To Anyone Still Pretending This Is a Normal Market
MEMO: To Anyone Still Pretending This Is a Normal Market
FROM: The part of your brain that reads the fine print
RE: The Microsoft-OpenAI divorce, Intel's ghost, and the consumer who hasn't broken yet
DATE: April 28, 2026
CLASSIFICATION: Read before you open your positions
Let's start with the thing nobody wants to say out loud.
Microsoft and OpenAI just tore up their exclusivity arrangement, dressed it in a joint statement full of words like flexibility and predictability, and handed it to the press on a Monday morning while oil traders were busy watching tanker counts in the Strait. The timing was either brilliant or desperate. Possibly both.
Here is what actually happened:
OpenAI has been bleeding enterprise deals because AWS and Google Cloud customers couldn't integrate its products without going through Azure. Demand on Amazon's cloud since OpenAI started selling there was, in the words of an internal memo reported by CNBC, staggering. Microsoft's exclusivity wasn't protecting a competitive advantage anymore — it was throttling OpenAI's growth and handing Anthropic the enterprise market on a platter. So they renegotiated. OpenAI can now sell to anyone. Microsoft keeps a non-exclusive license through 2032, stops paying revenue share on what it resells, and gets freed from having to build out all the datacentre capacity that OpenAI's ambitions required.
Microsoft shares dipped 1.3% on the news and then recovered to flat. Alphabet popped 1.5%. Read into that what you will.
What this actually means — past the press release — is that the most consequential commercial arrangement in the history of artificial intelligence just quietly pivoted from a bilateral monopoly into a multi-cloud arms race. Anthropic, Google DeepMind, and Meta's open-weight models all just got a little more room to breathe in the enterprise. Sam Altman gets his freedom; Satya Nadella gets his datacenter capital back. Everyone wins. Except whoever was pricing in indefinite Azure dependency.
Now hold that thought, and look at what Intel has done.
INTC hit a record high last week for the first time since August 2000. Twenty-five years. A quarter of a century of shareholder pain, of watching AMD eat your margins and Nvidia build an empire on your failures, of reading analyst notes that treated you like a cautionary tale for business school students. And then Q1 came in at $13.6 billion in revenue, EPS of $0.29 against a $0.01 consensus — not a beat, a demolition — and the stock went up 24% in a single session, its best day since 1987.
The Philadelphia Semiconductor Index ran for 18 consecutive days of gains. Eighteen. AMD jumped 14% in sympathy. Broadcom 11%. Nvidia, which had already retaken the $5 trillion market cap crown, added more. The SOX hit 10,000 for the first time. The VanEck Semiconductor ETF is up roughly 40% over those 18 sessions.
Let that sit next to this: the University of Michigan Consumer Sentiment index is at 49.8. The lowest reading on record. Below 2008. Below the pandemic lows. Below the post-Ukraine inflation shock. Year-ahead inflation expectations among survey respondents just jumped to 4.7% — the highest since September 2025 — with consumers citing the Iran conflict and $97 crude as the culprits.
So the people who make chips are having the best month in a generation. The people who buy things are at their most pessimistic in living memory. These two populations live in the same country, use the same currency, and are being priced by the same Fed, which meets today.
Something is going to have to give way, eventually. Markets keep deciding it won't be the chips.
The logic runs like this, and it is not crazy: AI infrastructure spending is a capital expenditure decision made by hyperscalers — Microsoft, Amazon, Alphabet, Meta — not by the consumer sentiment index. Households can feel terrible about the future and still generate exactly zero reduction in demand for H100s. Corporate capex cycles and consumer confidence have been decorrelated before; they're being decorrelated now, and violently, because the investment thesis for AI doesn't require a confident American consumer to remain intact. It requires continued quarterly earnings beats from the Mag Seven, which — with five of them reporting this week — we are about to find out about in real time.
JPMorgan has raised its year-end S&P 500 target to 7,600 from 7,200. Fabio Bassi wrote last Friday that markets remain "jittery but broadly resilient." Follow the winners, not the laggards. The structurally impaired stay impaired. The tech complex keeps printing.
Fine. But let's not pretend the consumer data is irrelevant. March retail sales of +1.7% month-over-month look strong until you notice that nearly 16% of the gain came from gasoline stations — people paying more for the same or fewer gallons. Strip out autos and gas and you get +0.65%, which is decent but unspectacular. The consumer hasn't broken. The consumer is paying $97-crude prices at the pump while feeling worse than they did during COVID, and still showing up to spend. That is not resilience. That is inertia. Inertia eventually runs out of road.
PCE prices drop Thursday alongside the advance Q1 GDP estimate. If the number comes in hot — and with gasoline station sales inflating the retail figures, it probably will — the market will need to decide whether it cares. So far in 2026, the market has repeatedly decided it doesn't. The Fed has been on hold at 3.50%–3.75% since the conflict started. No cut is coming today. Possibly none this year.
Kevin Warsh is presumably watching his Senate vote count. The DOJ probe into Powell got dropped Friday, which cleared the last procedural obstacle. If Warsh takes the chair by June, the new Fed era begins with oil above $95, consumer sentiment in the gutter, chips in euphoria, and a geopolitical negotiation being conducted via phone calls that may or may not happen.
One more thing.
Intel's 18-day SOX streak was already wobbling Monday. The semiconductor complex was set to snap its historic run. INTC was described as "partying like it's 2000" — and the dark irony of that phrasing clearly didn't register, because Intel was the second-largest stock by market cap at the dot-com peak. Investors who bought at that record in 2000 waited twenty-five years to break even. The stock returned roughly 0% nominal over that stretch while the S&P 500 compounded at 650% including dividends.
Entry price is still the only thing that matters. The story can be completely true and the stock can still be wrong.
That was 2000. This is probably different.
It usually is.
Published April 28, 2026