The number of S&P 500 companies trading at 40 times sales or more is back to bear-market levels. The index itself is within a couple of percent of a record high.
The usual way to describe an AI trade at these levels is a bubble in valuations. Luke Kawa's reading is that the expensive tail has already been cut, which leaves the argument sitting entirely on whether the earnings arrive.
"So that tells me the tails are extremely fat in where we are in AI."
Kawa runs markets coverage at Sherwood News and writes its EntryPoint newsletter three times a week, built partly on Robinhood's own data on what retail traders are doing in close to real time; before that he led US rates and commodity research for multi-asset portfolios at UBS Asset Management.
The full interview is covered here so you can skip it. 37 minutes of audio, 22 minutes of reading.
Here are the 13 insights that matter.
👤 Guest: Luke Kawa, Markets Editor at Sherwood News, who writes the EntryPoint newsletter three times a week for active traders using Robinhood's own data on retail positioning, and who previously led US rates and commodity research for multi-asset portfolios at UBS Asset Management
🎙️ Host: Phil Rosen, Chief Market Strategist at ProCap Financial and co-founder and Chief Executive of Opening Bell Media, whose daily markets newsletter reaches more than 200,000 investors
📰 Published: 16 September 2026 on YouTube (Full Signal)
🔴 YouTube | ⏱️ 37 min | ✅ Time saved: 15 min
Key Takeaways
Apple is not being bought as a phone or services company but as the place to hide from hyperscaler capital spending
Its negative correlation to the rest of the group is what has kept index drawdowns shallow all year
The Magnificent 7 are negatively correlated with momentum for the first time ever, and the reason is mechanical
The best-known momentum ETF rebalances twice a year and holds only one of the seven
Hyperscaler free cash flow has turned negative, and the supply of "magnificent" assets keeps growing through equity and bond issuance
US earnings revisions are far better than the rest of the developed world, and the stocks have performed the same
His explanation is that negative free cash flow and issuance are compressing US multiples against that earnings growth
The chip designers with the strongest earnings revisions have de-rated, while smaller names have outrun theirs
Rising oil, rising food and rising yields are all moving together, and either they pull back or stocks do
He is deliberately not buying right now, and is waiting for the market to give him a level to trade against
The AI question will not be settled this year: he puts resolution around mid-2027 and ties it to credit-market appetite
1. Unsettled at a Crossroads
Rosen opened by noting the market was still a few percent from record highs and asked for the ten-thousand-foot view.
"I'd say that the main thing that stands out is how unsettled everything is across the investing backdrop."
He listed what had happened since Labor Day, in order. The traditional post-Labor-Day hangover in stocks; momentum names bouncing off year-to-date lows, with the long-short technology momentum pair picking up; and the equal-weight index breaking below its 50-day moving average after a long stretch in which broadening had been the constructive story.
A closely watched UBS index of software and software-as-a-service companies did a full round trip. It made a fresh record high and then, in his word, "oopsied" all the way back below the previous highs.
The level he is watching on the S&P 500 is 7,600, which the index broke above and has been trading either side of since. With the Fed decision imminent, he said the market is not giving any great theme to latch on to and there is no clarity about the next leg.
2. The Big Stop Getting Bigger
The first chart Rosen put up showed the S&P 100 and the Magnificent 7 trading roughly flat since March.
Kawa treats this as one of the more constructive parts of the market, on the simple grounds that both are close to record highs.
The divergence is the detail worth having: the S&P 100 — the largest hundred stocks in the index — made a fresh record high in recent weeks even though the S&P 500 did not. "That's the first time you had a divergence of that nature since I believe October 2025."
In a market still looking for leadership after a violent momentum unwind, he said the heavyweights starting to pull their weight is an interesting place to look.
The single-stock fact he volunteered is the most surprising number in the section: "I was astonished to find that a stock that is, pretty much done nothing on the year is also a stock that's contributed the most to the S&P 500 positively from the early June record high through about now."
"It's Microsoft, which has done, pretty much nothing." He described it as having been in the software penalty box and the OpenAI penalty box, and said positive earnings are what got it there.
The only time the group has led this year, he said, was coming out of the Iran war bottom, and even then semiconductors were doing the leading. He is watching to see whether that changes, and said it would have implications for the US-versus-rest-of-world trade.
3. Only So Much Magnificent
Rosen asked whether the Magnificent 7 are emerging as a new leader, given how out of favor the group has been as a single holding.
Two things have weighed on it, and the first is cash. For the hyperscalers within the group, free cash flow was on track to turn negative six months ago and has since done so.
