HN Debrief

Mag 7 starting to underperform [pdf]

  • AI
  • Economics
  • Public Markets
  • Infrastructure

The PDF is an Apollo market note claiming the Magnificent 7 are beginning to underperform, with the implied culprit being the AI buildout. Its slides point to falling free cash flow at hyperscalers, huge planned capital spending on data centers, elevated valuations, and a broader concern that the market is rewarding chip and infrastructure suppliers while punishing the companies writing the checks. People did not buy the presentation style. A lot of the reaction was that the deck looks like “analysis by slide title,” mixes categories carelessly, and leans too hard on a very short recent period. Still, most of the useful discussion landed on the same core point Apollo was trying to make: these companies are becoming more capital intensive, and that changes the investment case even if the businesses themselves remain strong.

If you own big tech through public markets, the key question is no longer just revenue growth but whether AI capex turns into durable cash flow before buybacks, margins, and balance sheets weaken further. Treat the group less like a single unstoppable trade and more like a set of very different businesses with very different exposure to compute costs and monetization risk.

Discussion mood

Skeptical but not dismissive. People mostly thought Apollo’s deck was sloppy and overclaimed from cherry-picked charts, yet they agreed the underlying issue is serious: AI capex is crushing free cash flow, making hyperscalers more capital intensive, and weakening the old buyback-driven big tech investment story.

Key insights

  1. 01

    AI buildout looks like telecom economics

    The better analogy is not software scaling but infrastructure booms where demand is undeniable and returns are mediocre. If models keep getting swapped out and value shifts toward cheap inference, the winners may be users and hardware vendors rather than the firms pouring hundreds of billions into data centers. That turns AI into a Red Queen race where everyone must spend heavily just to avoid falling behind.

    Model your AI exposure like infrastructure with competitive margin pressure, not like high-margin software. Watch pricing power and utilization far more closely than top-line adoption.

      Attribution:
    • guluarte #1
    • gabriel-uribe #1
    • tptacek #1
  2. 02

    Capex is replacing buybacks and adding leverage

    The sharper financial change is not just lower free cash flow in a quarter. It is that some of these companies are moving from cash-gushing, buyback-heavy machines toward debt-funded capex programs that can dilute the old equity story. One commenter argued Meta and Oracle in particular have piled up large obligations that will hang around for years if AGI-scale returns do not show up.

    Do not rely on past shareholder return mechanics when valuing big tech. Check debt, lease commitments, new share issuance, and whether buybacks are still doing the work investors got used to.

      Attribution:
    • epolanski #1
    • davidpapermill #1
    • 2748484848 #1
  3. 03

    Google has both higher costs and more escape routes

    The most useful debate was around Alphabet. Bears see AI as an expensive overlay on products people already used, with unclear monetization and worse unit economics for search, docs, and assistants. Bulls answered that Alphabet is unusually hedged. It has GCP, TPUs, YouTube, Waymo, and stakes like Anthropic and Databricks. Commenters also claimed Gemini and NotebookLM already have meaningful enterprise use. That makes Google less of a pure AI bet and more of a company with multiple ways to win or at least survive.

    Avoid treating Alphabet as just a search multiple with AI attached. Its valuation now hinges on whether cloud, enterprise AI, and other businesses become large enough to offset higher inference costs in the core franchise.

      Attribution:
    • zerobees #1
    • haberdasher #1
    • blehn #1
    • scarmig #1
    • brainwad #1
    • epolanski #1
  4. 04

    Data center growth is the real signal

    One concrete datapoint people took seriously was the sheer scale of planned U.S. data center expansion. A 60 percent increase in sites planned or under construction suggests the spending wave is still early, and likely larger in dollars than simple facility counts imply because new builds are bigger. That matters more than short-term stock charts because it shows the capex pipeline is still accelerating.

    Track physical buildout, power availability, and construction backlog as leading indicators. They will tell you sooner than earnings calls whether AI spend is peaking or still compounding.

      Attribution:
    • jnwatson #1
    • Creamsicle47 #1
  5. 05

    Diversification matters when winners get too concentrated

    Several comments widened the lens beyond this quarter’s AI panic. The useful reminder was that market wealth creation is highly skewed, yet chasing recent winners still often ends badly over long periods. The Japan example and the “lost decade” reference both point to the same risk. A great twenty-year run can be followed by a long stretch of weak returns while still looking fine in half-century averages.

    If your portfolio is effectively a concentrated tech bet dressed up as broad market exposure, rebalance intentionally. Separate your view on U.S. equities from your view on one sector’s current dominance.

      Attribution:
    • throw0101d #1 #2 #3

Against the grain

  1. 01

    The underperformance call is too early

    The biggest pushback was that Apollo is dressing up a tiny recent move as a regime change. Several readers said you can find many short windows where the Magnificent 7 lag temporarily before resuming leadership. Even commenters who expect an eventual AI bubble pop did not think this deck proved it has started.

    Do not confuse a provocative chart with a durable trend break. If you want to act on this thesis, demand more than a month or two of relative weakness.

      Attribution:
    • mattas #1
    • geori #1
    • bArray #1
  2. 02

    Big tech may not mean-revert like typical winners

    A credible minority argued that applying generic historical studies of past outperformers to firms like Microsoft or modern big tech misses the point. These companies sit on unusual software, data, and capital flywheels that can persist for decades. Ten years ago, the same mean-reversion warning would have looked foolish against what actually happened.

    Be careful importing broad stock-level factor results directly into platform businesses with structural advantages. Check whether you are analyzing ordinary momentum reversals or a genuinely exceptional class of company.

      Attribution:
    • uejfiweun #1
    • M3L0NM4N #1

In plain english

AGI
Artificial general intelligence, the idea of an AI system with broad human-like reasoning ability across many tasks.
capex
Capital expenditure, meaning money a company spends on long-lived assets like data centers, chips, buildings, or equipment.
free cash flow
Cash a company has left after paying operating costs and capital expenditures, often used as a rough measure of financial flexibility.
GCP
Google Cloud Platform, Alphabet’s cloud computing business.
inference
The process of running a trained AI model to generate answers or predictions for users.
Mag 7
Short for the “Magnificent 7,” a label for seven large U.S. tech-oriented stocks that have dominated market performance in recent years.
Red Queen race
A situation where competitors must keep spending and moving just to avoid falling behind, without guaranteed improvement in profits.

Reference links

Research and market notes

Stock return concentration and diversification

Company financials and AI spending

AI market structure and usage