Listen to the audio version of this article (generated by AI).

A week that tested the AI bull… capability pacing vs. capacity racing… the safety promise that could backfire… and the competing priorities no one escapes

As I write on Thursday morning, the stock market is roaring higher – and it’s the tech and AI names, the very ones that took a beating all week, leading the charge.

The Nasdaq is up nearly 1.5%, small caps are joining in, and Treasury yields are easing back from the near-5% levels that rattled everyone this week.

It’s quite the reversal. Just yesterday, the Dow Jones Industrial Average plunged more than 600 points after the Federal Reserve hiked and signaled it isn’t done. Today, the mood has flipped: Investors have decided that a Fed willing to get tough on inflation is actually reassuring, and risk appetite has come rushing back.

There’s no single new headline driving it – no blockbuster earnings, no policy surprise. Just sentiment, swinging hard in the opposite direction less than 24 hours later.

It’s the perfect setup for today’s Digest. It’s a reminder that in the short term, investor sentiment can and will swing the market up and down. But the daily mood isn’t what will decide the next three years for the AI trade. One thing is. And this week’s Digests have been building toward it…

On Monday, we covered the warning about AI from an Anthropic researcher. It resulted in widespread fear, a bipartisan political scramble for regulation, and AI’s own founders calling to slow the pace of frontier development. The market read it as a reason to sell.

In Tuesday’s Digest, we highlighted how that fear collided with a trio of macro headaches – oil back above $100, the 10-year Treasury at 5%, and a Fed poised to hike. But we shared the historical stats on why the first rate hike in years is actually bullish if we look six months or more out.

Yesterday, we put those reassuring stats under a microscope and found the one condition where they fail: a supply-driven hiking cycle. That led us to the bedrock point…

The AI trade – the engine under this entire bull market – rests on one thing: the torrent of spending from a handful of hyperscalers.

Sentiment can cause it to wobble, but it’s earnings from AI infrastructure companies (spend from the hyperscale data center builders) that will ultimately make or break it.

So, today, let’s tackle the obvious related question…

Will the hyperscaler spending keep flowing as AI fears explode and calls for regulation grow louder?

The bull case: capability pacing, capacity racing

On the “yes” side, we’ll go to our technology expert, Luke Lango, editor of Innovation Investor. He’s in an especially good position to make this case because he’s spent this week in Los Angeles at the All-In Summit, a gathering of some of the biggest names in tech and finance with a hand-selected attendance list. Luke heard from Jensen Huang, Elon Musk, and even President Trump, who called in.

For Luke’s first point, he flags the distinction that Wall Street is blowing right past. From Monday’s Innovation Investor Daily Notes:

The immediate debate centers on when a frontier model should be released and who gets to inspect it first.

None of it touches the amount of computing power being purchased, built, or consumed. That distinction is the single most important thing to us.

To understand why, recognize that an AI model burns computing power on two very different jobs. “Training” is the massive, one-time effort that builds a new frontier model. “Inference” is everything after – every answer and task the AI bot performs once it’s live.

Luke notes that pre-training – the stage the entire safety conversation is trying to moderate – has collapsed from more than 60% of frontier computing power in early 2024 to under 10% today. Meanwhile, the workloads that took its place, post-training and inference, are barely touched by any safety procedures being discussed.

So, even a real slowdown in the one thing the safety hawks are targeting would barely dent total AI spending. Inference (use by customers) already makes up roughly two-thirds of AI compute, according to Deloitte, and McKinsey expects it to grow about 35% a year through 2030.

Luke notes that Bloomberg Intelligence landed in the same place this week: With the industry capacity-constrained for years, there’s little reason for anyone to walk away from their commitments now.

Here’s Luke’s overall bottom line:

Capability pacing. Capacity racing.

Model releases may slow. The infrastructure required to train, test, monitor, and run AI keeps growing.

Then there’s his second point – the one I’d underline for anyone rattled by this week’s doom-and-gloom headlines:

Follow what they do, not what they say.

And those actions are hard to argue with.

Back to Luke:

At All-In, I didn’t hear much talk about slowing down.

I paid particular attention to [Microsoft CEO Satya] Nadella. Microsoft is one of the companies writing the biggest checks in the AI Boom.

And the check writer didn’t say anything about writing fewer checks. I didn’t hear anything about cutting AI spending, reducing infrastructure commitments, or backing away from new data centers.

A fresh Bank of America industry report, in Luke’s telling, put it just as plainly – analysts there wrote that they “see no signs of slowing in customer orders, capacity commitments, or semis pricing,” while projecting global semiconductor sales to nearly double to $3.2 trillion by 2030.

This theme was the throughline of the All-In Summit

According to Luke, the conference amounted to a coordinated counterpunch to the recent wave of AI doom.

Beyond Nadella saying nothing about slowing capex, Luke reported that Nvidia Corp.’s (NVDA) Jensen Huang called the human-extinction fears “irresponsible and wrong” and sounded as bullish as Luke’s ever heard him.