The second is supply of the asset itself. There is, he said, "only so much magnificent the market can want" — and much more of it is coming to market, through Alphabet's equity and much more through the bond side.
A cross-asset investor has a finite appetite for Microsoft and Alphabet across both a credit and an equity book. "So I see that as a pretty kind of big cap on a lot of multiple expansion for the group."
The earnings outlook, he allowed, has been supportive for the cohort for a while.
His framing for the group's internal behavior is "like the Avengers" — all powerful, each with a different power. Several of them are in the same businesses and are nonetheless massively uncorrelated.
He attributes a lot of that to Apple, which he does not see as an AI player so much as a toll collector or gatekeeper on it.
The resulting oddity is that the group is near its all-time high while none of its individual members are. "That's a story of how well the group internally plays defense, how low the correlations are, that protects draw downs." Going down less, he said, reduces how much has to be made back.
His two-part explanation for a shallow-drawdown year in the S&P 500: correlation disinflation within the Magnificent 7, and the violent swings between software and semiconductor names.
4. Momentum Doesn't Own Them
The next chart showed the Magnificent 7 negatively correlated with momentum for the first time on record. Rosen asked what it means in plain English.
"I really like this because you can tell a lot of stories with it." The first is a "know what you own" story.
The mechanism is the index construction. The widely followed momentum ETF, MTUM, rebalances only twice a year. "So what's momentum in there is, largely what's been momentum a while ago."
He believes only one of the Magnificent 7 — Alphabet — is in that momentum index at all. So the chart is partly a case of momentum, in this form, simply not owning the names that have been working.
Shorter-horizon momentum indexes run by the big banks tell a different story: over the past two to three months they have become dominated by software rather than semiconductors.
The second story is about how people read the chart. "Most of the time the reaction I've gotten to this is dang the Mag 7 just ain't got the juice anymore." What actually happened on the chart is momentum going nowhere or lower while the group climbed back toward record highs.
His conclusion is that the narrative has drifted away from the price action, to the point where the meme about the group has divorced people from what the stocks are actually doing.
5. Apple, the Hidey-Hole
The third chart was the average pairwise 21-day correlation among the Magnificent 7, set against Apple's correlation to the rest of them.
Kawa explained the construction plainly: it measures how much two members move together on an average session, averaged across every pair Apple is in.
"Apple has been a phenomenal hidey-hole for the group" when investors dislike the capital spending.
The mirror image is Nvidia, which has been lumped in with its own biggest customers in the capex group — a story he called one of 2026's. He thinks that is fair, given the steps Nvidia has taken on the financing side, which make it more of what he called an amorphous AI blob play.
The characterization of Apple is the line that carries the section: "Apple is not a services company. It's not an iPhone company. Apple's a stock that you go to when you do not like how much the hyperscalers are spending on AI capex and it's basically where you hide out and wait to weather that storm."
He saw it again in the week after Labor Day, with markets struggling and Apple holding things together — and said it has nothing to do with enthusiasm for a foldable phone.
The index consequence is the one that matters for risk. To take a market-cap index down a lot you need many members falling a lot at once, or software and semiconductors falling together, and neither has happened this year. "So that's why the VIX has been, you know, pretty dang low" for most of 2026.
Rosen put the obvious question: "So Apple is almost like the defensive play of the Mag 7, but maybe even arguably just a defensive stock in general. Is that fair to say?"
"Yeah, I think that's a pretty safe argument." Unlike most defensives, he said, it still grows decently — something like twice the growth of consumer staples — which supports an above-market multiple while still being treated as a defensive.
The mechanical version is about how pair trades unwind. To be overweight one hyperscaler you have to be underweight another hyperscaler or Apple, so Apple sits on the short side of a lot of AI expressions — and buying pressure arrives when those trades are unwound.
Rosen's addition was that Apple has been called the loser of the AI race for two years and is still beating the S&P 500 this year, and that a company which can use whatever models it likes on top of its own hardware base does not need to participate. The framing offered on the show was that markets reward what is scarce, and what is scarce now is a large company still making decent shareholder returns without spending heavily.
6. The Broadening Cracks
The next chart was the S&P 500 equal-weight index closing below its 50-day moving average for the first time since April.
"I think why it's relevant to me is it cuts against a lot of prevailing narratives."
He explained why broadening was in vogue in the first place. During the momentum unwind, the only reason the market did not fall apart while semiconductors did is that everything else had to pull its weight.
It also coincided with US economic data surprising to the upside, particularly outside the labor market — spending data in the US and globally, and industrial data, which he said continues to surprise.
The question he now poses, with higher rates and higher commodity prices: "how long can the many outperform the few?" Right now the many have shown some weakness, and the equal-weight ETF's year-to-date lead over the S&P 500 tracker has narrowed to very little.
