Anyone still holding semis at Thursday's close

TO: Anyone still holding semis at Thursday's close
FROM: The part of your brain that reads balance sheets
RE: What just happened, and why you knew it would
DATE: Saturday, 16 May 2026

---

Cerebras priced at $185. Opened Thursday at $311. Closed up 68% on its first day. Market cap at peak: north of $90 billion. Full-year 2025 revenue: $510 million. Operating loss: $146 million. That's a revenue multiple somewhere above 170x on a money-losing chip startup whose largest customer — OpenAI — accounts for a concentration risk the S-1 describes, in the careful language of lawyers who know what's coming, as material.

None of this stopped anyone. The deal was oversubscribed 20 times over. They upsized it twice. And in the same week that April CPI came in at 3.8%, the 10-year yield hit a one-year high, and the Strait of Hormuz remained closed to 97% of its normal shipping volume — the market decided that what it really needed was a $56 billion bet on wafer-scale AI inference.

By Friday, CBRS was down 2% in premarket. Intel had fallen 7%. AMD was off 5.7%. Micron 6.6%. The entire semiconductor complex spent the week ripping higher on AI enthusiasm, peaked at the Cerebras debut, and is now correcting into a macro backdrop that has no intention of cooperating. You couldn't have scripted the timing with more economy if you tried.

---

Here is the thing about the AI trade in May 2026 that nobody in the middle of it will say plainly: the earnings are real, and the valuations are insane, and both of those things are true at the same time. Cisco's Chuck Robbins called it a "networking supercycle" and the stock jumped 13.4% in a single session. He's probably not wrong about the cycle. The S&P 500's blended Q1 net margin just hit 13.4% — the highest FactSet has ever recorded. Information technology sector margins came in at 29.1%, up from 25.4% a year ago. The earnings power is documented, audited, and real.

But the forward P/E on the S&P is 20.9x, above both the five-year average of 19.9x and the ten-year average of 18.9x. And that multiple is being sustained by a market that spent most of this week pricing in a quick resolution to a conflict that, as of this morning, has a vessel seized near the UAE heading toward Iranian waters with no identified responsible party. The market wants to believe the geopolitical problem is temporary and the AI earnings story is permanent. That could be right. It is also the most convenient possible belief, which should at minimum prompt some epistemic caution.

---

The Empire State Manufacturing Index for May came in at 19.6 against a 6.2 estimate — a four-year high. Someone tell the macro bears. Don't tell the oil market; it didn't get the memo.

---

The Ford situation deserves more attention than it's getting.

This week, Ford launched a subsidiary called Ford Energy, formally pivoting away from electric vehicles toward selling battery storage capacity to AI data centers. The tech is licensed from CATL — the Chinese battery giant. Morgan Stanley valued Ford Energy at $10 billion within days of its announcement. Ford's entire market cap has been under pressure for two years. And now a car company, having failed to profitably manufacture EVs at scale, is being reborn as an energy infrastructure play for hyperscalers, using Chinese IP, in a war economy where supply chains to China are politically radioactive.

If you wrote that in a speculative fiction piece about 2026 in 2020, an editor would have asked you to dial it back.

But the logic is not absurd. Hyperscalers need power. Data centers need battery backup. AI training and inference runs around the clock. The grid can't keep up. Ford has manufacturing capacity, a balance sheet, and relationships. CATL has the best battery technology on the planet. The partnership exists in a regulatory gray zone because the licensing arrangement keeps CATL off the direct ownership structure while still putting their technology at the core of the product. Whether that survives congressional scrutiny — given that Hormuz is closed, China pledged not to arm Iran but offered no mechanism to enforce that pledge, and the Taiwan arms package is still unresolved — is a question that appears to be on no one's immediate agenda.

---

Now the part nobody wants to hear about Bitcoin.

CBRS closed at a $90 billion valuation on Thursday. Bitcoin fell back below $80,000 on Friday, down nearly 3% for the week. Coinbase lost 8%. Strategy dropped 6%. Circle, fresh off a post-earnings rally, fell 8% and flipped negative for the week.

Bitcoin is widely theorized to be the inflation hedge for a generation that distrusts gold. The gold story isn't going well either — spot gold has now fallen four consecutive sessions to around $4,558, getting crushed by the same rising real yields that are hammering long-duration equities. The 10-year yield at 4.58% with inflation at 3.8% gives you a real rate of roughly 80 basis points. Not spectacular, but enough to make holding a non-yielding asset less appealing when everyone is repricing rate-hike risk.

There's a structural crypto argument that survives all of this — scarcity, sovereign debt deterioration, dollar credibility under a new Fed chair who was confirmed 54–45 in the most politically divisive central bank vote in American history. That argument doesn't go away just because Coinbase sells off in a risk-off Friday. But the short-to-medium term behavior of crypto continues to be that of a leveraged tech proxy, not a monetary safe haven. The institutional flows that were supposed to change that have, in practice, mostly made the correlation to the Nasdaq tighter, not looser.

---

The retail data next week will matter more than people think.

Home Depot, Target, Walmart, TJX, Ralph Lauren, and VF Corp all report. This is your consumer read. Michigan consumer sentiment came in at 48.2 in the preliminary May print — a record low. The Conference Board measure has held up better because it weights labor market conditions more heavily, and the labor market remains, on its own narrow terms, resilient. But there is a widening gap between what consumers tell pollsters and what they actually spend, and at some point those two things converge. If the retail prints show demand destruction — not just in discretionary but in staples — then the inflation is doing its job in the ugliest possible way, and the earnings revision cycle for H2 2026 starts sooner than the models have it.

FOMC minutes drop May 20. Nvidia reports the same day. Two distinct readings of the same economy, priced by two entirely different market participants, arriving in the same twelve-hour window.

---

One last thing about Trump's financial disclosure.

Over 3,600 stock transactions between January and March 2026, involving up to $750 million in trades. Heavy exposure to Nvidia, Microsoft, Amazon, and Apple. Reportedly 20%+ profit on most positions.

I'm not going to say what I'm thinking. You're thinking it too. We can leave it there.

---

Anonymous. Not financial advice. Though at this point, what even is.

---

Closing Numbers · Friday Close

Calendar