Then, President Trump phoned in to vow no regulation and no slowdown, dismissing the whole episode as a Chinese psyop. Finally, Elon Musk pitched a private peer-review system in place of heavy-handed government oversight.

Here’s Luke, summing it all up:

The people who control the capital, the chips, and the policy just told you they aren’t slowing down — and selloffs built on the assumption that they will tend to reverse quickly…

It’s worth waiting through this volatility because we deeply believe that on the other side of this, we will see a massive and sustained rally in AI stocks.

So, is it time to sound the all-clear?

Perhaps, but keep these overhangs in mind

First, Luke is clear that despite his overall bullishness, things could get very bumpy on the road directly ahead:

The short-term setup has gotten tougher… markets trade on fear as well as fundamentals… That could keep pressure on AI stocks over the next few weeks.

But the immediate aside, I still wonder about risks out on the horizon.

First, Luke told us to “follow what they do, not what they say” – that’s reassuring on spending. But if we apply it to “safety,” it creates some issues.

If the labs’ actions tell us they aren’t really slowing down – if the capex keeps racing while the safety talk remains just talk – then the thing that actually frightened people this week isn’t being addressed. And unaddressed risk is exactly what feeds the public and political backlashes.

Regular readers will recognize the shape of this. In our April 6 Digest, I laid out the Prisoner’s Dilemma running through every layer of AI – the competing priorities that leave everyone worse off when each player does what’s individually rational. And the conversation this week looks like another Prisoner’s Dilemma, maybe the biggest of all…

AI leadership versus AI safety.

A frontier lab can’t fully maximize both at once. Lean all the way into leadership – keep spending, keep racing – and safety gets shortchanged, leading to more AI agent hacks and handing ammunition to every politician looking for a cause.

In this case, even if the hyperscalers want to spend, the government can find ways to interrupt those dollars.

But lean all the way into safety – actually slow down – and you validate the capex fear that hammered these stocks this week (not to mention the geopolitical fear of China “winning” the AI race).

This leaves us in what I’ve called “The Messy Middle” – pacing in their rhetoric, but racing in their capital budgets. That middle is bullish for spending right now. But the longer it holds up – potentially resulting in more AI security breaches – the louder the case grows for someone in Washington to force the issue.

On that note, a headline from CNBC this morning reads “OpenAI reports 6 new instances of ‘concerning model behavior’ since March.”

If the AI industry can’t control itself, the politicians will – in a far more heavy-handed way. Luke himself has said that what ends this trade won’t be a tech failure or a recession – it’ll be politics.

Plus, the bull case rests on one thing

The broader bull case sits on a single assumption – demand for all this AI compute will keep compounding.

I think there’s a strong case for this, but if that demand ever wobbles – if enterprises decide the returns on their AI spending just aren’t there yet – then “supply-constrained” can flip to “overbuilt” faster than anyone expects. And we have history to tell us what could happen then…

In 2000, telecom companies laid enough fiber to wire the world for a decade, all of it justified by demand curves that pointed straight up. Then the spending paused, and the suppliers who’d bet on it got crushed – even though the internet ultimately proved every bit as transformational as promised.

The technology can be real, and yet the stocks can get hurt. That’s the lesson of 2000.

So, where does that leave us?

Luke’s “capability is pacing, capacity is racing” analysis is reassuring, and the money is still moving in one direction. So, this isn’t a moment to run from the AI trade. But it is a moment to own it the way we’ve been describing all week.

On Wednesday, I made the case for stocks built to survive higher-for-longer rates – companies with real earnings and real cash today, not rich multiples riding on profits promised a decade out. That same lens is your protection here.

When multiples compress – and in this environment, they most certainly can – the names supported by actual cash flow should hold up better. The ones priced purely on story will have a tougher time.

Finally, on Monday, we handed out a homework assignment: Sort every AI position you own into a bucket. Bucket 1: the businesses you believe in deeply enough to hold through any drawdown. Bucket 2: the momentum trades you’ll exit the moment they turn – knowing exactly why, when, and how.

If you haven’t done it yet, the recent fireworks are a good illustration of why it’s important.

A quick heads-up

Luke is putting together a full recap of everything he saw and heard at the All-In Summit – the conversations on stage and off – and we’ll bring it to you here in the Digest in the days ahead. In short – he remains very bullish.

But for now, one corner of the AI boom that he’s especially bullish on is robotics. In fact, he recently recommended one young private robotics company that he believes is particularly well-positioned.

Our Digest is running long, so I won’t dive into those details today. But to hear more about it from Luke directly, you can check out his free 2026 AI Megadeal Event right here.

For now, watch the spending, not the Fed. It’s still flowing. Just make sure you’re holding the names built to survive whatever the coming months throw at them.

Have a good evening,

Jeff Remsburg