"Consumer discretionary stocks have been completely shot to hell I would say lately." A lot of them bounced off resistance and the whole group has gone lower.
The observation he finds most notable is a reversal of the usual hierarchy: "This to me is just very interesting in talking about for probably the first time I can remember there's a lot more confidence in the global industrial cycle and AI but not just AI doing well than there is in the US consumer doing well."
He flagged that against his own priors. "I come from the school of the US consumer is the engine of the US economy and that's the engine of global growth." The open question is how far the industrial side can pick up while US consumption growth decelerates in the back half of the year.
7. What Tightening Slows
Rosen said he expected the megacaps to pick up in the second half and asked whether that matched Kawa's view. The answer turned into a question about what a tightening cycle can actually slow.
His starting point is the purpose of the policy: "It's to slow inflation and slow nominal activity."
The available dials are consumption or rate-sensitive industrial activity, and he ruled one out immediately. Housing, he said, probably does not have much room to slow from here.
"So, you're talking about can you slow AI or do you continue to slow through the US consumer?" He said he has heard good arguments on both sides.
His own answer is about where the imagination is: "From my vantage point, the opportunity in AI and the possible growth that people are getting and extrapolating from that just runs a lot more than the growth you can dream on the consumer space."
The capital-allocation conclusion follows: "So in terms of what gets more capital I would say capital will go to the place where you can dream on it a little more a little better."
Whether that benefits the Magnificent 7 themselves or the beneficiaries of their spending, he would not say. "Not sure. Not sure to be honest."
"And I think I'm not trying to be a hero right now with things at a crossroads." He expects the market to declare itself over the next few weeks through the midterms, which will offer a better chance to react than guessing now.
8. US vs Rest of World
The next pair of charts set the S&P 500 against the MSCI World ex-US index — roughly level year to date — and then set US earnings revisions against those of the developed world outside the US.
"Well, usually there's a bit more of a performance gap between the two." Historically, he said, there have been stretches of sustained US outperformance, and the market now appears to be on the other side of one.
What makes it strange is the earnings: "US earnings revisions have been so much better than DM XUS." Stocks generally follow earnings, he said, but the magnitude can differ a great deal.
His explanation for the gap is the same cash-flow point from earlier: hyperscaler issuance and negative free cash flow are compressing price-to-earnings multiples across a big chunk of the US market, offsetting some of the earnings growth.
The part that surprised him is Europe's resilience. High commodity prices and high rates would normally make him fear the worst for Europe: "And Europe's hung in there, I think, very extremely well."
So he expects one of two things to give. Either the commodity and rate headwinds back off, or rest-of-world performance — Europe in particular — gets more challenged while those two macro risks stay on the front burner.
His reasoning on relative sensitivity: the US is much less sensitive to energy costs than Europe, and between US AI spending and European non-AI cyclical spending, he thinks the European side is the more rate-sensitive.
Rosen pushed on why nobody is calling it a catch-up trade, given how much better US forward revisions are.
Kawa's answer is that the cheapness depends on the metric. "US has gotten a lot cheaper on PE. US has gotten a lot more expensive on free cash flow."
He framed the market as close to a binary bet on whether the return on investment in AI is good, and said that is not resolved. "I don't necessarily think we'll get quick resolution on that in you know in a quarter or two." He put it closer to a mid-2027 story, and tied it to the credit market's appetite to keep financing the spending.
So the gap makes sense to him: investors have doubts given the scale of the spending and clear eyes about what usually happens at "the back end of a capex binge."
9. The Great De-Rating
The next chart plotted Broadcom, Dell, Marvell and Nvidia by 2026 price change against revisions to 12-month forward earnings. Kawa built it ahead of Nvidia's and Broadcom's results.
The point was that the major chip designers are not being rewarded in proportion to their expected earnings growth over the coming year.
Six months ago, he said, there was a clear safety trade in the market — and he included two things people would not immediately file that way. "Like people forget consumer staples started off the year doing pretty darn well." The crowding into memory was the other: going for the bottleneck, where supply constraints and pricing power are, which he treats as a form of safety.
What the chart shows now is Dell and Marvell far outperforming their earnings revisions while Nvidia and Broadcom sit below theirs — which he reads as an increase in risk appetite.
The arithmetic underneath it is size: "And I think it does get to the point that hey, it's a lot easier for probably a 500 to $600 million company to double sales, double market cap, double EPS." Doing the same from a trillion-dollar base is another matter.
