Mid-Year Check: How Much of AI's Growth Is Just Inflation?
Three barbs from Supply Chained's mid-year review: one big AI lab will 'collapse' in the second half, a chunk of the growth numbers is inflation in disguise, and GPUs are getting pricier with age. Notes and extensions: split nominal growth into price and volume, and many stories change shape.
Plenty of shows do mid-year reviews. Few open one with a hard prediction. Supply Chained’s latest episode (August 11, 2026, hosted by Jon Y of Asianometry and former Bloomberg tech columnist Tim Culpan) starts with Jon’s call for the second half: “I think we’re going to see one big AI lab collapse.”
Here are the three barbs I took from the hour, plus — as usual — the homework they pointed at but didn’t do.
One lab goes down, but not the way you think
Jon doesn’t mean bankruptcy. His chain of reasoning: AI labs are now a year or two into shipping. Models are out, revenue exists — and the side effect of revenue existing is that insiders and investors can finally tell who’s actually behind. Once a model becomes a product, it needs sales, marketing, advertising — recurring burn not every lab can carry. When monetization can’t catch up with valuation, the dignified exit is a sale, or something dressed up as one. Tim asks whether the lab gets rescued or put down. Jon’s answer: “humanely euthanized.”
The prediction has an obvious weakness: no time window, no observable, unfalsifiable after the fact. So I built my own set of leading indicators to track it: funding rounds spacing out, API price cuts turning into a price war, senior researchers leaving in batches, contracted cloud compute quietly being resold. When those four show up at the same shop at the same time, the countdown has started.
The inflation hiding inside the growth numbers
The most valuable thread in the episode is a question Tim keeps jabbing at: how much of this AI boom is volume actually growing, and how much is things simply getting more expensive?
His examples (as told on the show): TSMC raised its full-year revenue growth forecast from about thirty to about forty percent — but a good chunk of the capex increase is equipment getting pricier. Foxconn’s business model is buying components and selling them onward, so component inflation flows straight into its revenue line. Downstream, hyperscaler capex climbs every quarter — but are they buying more compute, or the same compute with data centers, power, and cooling all costing more? The hosts’ honest answer: you can’t split it yet; watch the next few quarters.
Can’t split it yet doesn’t mean you shouldn’t try. Revenue equals price times volume — the first lesson of financial analysis, and the first thing forgotten when a narrative runs hot. Split every “forty percent growth” into a price line and a volume line and many stories change shape. The show points at this and stops; the actual work is doable: wafer shipments against blended wafer prices, server units against per-rack prices, capex dollars against megawatts actually connected to the grid.
GPUs aging into higher prices is an anomaly, not a regime
Third barb: by twenty years of habit, a GPU six months after launch should be older and cheaper. Right now they appreciate after launch. Moore’s law, briefly flipped.
Tim’s stance is strong enough to quote: if someone launched a fund betting that same-model GPU prices keep rising quarter after quarter, “I would short you.” He reads it as a blip from mismatched supply and demand, not a trend to extrapolate. The episode adds a lovely supply-chain mechanic along the way: a hyperscaler that wants to slow down installation just tells its system integrators to hold shipments, and the signal squeezes back up through module makers and board makers, buying a quarter or two of financial breathing room. The built-in penalty: inventory that sits too long gets lapped by the next-generation chip.
One thing to take with you
The idea worth keeping: nominal growth has to be split into volume and price. The same “up forty percent” is two entirely different stories depending on which drove it — one is demand actually growing, the other may be inflation wearing growth’s clothes.
A practical exercise: take the growth number you trust most — your stock’s revenue growth, your own salary bump, your neighborhood’s home prices — and split it once into volume and price. Salary up five percent: what did prices do over the same stretch? Revenue up forty: how much was units, how much was pricing? You may find that some things you thought were getting better were only getting more expensive. Telling those two apart is where judgment starts.
This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.