1AI investment has passed the peak share of GDP that housing reached in 2005
The largest fact of the day is an accounting one. On The Compound and Friends (TCAF 245), Neil Dutta of RenMac and Skanda Amarnath of Employ America worked through what the AI build-out now is in national-accounts terms: roughly 7% of GDP, which is past the share the housing boom reached at its 2005 peak, with tech equipment, software and industrial equipment each above their dotcom and telecom-boom highs. The conclusion drawn on the episode is that tech is no longer a growth sector that happens to be volatile — it is the cyclical sector.
The reclassification is already visible in correlations. The episode counted 15 non-tech S&P 500 names, roughly $2trn of market value between them, that move with the semiconductor complex at a correlation of 0.5 or better. Twelve of the 15 are industrials, among them , , , , , , and — and the reason is mechanical rather than sentimental: their order books are AI-capex order books. Sector classification has not caught up with that.
Breadth points the same way. The equal-weight S&P 500 is up 12.6% year to date against 7.3% for the Mag 7, so the money this year has gone to the AI-touching industrials rather than to the mega-caps that carried the earlier phase of the trade.
The consequence is a hidden concentration: a portfolio holding industrials as ballast against technology may be holding one order book twice under two sector labels. The episode was specific about what would mark the turn — credit spreads widening, the return non-tech corporates actually earn on their own AI spending (as distinct from hyperscalers spending to defend a moat), the fiscal tailwind fading at the end of 2026, and a Fed on hold acting as passive tightening on the labour market.
2The binding constraint on the build-out is now the grid, and the lead times run to four years
Ross Givens, on his own channel, made the case that the throttle on AI capacity has moved from silicon to electricity, and priced it: transformer lead times of about 128 weeks — four years — with transformer prices up 77% since 2019 and grid capacity now treated as a national-defence priority. His four names are the pure expressions of that. (Eaton) did $7.5bn of revenue, up 17% year on year, on record earnings with a $23bn backlog and a 30% year-to-date gain. (Quanta), which builds the lines and substations themselves, grew revenue 26% to about $8bn with roughly 30% EPS growth and a record backlog near $50bn — at about 49 times forward earnings, a premium he argues is justified. (GE Vernova) covers generation, grid equipment and nuclear, trading near $969 and up 40% since January, with a split possible. (Vertiv) is the most direct data-centre power and cooling exposure, with organic orders up 250% year on year, backlog above $15bn and EPS growth of 40% — the fastest of the four.
Brian, on Business with Brian, reaches the same constraint from the generation end: , and on power supply, on the switchgear that connects it. His overarching case is a supply-shock one rather than a grid one — that argument is below — but the power bottleneck is the single theme two channels raised independently in one day, which is the strongest corroboration pattern in the day's material.
Two things temper it. Both Ross Givens videos carry a promotion for a $5-a-year paid service, so the framing is selling something even where the company-level numbers are checkable. And , and surface independently in the correlation list from the first section — but that corroborates that these names now trade as AI derivatives, not that they are cheap. The two findings are about the same names for opposite reasons.
The point about a 128-week lead time is that it is a physical fact rather than a pricing opinion. It does not clear on a quarter's notice, and for as long as it binds, it hands pricing power to whoever already has capacity booked.
3The rate argument is now about direction, not pace
The most contested material of the day was the same episode's Fed segment, and the disagreement in it is about direction, not pace. Skanda Amarnath takes the hawkish side: he sees hikes as possible, on the order of 75 to 100 basis points to remove the insurance cuts delivered in 2025, on an inflation picture he reads as broad and sticky rather than concentrated — oil via the Strait of Hormuz, tariffs, memory, airfares, cattle and apparel all pulling the same way. Neil Dutta takes the dovish side on labour and housing: soft hiring, sluggish income growth, a housing market close to frozen with new-home sales down 6 to 7% year on year, and a K-shaped consumer where the high end is spending freely and travel is at records ( and both at all-time highs) while GLP-1 drugs dent food and restaurant demand (, , and all weak).
What makes it more than a two-handed economist argument is that the split runs through the committee itself. The last meeting produced three hawkish dissents against one dovish one. Warsh, now the chair, is pushing the opposite case from Amarnath's — a golden age of productivity growth with spare capacity, and therefore room to cut. The episode put the market as pricing a possible hike, though no figure was attached to it.