He tied it to Jensen Huang's question about whether Marvell is the next trillion-dollar company. "There's an appetite to find the next one and the companies that have already been the one. Well, once you're the one, it's harder to be than the next one."
Rosen used the chart to relay a previous guest's call and disclose his own position. Jay Hatfield, a chief investment officer and portfolio manager who had recently been on the show, called Marvell the idea of the decade: "Look, this might be my favorite stock of the last five years and the next five years."
Rosen owns it: "I personally own Marvell. I think it's a great company. It's been a dog for the last few months, and I don't know why because nothing has really changed in its business."
10. Expensive and Slow
The next chart set Walmart against Tesla, Micron and SanDisk on forward earnings and year-over-year revenue growth.
Kawa said Tesla and SanDisk were there as outliers, but the chart's real use is as a filter — working out what not to own.
The conditions that make the filter work: nominal growth is firm even if coming off the boil, there is a lot of demand for capital, interest rates are elevated, and the top line of the index is growing quickly.
On valuations of private AI stakes he was dismissive of the debate and pointed at revenue instead: "topline don't lie" and the top line is growing rapidly.
The question that follows is the section's whole argument: "Why in a world where the top line of corporate America seems to be growing quite well, would I then want to own something expensive that is also growing, relatively slowly?"
That, he said, is most of the reason consumer companies have ended up in the penalty box, in both staples and discretionary. "It just has to do with the growth not being there." He noted those stocks have since become less expensive.
The evolution he sees over 2026 is investors becoming willing to not want safety, in a world where top lines are growing quickly and growth has surprised to the upside even through a large oil shock.
The historical detail he added is how differently the Fed first read that shock. In the minutes of the meeting that followed the start of the Iran war, most officials were worried about downside risks from the war and the rise in commodity prices, and thought it might be a reason to cut rates further.
Rosen's reaction marks how fast that changed: "Man, I remember so clearly when we had more people talking about cuts than hikes."
11. Bad Momentum
Two more charts: the Citi global economic surprise index at its most positive since 2022, and the 10-year Treasury yield plotted with an agriculture fund and an oil refiners ETF.
The surprise index, for Kawa, is mainly shorthand for the move in long-term yields. Despite a big oil shock, the economy is holding up on still-decent consumer spending, a large tax refund season, and an industrial side that is doing well worldwide.
His advice is not to over-complicate the move. "I don't think you have to over complicate the rise in yields too much. I don't think you have to start talking about Fed independence too much." Nor, he said, does it require a discussion of supply and buybacks.
"I think you can simply say long-term yields are rising because global growth has been fairly resilient and better than expected all over the world and the Fed seems poised to start a tightening cycle."
The two facts he used to size that tightening cycle: it would begin nearly 100 basis points above where the last one ended, and from the highest starting level since 1999. Resetting rate expectations that dramatically, he said, is why it makes sense to be talking about yields at their highest since 2007 or 2008.
The three-line chart is the one he says the market has to resolve. "this is bad momentum to me." Higher oil slows growth; higher food prices hit discretionary spending; "Higher yields generally slow activity, not good for stocks," and they compete with stocks for a place in a portfolio.
What puzzles him is that the index has stayed near record highs while momentum migrated from semiconductors into all three of those. He reads that as a testament to the resilience of the economy and of earnings growth, and not something people will extrapolate forever.
His conclusion is a straight either-or: "So either these need to pull back I would say or stocks probably need to pull back but I would not expect both to be remaining at very elevated levels and rising forever."
12. Waiting for Pitches
Rosen asked the question the episode's title promises: where are the best opportunities right now?
He started with the exclusion. "So, again, staying away from, companies that are posting slow topline growth." They might bounce on an oil pullback, but in a world where the worry is too much growth, he does not think slow growth gets rewarded.
The inclusion is a method rather than a name: "My general strategy and this is how I got into Marvell too at one point is buying tech companies when they get bombed out whether that's in software or semiconductors."
The unusual part is his warning against fundamentals. "I do think there is a danger in this market and particularly in this regime to be focusing, too much on the fundamentals. It sounds weird to say that." His phrase for it is "a standing too close to an elephant problem" in technology stocks.
He is explicit about the limits of his own edge: "I'm never going to be the person who can, tell you why a certain tech breakthrough, a design breakthrough, that a company has made or the benefits of a certain kind of TPU versus GPU for inference, means that one company's going to do much better than another."
What he can do is establish levels — a low and a high to benchmark against — so that he knows when he is wrong.
The reason he leans technical is a view about the earnings themselves: "I think at some point there's going to be a point in time in the AI boom in which the earnings that we're looking for might not be realized." So he prefers to "respect the price action" and play higher highs against lower lows.
