Sløk's single-chart Daily Spark plots YTD total returns for the Magnificent Seven against the equal-weighted S&P 493 and shows the lines have crossed, with the 493 now ahead. By publishing this as Apollo's house view to LPs without commentary, he is signaling that the consensus 'you must own the Mag 7' trade is breaking down in real time.
By submitting the Apollo PDF to HN with no editorializing and driving it to 172 points, mooreds is amplifying the underlying claim that the Mag 7's index dominance is reversing. The submission frames the chart itself as the news.
A faction of the HN thread argues the chart is methodologically unfair because Nvidia alone moved roughly a quarter of last year's index return, so comparing seven mega-cap names equal-weighted against 493 others isn't a like-for-like read. They contend the Mag 7's pullback after such an outsized run is statistical mean reversion, not a regime change.
The editorial argues the real significance isn't that seven stocks dipped, but that these specific seven balance sheets underwrite essentially all US AI infrastructure spending — hyperscaler buildouts, Nvidia's order book, data-center power contracts, and training runs. With Microsoft, Alphabet, Meta and Amazon alone on pace for ~$300B of capex, a sustained de-rating of their equity directly threatens the financing model for the AI buildout.
Roughly half the thread posted screenshots of their own broker YTD tabs confirming the gap, treating the chart as validation of what they were already seeing in their portfolios. Their take: utilities, industrials, regional banks, and healthcare — the unsexy 493 — are up high single digits while the AI-narrative names are flat, and that broad-based participation is healthier for the market than narrow leadership.
On June 28, Apollo's chief economist Torsten Sløk pushed a single-page Daily Spark titled *Mag 7 Starting to Underperform*. The PDF is one chart and a paragraph: year-to-date total returns for the Magnificent Seven — Apple, Microsoft, Alphabet, Amazon, Meta, Nvidia, Tesla — plotted against the equal-weighted S&P 493 (the index minus those seven). The lines crossed. The 493 is now ahead.
That is the entire post. No thesis, no recommendation, no 40-slide deck. Sløk's house style is to publish the chart that breaks the consensus and let the consensus argue with it. The HN thread (172 points, climbing) is doing exactly that — half the comments are screenshots of their own broker year-to-date tabs confirming the gap, the other half are arguing about whether equal-weight comparisons are fair when Nvidia alone moved a quarter of last year's index return.
The numbers behind the chart are not subtle. Through late June, the Mag 7 basket is roughly flat to mildly negative on the year. The S&P 493 — utilities, industrials, regional banks, healthcare, the boring two-thirds of the economy — is up high single digits. Twelve months ago this comparison would have been a punchline. It is now Apollo's published view to LPs.
The Mag 7 was never a fund construct. Bank of America's Michael Hartnett coined it in 2023 as shorthand for the seven names that had collectively eaten the index. By 2024 they were ~32% of the S&P 500 by weight and delivering ~60% of its return. Every long-only manager in the US had to own them or explain why not. That trade is what's rolling over.
The interesting question is not whether seven stocks are down — it's *which* seven and *when*. These are the same seven balance sheets currently funding essentially all US AI capex: the hyperscaler buildouts, the Nvidia order book, the data-center power contracts, the model training runs. Microsoft, Alphabet, Meta and Amazon alone are on pace for roughly $300B of capex in 2026, the majority of it AI-adjacent. Nvidia is the counterparty. Apple and Tesla are the wildcards funding their own bets. When the market re-rates this cohort downward while they're mid-buildout, it's not a sector rotation — it's a financing signal.
This is the second time in two weeks the same alarm has gone off from a serious institution. The BIS annual report, covered here yesterday, drew an explicit parallel to the 1999–2001 telecom capex cycle: tower companies and CLECs that raised debt against forecasted demand, built fiber that took a decade to light up, and dragged their lenders down with them when the cash flows didn't show up on schedule. The BIS wasn't worried about transformers; it was worried about the capital stack funding the GPUs. Apollo's chart is the same worry expressed in equity prices instead of bond spreads.
The community reaction on HN is split along a predictable line. The "this is healthy" camp points out that the 493 had been historically cheap, earnings revisions there are turning positive, and a broadening rally is what every strategist asked for in January. The "this is the top" camp notes that every prior AI-adjacent capex peak — fiber in 2000, solar in 2011, crypto mining in 2021 — was first visible in the equity of the companies doing the spending, not the companies selling the picks and shovels. Nvidia is the pick-and-shovel name. It is also down on the year.
There is a third reading that neither camp is making loudly enough: the Mag 7 trade was always partly a *passive* trade. Index inflows mechanically bought these seven names in proportion to their weight, which pushed the weight higher, which pulled in more inflows. Once that flywheel stops — because returns disappoint, because regulators cap concentration in target-date funds, because the BIS gets loud enough — the unwind is mechanical too. Apollo isn't predicting that unwind. They're pointing out the line has already crossed.
If you're an engineer reading equity charts: normally you shouldn't, and this is the exception. Three concrete reasons it matters for the people shipping code.
First, vendor risk. The hyperscaler pricing you're modeling for 2027 — egress, GPU-hour, managed inference — was set when these companies could fund capex out of multiple expansion. If their cost of capital re-rates because the stock stops working, the unit economics they're charging you become the unit economics they actually need. The era of below-cost AI inference subsidized by stock-comp'd engineers and patient shareholders has a half-life. Price your roadmap accordingly, and treat any 3-year committed-use discount as the option it actually is.
Second, comp. A non-trivial fraction of senior IC compensation at these seven companies is RSU-denominated and back-loaded. The 2023–2024 hiring class is sitting on grants priced at the top of the move. The 2026 grants are being struck against a chart that just crossed. If the Mag 7 spend 2026 trading sideways, the talent market re-prices — and it re-prices in favor of profitable mid-caps, sovereign AI labs, and the 493 names that are quietly hiring platform engineers for the first time.
Third, your build-vs-buy calculus. The implicit assumption behind "just use the API" is that the API provider is a perpetual-motion machine with infinite balance sheet. That's the trade Apollo is flagging. It doesn't mean self-hosting Llama is suddenly cheaper — it means the option value of being *able* to self-host, of having your prompts portable, of not having a single OpenAI/Anthropic/Bedrock dependency in your critical path, is worth more than it was last quarter. Architect for substitutability now, while migration is a refactor rather than a fire drill.
One chart from one economist isn't a regime change. But Sløk has a track record of publishing the chart three to six months before the narrative catches up — he was early on regional bank deposit flight in early 2023 and early on the commercial real estate roll in 2024. The Mag 7 may bounce, the 493 may give back the lead by August, and this post will look like noise. The thing worth watching isn't the line itself — it's whether the next two earnings cycles validate the capex narrative the seven balance sheets are betting on. If they don't, the chart that crossed in June is the chart that gets cited in the 2027 postmortem.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.