A publication by EquityTribes
TINO
Second SeriesEdition 2
Everyone Is Long the Build-Out
AI has become the default holding of ordinary savers, financed through private credit that is rarely marked to market, while its costs surface in power bills, gas balances and the price of a phone. Exposure to the build-out is no longer a choice; it is what everyone already owns.
Tectonics · Inflections · Narratives · Outliers
Tectonics
AI has become the default holding, and it sits mostly in private markets
The Force
In his Money Stuff column of 15 October, Bloomberg Opinion's Matt Levine put the shift in one line: "AI in 2016 was esoteric alpha; AI in 2026 is bog-standard beta." A large share of expected earnings growth now comes from AI, and when investors worry about an AI slowdown, both stocks and bonds fall. An investor who wants broad exposure to economic growth gets AI exposure whether they asked for it or not.
Much of that exposure is not in the stock market at all. OpenAI and Anthropic are still private and, Levine writes, probably worth more than a trillion dollars each; Bloomberg reported the same week that Anthropic's revenue rose to $16.8 billion in the quarter before its planned IPO. The financing for the build-out comes in large part from private credit and infrastructure funds run by the large alternative asset managers, whose money comes from endowments, pensions, insurers and, increasingly, individual savers.
Blackstone shows the scale. Bloomberg reported on 15 October that its president, Jon Gray, has built a $200 billion data-centre business, is building a cloud venture with Google, has taken stakes in labs including OpenAI, has lent against Broadcom and Nvidia chips, and is committing capital to power plants and data-centre cooling. Gray calls the AI investments "the single most important thing" under way at the firm. The chip lending began in 2023 with a multibillion-dollar GPU financing for CoreWeave, which Gray says he initially balked at.
The Consequence
The return on a large part of household savings now depends on the AI build-out, but it reaches savers through vehicles that trade rarely, are valued by their managers and charge higher fees than an index fund. The public markets that financed earlier infrastructure booms are, as Levine notes, not doing most of the work this time.
Inflections
BloombergNEF raises its data-centre outlook by half, and gas takes the strain
The revision
118 gigawatts by 2030, up 52% in ten months
BloombergNEF's US Clean Energy Market Outlook, released on 16 October, puts installed US data-centre capacity at 118 gigawatts in 2030, 52% higher than its projection of last December. Average data-centre power demand could reach 120 gigawatts by 2035, roughly five times current levels, with AI-related computing responsible for nearly half.
The fuel
A gas surplus that may become a deficit by 2028
BNEF expects 67.8 gigawatts of new gas-fired capacity could be added through 2030, against only 14.3 gigawatts slated to close. A significant portion of new data centres will be powered by on-site gas, and BNEF expects US gas supply to swing from surplus to deficit as soon as 2028.
Renewables are growing too: BNEF expects a record 89 gigawatts of US clean energy this year and more in 2027, but the surge is driven largely by developers racing to claim expiring federal tax credits. "We still see data center growth as a tailwind to renewable energy, but it is tainted by the growing gas pipeline," said Hall Wellborn, an author of the report.
The bill
Who pays shows up in the power bill
Electricity bills have become one of the most potent issues in next month's US midterm elections. A Bloomberg News analysis of EIA data shows prices for the two-thirds of US households served by investor-owned utilities have risen 34% since 2020, against 19% for customers of publicly owned utilities. Many politicians blame data centres; some are now questioning the investor-owned utility model itself.
Narratives
Private credit is insulated from public-market swings
The Claim
Private credit loans hold their value when public debt markets fall, which makes them a low-volatility, low-correlation way to finance the AI build-out.
The Evidence
In early October Apollo Global Management began rolling out daily valuations across its roughly $850 billion credit business. In mid-October, as publicly traded bonds and broadly syndicated loans fell several points, the move drew open pushback at the CAIS Alternative Investment Summit, Bloomberg reported. On a panel on 14 October Ted Koenig of Monroe Capital said: "Daily marks in illiquid private assets is an illusion of precision, not accuracy."
Drew McKnight, co-CEO of Fortress Investment Group, said a book marked towards a leveraged loan index could be valued daily on a spread basis, but "it wouldn't show you if the loan is at par unless you're doing the actual analysis." Both objections are about the quality of the mark, not about whether the loans move with public credit.
Levine's reading in Money Stuff: low correlation with public markets is the pitch that justifies private-credit fees, and moving marks with jumpy public prices undermines it. But a portfolio of loans to AI-exposed companies is plainly related to public loans to similar companies; if it were perfectly correlated with a loan index, the manager would not be earning its fee, and if it were wholly uncorrelated, marking to that index would make no sense.
Outliers
The AI build-out just caused the worst quarter on record for smartphones
The Connection
IDC's preliminary figures, published on 14 October, show 255.3 million smartphones shipped worldwide in the third quarter, down 22.0% from a year earlier and the largest year-on-year decline of any quarter IDC has tracked. IDC attributes it to "the AI-led memory crisis": Nabila Popal, a senior research director, said memory costs have risen about 350% since last year and are still rising, and that "the worst is yet to come".
The damage is uneven. Samsung's shipments fell 5.8% and Apple's 4.0%, while Xiaomi fell 29.2%, OPPO 34.5% and vivo 31.5%. "The pressure is now moving from vendors to consumers," said IDC's Francisco Jeronimo. Meanwhile TSMC projected December-quarter revenue of as much as $54.3 billion, and SK hynix shares rose 4.2% on 16 October, according to Bloomberg.
What Transfers
When an input is shared between AI servers and consumer products, the build-out rations it by price, and the cost lands on the buyers least able to pass it on. To see where the build-out's bill falls next, look for the next component that data centres and households both need.