1Tech's lead over the index reached a record extreme in June, and the give-back since is the whole argument
On 2 June, tech — tracked here by — had beaten the S&P by 29 points over the previous 50 days. Jessica Colas, of Data Trek Research, speaking on The Compound's "What Did We Learn", puts that at six standard deviations and the most extreme reading in her dataset. Since that date tech is down 6.3% against the S&P's 0.6%, so roughly a fifth of the extension has come back. She frames the unwind against December 1999 through April 2000, which is a note on what a lead this wide has done before, not a forecast.
Today's version of it is concentrated in memory. is down 3.5% and the most active name on the tape, on what the coverage calls a reassessment of the AI trade. is in what Benzinga described as DRAM panic selling, is falling with it, and is lower alongside them even after announcing a five-billion-euro fab in Ireland. SK Hynix, off 8 to 9% and the epicentre of the move, is a Korean listing. But the selling is not tech-wide: is up 5.96% pre-market. Memory hit while equipment is bid is the shape of an idiosyncratic unwind, not of a demand shock.
Two readings of that unwind sit inside one video. Katie Stockton, the technician on Excess Returns, says the ratio of the Mag 7 to the S&P has traced a bearish head and shoulders, while in absolute terms this is a run-of-the-mill correction — leadership narrow and confined to semis, momentum stalling, but breadth and participation holding. Jim Paulsen, on the same show, has turned more cautious: he expects a 10 to 20% correction over the next several months, with tech down more than 20% and the broad market around 10%, while staying a long-term bull into the 2030s and saying he would underweight mega-cap tech in favour of the average stock. S&P tech is already down 10%.
Underneath both readings sits Data Trek's frame that this economy is recession-resistant rather than recession-proof: roughly sixteen years without a US recession through repeated shocks, a services economy that burns less energy, a policy response that has always arrived, and a deficit near 6% of GDP underwriting baseline demand. Steady economy, steady earnings, and therefore a market that can carry 20 times earnings with credit spreads in their first percentile and a VIX under 20. Polymarket agrees on the recession question, pricing a US recession by end-2026 at 10% on $1.66m of volume. So the drawdown in tech is being priced as a correction inside a cycle rather than the turn of one — a reading that only fails if earnings break.
2Inflation's monthly run-rate is close to zero while its annual level is not, and the rate market is betting on the level
June CPI resolves today, and Polymarket's two contracts on that single print say things that sound opposed. The annual contract has 3.8% as its modal bucket at 49% implied, with 3.7% at 24% and 3.9% at 26%, and anything at or below 3.6% at just 2%, on about $799,000 of volume. The monthly contract has 0.1% or lower at 98% implied, on about $39,000. A monthly rate of 0.1% annualises to roughly 1.2%. So the expected headline near 3.8% — nearly double the Fed's 2% target — is assembled out of base effects rather than current momentum.
The rate market has picked the level over the run-rate. Polymarket's "Fed rate hike in 2026" contract is the deep one at $3.97m of volume, and it sits at 70% yes, with the funds rate at 3.63%. Yahoo's markets coverage carries the same thing as traders lifting hike bets. Warsh gave his first Humphrey-Hawkins testimony this week, and Data Trek's read is that he inherits an economy that is resilient and inflation-prone at the same time — which is the combination that argues for a hike rather than a cut.
Energy is the loose variable, and the administration is on both sides of it. President Trump wrote that the Strait of Hormuz is open while reinstating a blockade aimed at Iran's own ships and customers, which tightens Iranian supply while removing the broader tail risk of a closed strait. In the same window he was claiming falling oil and gas prices as a win. Cheaper energy would take pressure off the level; a squeeze on Iranian barrels would add to it, and the two posts point opposite ways.
The consequence is worth stating plainly, because the two contracts get conflated. If the monthly print lands at or below 0.1% while the annual sticks near 3.8%, then the 70% hike bet is a bet that base effects hold the annual number too high to cut into — not a bet that inflation is re-accelerating. Those are two different arguments wearing the same number.
3The AI build-out is being funded in the bond market, and the supply is holding investment-grade tech yields near 6 to 7 percent
Jeff Klinghoffer, the bond manager on Excess Returns, sizes roughly $600bn of AI capital spending coming from about four companies, and puts that at approximately the whole of America's 3% growth. Take his arithmetic at face value and one sector's building programme is the growth number.
The funding shows up in his own market. Hyperscaler bonds — , , — yield 6 to 7% on paper he rates high quality, with spreads held wide by the sheer volume of issuance rather than by any doubt about the issuers. He notes SpaceX tapping the debt market about a week and a half after its listing, which is a fast turn from equity raise to bond raise.
That splits the AI argument across two markets. Equity investors are debating whether the spending earns a return; bond investors are separately repricing what it costs to finance. If issuance keeps spreads wide, the cost of the build rises before the demand question is settled at all — and it shows up in the yield first, in the multiple second.
4The market is at a 150-year valuation extreme without the froth readings that marked 2000 and 2021
Ross Givens, on his own channel, lays the valuation case out in three numbers: the Buffett indicator at 219% of GDP against 146% at the dot-com peak, a CAPE of 41.66 that he calls the second-highest in 150 years, and the S&P at three times sales. His own conclusion is that valuation forecasts returns rather than timing them, so what it buys you is a leaner next decade, not an exit date.
He is not selling. His euphoria gauge — the share of stocks that double within a quarter while in an uptrend — reads 1.1%, against 6.5% in 2000 and 5.9% in 2021. On that basis he puts this in the seventh or eighth inning with the blow-off still ahead, and says he will stay long, sell some into strength and lower his return expectations. For protection he sketches a long-dated index put on expiring December 2027 at about 2.8% a year, or a collar capping roughly 13% of upside against 10% of downside for a small net credit. Worth knowing that the same video pitches a paid subscription several times; the promotion is disclosed and makes no fabricated return claims, but the channel is selling something.
The IPO calendar is his eighth-inning marker, and it is the one piece of evidence with a second source today. SpaceX came public at $1.77trn, the largest listing ever, popped 19% and faded; OpenAI has filed at around $1trn and Anthropic at $965bn; his read is insiders selling at record valuations. Prof G Markets takes the same listing apart from the other side. SpaceX now carries roughly 18 or 19 buy ratings against one or two holds and sells, the stock is falling anyway, and retail buyers who came in between about 200 and 225 are underwater. Their argument is that the rating skew comes from the underwriting banks, which had a financial interest in the deal — so the ratings say more about who wrote them than about the company.
A record price with a subdued froth reading is an awkward pairing, and it is why the day's most bearish numbers arrive with a stay-long conclusion attached. The two only reconcile if the froth is still ahead. If it is not, the valuation number is what is left.
The inflation argument gets its number before the open
June CPI is the day's driver and the VIX is already up 14% to 17 going into it. The two contracts above set the test: a monthly print at or below 0.1% alongside an annual near 3.8% would confirm that the 70% hike bet rests on base effects. Second-quarter bank earnings start this morning before the bell — with revenue expected near $50bn, with earnings per share expected near $14.50, plus , and — and they are the first direct read on credit and the consumer behind the recession-resistant framing in the first section. , , , and follow on Wednesday, then , , , and on Thursday. A presidential "Speech to the Nation" is scheduled for Thursday at 9pm Eastern.
