yield investors demand to hold Oracle's bonds, which widens when they get nervous).

This isn't hypothetical. Oracle's credit-default-swap spreads (the cost to insure its debt) blew out to levels we hadn't seen since 2009, and at least one big bank has already moved its debt to underweight on downgrade risk.

Third, Oracle has the weakest defense if OpenAI ever renegotiates. Microsoft (MSFT) has its own platform and revenue lines. Nvidia (NVDA) sells to hundreds of other customers.

CoreWeave (CRWV) has a fleet it could in theory redeploy. Oracle has Larry's narrative and a contract.

The market hasn't yet reconciled the Oracle story with the OpenAI delay. That reconciliation is coming.

Approach: if you're long, this is a name to trim into strength (sell some while it's still rising). The chart will give you opportunities. Use them.

For the LEAPS-minded (LEAPS are long-dated options, a year or more out), the asymmetry favors put structures (bets that pay off when the stock falls) over straight shorts. Momentum can punish you for weeks before the math wins. Members get the specific structure on the trade desk.

What to watch: Oracle's credit spreads, any management commentary on customer concentration (how much of its cloud revenue rides on a single client), and any disclosure on the take-or-pay terms of the OpenAI commitment. Silence on that last one is its own answer.

So there's your split: own the company that already has the orders, fade the company that's lending its balance sheet to a customer that might not show up.

Buy the scarcity. Be careful with the abundance. Same trade. I'll be making the case for live on Monday.

Because Monday morning, right after the 10 a.m. State of the Market, the whole crew gets together for the Ticker Highlight Show.

We each put up the names we love and the names we'd leave, then take the charts apart on the best of them.

GEV and ORCL are exactly the kind of split we'll be hashing out, and I'll be in the room with the team to defend my picks.

Come hang out.

It's the most fun we have all week, and you'll walk away with a watch list.

Here for a good time AND a long time,

Hans