1Datacentre cooling and power now costs 30–43× forward earnings; the cable and conduit behind it, 14–18×
Ross Givens, on his own channel, spent ten minutes on a premise the AI supply chain has been trading on for a year: the conduit, cable, trays, bus bars and cooling gear are ordered before the chips are installed, so that demand is already booked. His point this week was about price rather than demand. The visible half of the trade now sits at 30 to 43 times forward earnings — (Vertiv), (Modine, up 180% over a year at 37×), (Powell Industries at 43×) and (nVent at around 30×) — and he treats all four as prices he will not pay for a new position rather than as businesses to sell out of. (Amphenol) gets the same treatment: profitable, cash-generating, expensive.
The three he would buy sit at roughly half those multiples. (Atkore) makes electrical conduit, cable and trays at about 14 times earnings — a $2.5bn company on roughly $3bn of revenue, earning over $5 a share — and the discount, in his telling, is post-COVID PVC pricing normalising while the electrical and datacentre side keeps growing. (Belden) sells copper, fibre, connectors and enclosures at about 18 times, with revenue and earnings both up 11% quarter on quarter and a $1.8bn purchase of Ruckus Networks underway; he concedes the chart is downtrending and ugly while the numbers are not. (Methode Electronics), an $11 stock, he sizes as speculation himself: laminated bus bars for datacentre power, where the datacentre share of sales has gone 3% to 7% to about 9%, the stock tripled between March and May, and the case rests on the automotive and EV drag falling away against roughly $1bn of guided revenue and positive free cash flow.
Two things travel with those names. The video ran a paid trading service throughout — a subscription, a QR code, a mentoring funnel — so the picks are worth reading as raw mentions rather than vetted work. And moving from 43× to 14× is a view on the price of the theme, not a hedge against it: both ends of that range are levered to the same order book.
2If many companies are valued off the same enormous market, the valuations sum to more than the prize
Aswath Damodaran, in a 68-minute interview on Excess Returns, made the argument one level above the stock picks. His Big Market Delusion is arithmetic rather than sentiment: when many companies are each valued off the same enormous market, every one of those valuations can be individually defensible while the sum exceeds the prize they are competing for. The theme can be entirely real and the aggregate still wrong. His case study was (SpaceX), freshly public, where he is sceptical of the valuation and warns specifically against extrapolating the addressable market. He flagged the crowding into the same handful of mega-caps, among them, as a concentration risk rather than a directional view. It was an hour on valuation method, not on stock selection.
The tape spent Thursday arguing the other way. The Nasdaq closed up 1.91% against the Dow's 0.14%, a gain concentrated in semiconductors: rose 9 to 10% on a Trump-announced partnership with Apple on domestic chip production and its 18A-P process reaching risk production, joined the trillion-dollar club as memory and HBM price forecasts were raised, and each added about 9%, 2.95%, and was upgraded to buy. Note what drove two of the largest moves: 's pact, and (Energy Fuels) up 16% on a $725m US rare-earth loan, are both industrial policy rather than new demand. itself had its first pullback since the IPO debut week, while the underlying business kept selling — rose 3.7% on a Starlink in-flight wifi deal.
The two arguments on the page contradict each other in a way neither can settle. One says the AI theme is sound and only the entry multiple is wrong; the other says the theme can be sound and the total still cannot be earned. Moving down the multiple range answers the first objection and leaves the second untouched.
3Software carries a roughly $24bn short book on the bet that AI eats it
Felix, on Felix & Friends, calls software the most lopsided positioning he has seen in years: roughly $24bn of hedge-fund short interest, on his figures, riding on the idea that AI eats the application layer. What he is trading is the unwind, not the thesis — a crowded position squeezes whether or not the argument behind it turns out to be right.
His simplest expression is , the iShares software ETF and its 110 holdings, Oracle, Microsoft, Palantir and CrowdStrike among them; it bounced off a low near 74 and has reclaimed its 50-day average. From there he goes progressively smaller. (Commvault) does cyber resilience and data protection on roughly $1bn of revenue growing 19% a year, partners CrowdStrike, has bought an AI data-security company, and is down about 60% while making higher lows. (Expensify) is down about 97% from $50, with a 25-million-share buyback and a ChatGPT integration. (Mara Holdings, formerly Marathon Digital) has more than 25% of its float sold short and is pivoting from bitcoin mining into AI and energy infrastructure through the purchase of Longridge Energy. He labels the last two speculative himself, and the same video argued against buy-and-hold in single stocks and sector funds in favour of rotating between sectors.
The channel is running its own funnel alongside the analysis — a free report, a two-hour workshop pitched with a ticket-scarcity line — and the framing throughout is hyperbolic. That matters because the $24bn is a single unverified figure and it is load-bearing for the entire setup; remove it and a chart pattern is what remains. Note also what a squeeze thesis does not claim: if the short book unwinds, nothing about it settles whether AI eats application software.
4The Iran de-escalation trade reversed on a cancelled meeting, and the war has reached an earnings forecast
Switzerland said the US-Iran talks scheduled for Friday were off. That undid the previous session's optimism, which had run on the Strait of Hormuz reopening and crude sitting at pre-war lows: oil erased its early losses and turned back up, and pre-market trading went risk-off. Republicans criticised the terms of the agreement as details emerged, and the Pentagon told lawmakers it needs $80bn for the Iran war and other bills, per the Wall Street Journal.
The cost has already reached a company forecast. (Accenture) fell 17% after the war hit its outlook through consulting demand — a discretionary budget a long way from the Gulf. rose 10% alongside the rest of energy.
The timing matters more than usual, because the US equity market was closed on Friday for Juneteenth. A reversal of this size normally gets a price the same day. This one got futures and crude, and everything else waits.
What the sources recommended
Eleven calls came out of three videos, and the eleven come from the two channels that spent part of their runtime selling a subscription.
The buys. Three from Ross Givens on the AI supply chain: at about 14× as the multi-year hold, at about 18× with a chart he admits looks wrong, and at $11 explicitly sized as speculation. Four from Felix on the software squeeze: as the broad expression, as the fundamental one, and as the speculative pair. Each set is one thesis, not seven independent ideas.
The avoids. , , and are valuation cautions on new entries into datacentre cooling and power, not calls to sell an existing position — the distinction is the source's own, and he is rotating into cheaper suppliers rather than leaving the theme.
None of this gets marked to market until Monday
The next US session is Monday 22 June, and the earnings calendar gives it nothing to work with: 148 entries across the coming fortnight and no major report on either 19 or 22 June, the nearest notables being and on 1 July and on 2 July, all small and mid-cap. That leaves two live inputs. Iran headlines decide whether the escalation premium that came back into crude on Friday stays there, and Fed commentary decides how much of the hawkish read survives — CNBC framed markets as set for a considerably more hawkish Warsh Fed than had been expected, a view that has persisted since the last meeting. The multiple argument in the first two sections has no scheduled test at all before the next reporting season, which is precisely what makes it an argument rather than a datapoint.
