1The AI bubble argument narrowed from a demand question to a moat question
Jan van Eck, chief executive of VanEck, spent an hour on The Compound and Friends drawing a line through the middle of the semiconductor complex, and the criterion is not valuation — it is whether there is a moat underneath the earnings. Memory sits on the wrong side of it. and the SanDisk business inside are a duopoly whose profit surge comes from price rather than units, which leaves it exposed on two fronts at once: Chinese memory entrants on supply, and more memory-efficient models on demand. His formulation is unusually dated for a structural call — will "very obviously" not be a top-five S&P 500 earnings name in 24 months.
That is a larger statement than a single-stock view, because on his own arithmetic the index's top five earners — , , , and — produce roughly a quarter of its earnings. One of those five, on this argument, is a cyclical whose pricing breaks. He counts about 263 S&P names tied in some way to the AI buildout, which is what he means by calling the index "an AI momentum ETF", and cites Nvidia's own finance chief on $3–4trn of annual AI infrastructure spending by the end of the decade.
sits on the right side of the line: CUDA is the moat, the business is the "Walmart of compute", and the survivor analogy he reaches for is IBM rather than a dot-com. The quarter backs the framing — earnings per share up 125%, gross margin 75%, guidance about $4bn above, the buyback lifted to $80bn — and the stock still fell around 1.5% on it, which he attributes to a SoftBank-driven rotation into a blue-chip holder base rather than to anything in the print. Around that he is selective rather than uniformly bullish: is the only mega-cap he notes on the new-high list and the one that owns both the model and the chips; 's Reels is at roughly a $50bn run rate on his recollection, more than all of Netflix's revenue; , up 100% on the year, is his example of a move that does not hold; and is down about half from its high with capex debt sitting off the balance sheet.
Note what the tape did with all this. Google News's pre-market coverage has Nvidia's earnings pulling memory and semis higher together — one rally treating both halves of the distinction as the same asset, in the week the argument is that they are not. Two caveats travel with the call. It is one firm's view stated once, and it is hedged by the person making it, who says repeatedly that he could be wrong and that he is bad on timing. The memory thesis rests on a judgement about moats rather than on a dated catalyst, and several of the supporting numbers were quoted from memory.
2Most of the US market's 30-year lead over the rest of the world was re-rating, not growth
Cliff Asness of AQR, on Excess Returns, put a number on the thing US-versus-international arguments usually talk around. Of the US market's 30-year outperformance against non-US equities, he attributes 75–85% to re-rating — the multiple going up — and only 15–25% to faster real growth. The same decomposition shows up inside the US market: classic book-to-market value has gone roughly nowhere for three decades, and about 80% of that shortfall is growth multiple expansion rather than fundamentals. His conclusion is a portfolio default rather than a trade: he starts global, on the grounds that the US can lag over the next decade even if American companies do perfectly well.
The starting point he measures from is stretched. The Shiller CAPE sits near 40, the second-highest reading on record, and he refuses the usual escape hatch, that 2000 was different because it was worthless dot-coms. It was not: that index was led by profitable companies, among them at around 100 times earnings, which is structurally what today's AI leaders are.
Then comes the hedge, which is inseparable from the call. He will not short the market or AI at AQR, describes it as something like 19-in-20 odds rather than an arbitrage, and says flatly that whether AI is worth it is a single bet and not a question a quant process answers. He is explicitly not calling a top in 2026. The one place he is constructive is the value spread inside his own composite, around the 75th percentile — better than normal, he says, while much else looks worse than normal. The same care runs through his private-markets point: unobserved marks understate risk, what he calls "volatility laundering", the illiquidity premium may have shrunk or flipped, and forward private-equity excess returns over public equity could be low or negative — again with no date attached.
Both interviews end up looking outside the US, and it is worth not reading that as agreement. Van Eck arrives there on a ten-year view of AI and India; Asness arrives from an arithmetic decomposition that is indifferent to what the money is invested in. Same destination, unrelated reasoning.
3The long bond is back at a 2007 high, and the rate argument has moved from cuts to hikes
The 30-year Treasury yield is at its highest since 2007, and van Eck's account of why is entirely supply-side: hot CPI and PPI prints, defence spending, and strong nominal GDP. He describes a market flirting with hike odds rather than with cuts. His own worry about US debt and deficits comes with an unusually explicit trigger — the 10-year above 5–5.25% — and an unusually explicit admission that he has been bad at timing it.
