The argument over whether AI chip stocks are in a bubble got much louder on Monday, September 21, 2026. AMD closed above a $1 trillion market value for the first time, the S&P 500 set a record high, and 30 companies inside that same index fell to fresh 52-week lows. Bulls point at profits. Bears point at 1999.
Four separate things happened in one week, and most of the coverage picked exactly one of them. So here they are in one place, in plain language, with no buy-or-sell call attached. All figures are as of September 22, 2026.
Two rallies, one open question: profits or promises? (AI-generated image)Why AMD hit $1 trillion this week
AMD became the fourth US chipmaker valued above $1 trillion on September 21, 2026, following Nvidia, Broadcom and Micron. The trigger was software rather than silicon. Meta's personal AI agent Muse, launched in the US on September 8, reached No. 1 on Apple's US App Store free charts within 13 days, and traders read that as a demand signal for a very specific kind of chip.
The link is less obvious than it sounds. An agent that books a table, checks three tabs and chains small tasks together is doing ordinary general-purpose computing — CPU work — and AMD's EPYC server franchise lives there.
Training a giant model eats GPUs. Running millions of tiny agent steps eats plain processors. Bloomberg's report on the day tied the move to that exact read.
That split shows up in the share prices too. Nvidia still holds an estimated 81% of the AI chip market, yet through May 2026 its stock was up about 18% year to date while AMD's was up roughly 114%. The rally stopped being a one-company story some time ago.
The industry backdrop isn't imaginary either. Global semiconductor revenue is projected to grow 64% in 2026 to $1.32 trillion on AI processor and networking demand, and the PHLX Semiconductor index is up roughly 70% year to date.
The 1999 signal that flashed the same day
Under the record close, the market was much thinner than the headline suggested. On September 21 the S&P 500 hit an all-time high while 30 of its members touched fresh 52-week lows against only 7 new highs — the fourth time in market history a record high has coincided with 5% or more of the index at new lows, CNBC reported.
The three previous dates were July 1929, January 1973 and December 1999 — the last months before the dot-com crash. Three data points across a century is not a prediction, and nobody should read it as one. It is a strange kind of company to be keeping.
Just over half of the index was also trading below its 200-day average on that same record day. Translated out of trader-speak: a small handful of enormous stocks was carrying the scoreboard while the rest quietly slipped.
Michael Burry, the investor from The Big Short, disclosed short positions against Nvidia — opened near $198.09 — plus Tesla, Palantir and the iShares Semiconductor ETF, warning of a possible "1987-type" fall. Options tied to those bets reportedly made up as much as 80% of his roughly $1.1 billion fund. That's a very concentrated wager, not a forecast with a date on it.
Bar chart of the four dates when an S&P 500 record high coincided with 5%+ of members at 52-week lows — July 1929, January 1973, December 1999, September 2026Is the AI chip stock rally a bubble?
There is no settled answer as of September 22, 2026, and the two sides are arguing about different things. The bull case rests on earnings that already exist: Nvidia runs roughly a 53% net margin, chip revenue is climbing, and most 2026 AI capital spending has come out of operating profit rather than new debt. The bear case rests on concentration, thinning breadth and circular deals.
| Measure | 2026 AI rally | 2000 dot-com peak |
|---|---|---|
| Lead company P/E | Nvidia ~44x | Cisco ~472x |
| S&P 500 long-term earnings growth expectation | 20.2% | 18.6% |
| AI megacaps' share of S&P 500 value | ~40–45% (Nvidia alone 8%+) | about 14 points lower |
Two of those rows deserve an asterisk. The 44x-versus-472x comparison and the 40–45% concentration estimate come from single analysis sites rather than primary filings, so read them as direction, not decimals. The earnings-expectation row is from Fortune, and it's the one that deserves the most attention — analysts expect faster long-term growth today than they did at the 2000 top, which is either evidence of a real boom or evidence that optimism is doing the same job it did then.
Where the bubble debate stands now
The debate has narrowed to one question: is this buildout paid for with profits or with promises? Nvidia CEO Jensen Huang, in a CBS interview reported September 19–21, 2026, put a "0% chance" on AI ending the world by 2030 and called extinction warnings "not grounded in science" and "irresponsible."
Plenty of people who move real money disagree about the market, if not about the technology. JPMorgan's Jamie Dimon compared today's sentiment to 1972, 1986, 2000 and 2007 back in May 2026. Ray Dalio said in June that bubble indicators were "approaching 1929 and 2000 levels."
Then there's the financing. Nvidia has discussed guaranteeing up to $250 billion for an OpenAI data-center project in Ohio, with a $105 billion guarantee reportedly signed in August 2026. AMD granted OpenAI warrants for up to a 10% stake at $0.01 per share as part of a multi-gigawatt supply agreement.
The number we'd watch isn't a valuation multiple. It's that warrant — a chip buyer holding an option on its own supplier's share price. That's the thread part 2: circular AI deals explained pulls on.
At a glance
- Who this is for: readers trying to follow AI chip headlines, not traders looking for a signal.
- When: the week of September 21–22, 2026, which is when four separate stories collided.
- The numbers that matter: AMD above $1 trillion, a breadth warning last seen in December 1999, Nvidia near 44x earnings against Cisco's ~472x in 2000.
- One caution: the bull and bear evidence are both real and both partial, and none of it says what any individual stock does next.