Right now that framework gives him nothing to do. After the momentum unwind, few things are making decisive higher highs, and a set of AI-industrial stocks went from below the 200-day to above the 50-day and back below the 200-day inside a week.
"So, with that stuff going on, I think now is a great time to be waiting to see what pitches that the market is going to give you." He said plainly: "I haven't been, adding aggressively to anything right now."
The discipline he draws from it: "I don't have to intellectually pretend that I've got an edge in this stuff that I definitely know I don't." He would rather own higher-beta names that can bounce well when the market turns.
13. Fat Tails, Not a Bubble
Rosen's last substantive question was where we are in the AI bubble conversation, and he raised one of Kawa's own charts on the number of S&P 500 stocks trading above 40 times sales.
The count of stocks trading at 40 times sales or higher is at bear-market levels while the index is within a couple of percent of a record high.
"So that tells me the tails are extremely fat in where we are in AI." Both outcomes are live: aggressive earnings growth priced in alongside multiple expansion, or — for mainly credit-related reasons — multiples contracting further while the earnings fail to arrive.
"So I would say we're still at a point of extremely high optionality when it comes to where we are in AI." It is also why he prefers not to be married to a fundamental thesis: price can tell him where things are.
The paradigm he says has held from the start is that companies have insisted "the risk is spending too little not too much," capital-spending budgets keep rising, and the constraint is "limited by supply and not demand."
He is careful about what that has meant for valuations: roughly two years of this has not been a paradigm conducive to multiples expanding. "But I think the optionality on both sides continues to be massive."
Asked whether that amounts to saying it is not a bubble, his answer separated the two possible bubbles: if there is one it is in expected earnings, because "the bubble is clearly not in valuations." He called that the softer form of the statement.
Bonus Insights
Kawa's closing plug was for his own work: he is on social platforms as ljkawa, and writes the EntryPoint newsletter three times a week for Sherwood News. It is aimed at active traders and carries the charts discussed on the show plus proprietary Robinhood data on what retail traders are doing in close to real time, which he described as a way to see how sentiment is shifting.
Rosen's sign-off is a compact endorsement of the guest's value: "Luke, you have some of the best data in finance."
Rosen also offered a piece of buy-side observation of his own. In December he was talking to investors who planned to rotate out of the S&P 500 and into the equal-weight index or the S&P 493 on the broadening trade; his guess is that many of them have since rotated back into the market-cap-weighted index.
Kawa's response to that was a compliment rather than an endorsement: "Hey, well, if they have timed it like that, they've probably done a heck of a good job in terms of switching up there."
Kawa's bottom line is that the market is at a genuine crossroads rather than a turning point: the expensive tail has already de-rated, index-level volatility has been suppressed by low correlations inside the largest stocks rather than by calm, and the three things moving against equities — oil, food and yields — have to resolve before he is willing to add risk.
Products, Companies & Tools Mentioned
Apple (Not an iPhone or services company in his framing, but the place investors hide when they dislike hyperscaler capital spending)
Microsoft (Flat on the year and still the largest positive contributor to the S&P 500 since the early-June record high)
Nvidia and Broadcom (The chip designers whose share prices have lagged their own forward-earnings revisions in 2026)
Marvell and Dell (The other side of that chart: both have outrun their earnings revisions, and Marvell is a position the host discloses)
Alphabet (The only Magnificent 7 member he believes sits in the main momentum index, and a source of the new "magnificent" supply through equity and debt)
Walmart, Tesla, Micron and SanDisk (The expensive-and-slow-growing filter chart, with Tesla and SanDisk as the outliers)
MTUM, the iShares MSCI USA Momentum Factor ETF (Rebalances twice a year, which is why its holdings reflect what had momentum a while ago)
RSP, the Invesco S&P 500 Equal Weight ETF (The broadening trade in one ticker; its year-to-date lead over the market-cap index has nearly gone)
Books & Resources Mentioned
EntryPoint (Kawa's newsletter for Sherwood News, published three times a week with the charts from this episode and Robinhood's retail-positioning data)
Watch the full episode:
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens:
The supply point rarely makes it into the AI valuation debate: hyperscaler free cash flow turning negative while the same names issue equity and bonds means the supply of magnificent assets grows faster than cross-asset books can absorb it, and a cap on multiple expansion follows without anyone needing a view on the technology itself. The mechanical momentum detail earns its place too, since a twice-yearly rebalance that holds only one of the seven explains a first-ever negative correlation that would otherwise invite grand narratives. And tying resolution to credit-market appetite around mid-2027 turns the AI question into something that can be scored, which is rarer than it should be.