Sitting on top of that is a change of Fed leadership that ran through the interview and both news feeds: Powell's final FOMC remarks on inflation led Yahoo Finance's market coverage, and Kevin Warsh's incoming chairmanship recurs in the Google News feed and the interview alike. Van Eck expects a quieter, stabler chair — "less knob-turning" — and the only version of the handover that concerns him is an aggressive shrinking of the balance sheet. So the personnel change is being read as continuity, with the risk relocated from the rate path to the balance sheet. If the long end is moving on fiscal supply rather than on policy, a new chair does not fix it, and the 10-year level is the marker he says would tell him the market had stopped absorbing it.
4The coming mega-listings are a supply event before they are a growth story
Two very different sources described the same pipeline inside a day. Van Eck flags SpaceX at around $2trn with a rolling lockup and quick Nasdaq inclusion, and Anthropic at a roughly $40bn revenue run rate that is doubling, framing both as liquidity and wealth events rather than as anything to own. Yahoo Finance's coverage adds Reuters reporting that OpenAI is aiming for a speedy IPO of its own.
The Felix & Friends video takes the same event and works the mechanics from the other end. A SpaceX listing at $1.5–2trn is exit liquidity for the private holders already in; the listing itself offers maybe 2x on his estimate against 10–20x in the small suppliers; and the risk he names is a lockup expiry roughly six months after listing, for which he offers Voyager's roughly 70% post-listing decline as the analogue.
Put the two halves together and the calendar is the story. Fast index inclusion creates mechanical passive demand on one date, and a lockup expiry frees early holders to sell on another. Which arrives first, and how far apart the two sit, does more to the price in the first year than the operating story does.
What the sources recommended
Four buys, all single names, all from one video and all in one theme — and the video carrying them is built around promotional calls to action, which is part of the record on them.
(Redwire) is the picks-and-shovels name: solar arrays, antennas, 3D printing on the ISS, roughly 60% of revenue from government, with commercial space stations and NASA's Lunar Gateway as the named catalysts. Pitched as the lowest risk of the four, for a medium reward. (Voyager Technologies) is the other end — Starlab, an ISS replacement built with Airbus against a station decommissioning around 2030, with defence revenue funding the build; the highest risk and the highest framed conviction of the four. (Firefly Aerospace) has the Alpha small-launch and Eclipse medium-lift vehicles, a completed lunar landing and NASA missions behind it, listed near $24 and trades around $43, with about $300m of cash. (Orbit International) is a micro-cap in ruggedised military electronics that listed at $15, trades near $4 and is burning cash — the presenter's own description is a "tiny lottery-ticket position only". He caps the whole theme at about 5% of a portfolio.
Read those as one thesis from one source, not as four independent calls, because three things travel with them. The video repeatedly directs viewers to a free download, a "last time ever, no replay" workshop and a paid app trial, and leans on a past 500% personal winner for credibility. Its own details are loose in places: Voyager's ticker is given on air as "V", which is Visa, and Firefly's is never stated at all. And the "$14 stock" of the title does not correspond to what the four names actually trade at.
The two institutional guests made no single-name calls, and each said why. Van Eck keeps to thesis level because of the marketing rules that apply to a fund manager speaking in public; Asness declines single bets on principle, since whether AI pays off is not a question his process answers. Their directional views sit in the sections above, plus one that did not fit them: van Eck expects enterprise software to de-rate as AI compresses seat-based pricing, and says he shortened his own contract — a decision about his own spending, not a trade.
The week's calendar tests the software claim, not the memory one
US markets closed Friday higher — S&P 500 +0.37%, Dow +0.58%, Nasdaq +0.19% — and are shut again on Monday for Memorial Day, so the week opens on Tuesday 26 May with , , , , and .
Wednesday is the dense day and the relevant one. reports, which is the direct test of the enterprise-software de-rating; and land alongside it on the silicon side, with , , , , , and . Thursday brings , — one of Friday's largest gainers at +16.77% — and , whose target BofA reset going into the print, plus , , , , , , , and , which was Friday's worst performer at −27.53%. on 3 June, with , and around it, is the cleanest read on AI-silicon demand on the near calendar.
What the week does not offer is a memory print. The sharpest claim of the day — that memory's earnings are price rather than units — has no scheduled test on it, which leaves the software de-rating on Wednesday and AI-silicon demand the following week as the parts that can be checked at all. A structural call with no dated catalyst is exactly what the person making it said it was.