One number underneath the argument deserves its own line, because it cuts against the wage-inflation story: the corporate profit share of national income is at 12.1%, the highest since 1950, against a labour share of 51%, the lowest since 1947. Margins are expanding while real compensation growth is negative. That is the mechanism behind Dutta's "sluggish income growth", and it means the inflation Amarnath describes is not being generated by pay. The sticky components he named are what settle this — if oil, tariffs, memory and airfares keep printing, his case builds; if they roll, the chair's does.
4Friday's selloff hit the AI chain hardest, and gold fell with it
The tape behind this weekend's material was a sharp risk-off Friday: the S&P 500 down 2.64%, the NASDAQ down 4.18% and the VIX up 39.68% in a single session. Gold fell 3.1% alongside it, which is what distinguishes this from a flight to safety — equities and the usual hedge went down together, so it reads as broad de-risking rather than rotation.
The damage was concentrated in exactly the chain the rest of the day is about. fell about 12% pre-market on a guidance miss. was down 11.3%, down 9% even with a ByteDance agreement in the headlines, down 6.2% as the most-active loser, down 4%. was the day's contradiction — marking a $1trn-club milestone into the same session's weakness. Outside the AI complex, fell more than 12% on retail weakness and on a first-quarter miss.
That collides directly with the rest of the day's material. Every buy call named below sits downstream of AI capex — optical, storage, silicon, generation, transmission — and all of them were published into the session in which that chain was marked down hardest. Nothing in the day's material resolves the collision, and it is not the kind that argument resolves: the theses are claims about order books, Friday was a claim about price, and only earnings reconcile the two.
What the sources recommended
Fifteen buy calls came out of four videos, and they resolve into two theses rather than fifteen ideas. Nothing was framed as a sell: the bearish material of the day — , , , and — was put as a diagnosis of consumer demand, with the participants openly unsure whether the weakness is idiosyncratic or macro, and none of it was offered as a call.
The supply-shock ten. Brian ran a single thesis across ten names: disruption at the Strait of Hormuz tightening helium, LNG, sulfur and rare materials into the tech supply chain, delivering the kind of margin shock the 2011 Thailand floods delivered to hard-drive makers — but only for suppliers already positioned. His filter is unusually explicit: two quarters of margin expansion already on the record, signed hyperscaler contracts, and production booked through 2027. On the optical, storage and silicon side: (the only volume supplier of 200G-per-lane EML, with a $2bn stake), (WaveLogic 6, full-year guidance raised to $6.3bn), (Mosaic HAMR, gross margin above 50% for the first time), (a pure-play hard-drive business after the SanDisk spin, also past 50% gross margin), (roughly 73% share in active copper cables, revenue tripled) and (retimer, fabric and cable silicon, revenue up 93% at 76% gross margin). On the power side: (the largest US nuclear fleet, now with Calpine, guiding to 20% EPS growth in 2026), (Susquehanna plus an $18bn AWS agreement), (solid-oxide fuel cells, with Oracle and Brookfield deals behind revenue growth of 130%) and (small-cap switchgear, a $400m data-centre order, backlog moving from $1.8bn to $2.2bn). The caveat arrived in the same breath as the list: several of these names sit near all-time highs, and his instruction was to wait for pullbacks and not chase straight up.
The grid four and a basket. , , and from Ross Givens, on the transformer thesis above, plus — an ETF holding all four among its top positions, offered as the basket version of the same call and carrying the lower conviction that implies. His two videos both carry the paid-subscription promotion noted above.
The second Ross Givens video made no ticker call at all. It is a mechanics walkthrough of the SpaceX listing: priced at a fixed $135 with no range, 555.55m shares, allocations awarded on Thursday 11 June and trading from Friday 12 June, with Fidelity having cut its IPO-eligibility minimum from $500k to $2k of investable assets and five brokers taking retail allocation.
Wednesday's cloud print is the first hard test of the capex claim
reports after the close on Wednesday 10 June — the AI and cloud bellwether of the week, and the name the episode flagged as having recovered on a $300bn OpenAI agreement. It is the first order-book evidence after Friday's tape, and it lands directly on the argument in the first section. follows on Thursday after the close as the AI-software read, and the same day is the cleanest check on the frozen-housing claim. A run of consumer names reports through the week — , and on Monday, and on Wednesday — which is where the K-shaped consumer read gets marked.
The week ends on a flow event rather than a fundamental one: the SpaceX debut on Friday 12 June at $135, followed by a NASDAQ-100 addition after 15 days and Russell 1000 inclusion after five. That buying is mechanical and arrives on a schedule, so it carries no information about demand, and it moves prices regardless.
