Four days ago, I published 500 to 1 — the ratio of insider selling to buying across AI-hardware. Sixty-three sales, one purchase. But that headline number conflates two completely different kinds of selling. This post separates them.
The Distinction That Changes Everything
Rule 10b5-1 plans let executives schedule stock sales months or years in advance. Once set, they execute mechanically — rain or shine, top or bottom. A CEO selling under 10b5-1 is like an alarm clock going off. It tells you nothing about what they think right now.
Discretionary sales are different. No pre-set plan. The executive decides, that week, to sell. They sign the Form 4. They choose the moment.
One is a clock. The other is a decision.
The Audit
| Name | Company | Sold | 10b5-1? | Verdict |
|---|---|---|---|---|
| Gary Dickerson (CEO) | AMAT | $104.7M | NO | DISCRETIONARY |
| Omkaram Nalamasu (CTO) | AMAT | $14.4M | NO | DISCRETIONARY |
| G. Raja Prabu (President) | AMAT | $31.6M | NO | DISCRETIONARY |
| Mark Stevens (Director) | NVDA | $186M | Unclear | LIKELY DISCRETIONARY |
| Chris Koopmans (COO) | MRVL | $2.8M | Unclear | UNDER REVIEW |
| Confirmed programmatic (10b5-1): | ||||
| Jensen Huang (CEO) | NVDA | ~$226M | YES | PROGRAMMATIC |
| Lisa Su (CEO) | AMD | $57.6M | YES | PROGRAMMATIC |
| Akash Palkhiwala (CFO) | QCOM | $0.5M | YES | PROGRAMMATIC |
Dickerson Sold Both Rallies
This is the detail that elevates the signal above noise. Applied Materials' CEO didn't sell once. He sold twice, at two separate peaks:
A 10b5-1 plan can't do this. It can't sell at $590, watch the stock rally 25% to $735, and sell again. That requires a human being looking at a screen and making a decision. Twice.
What the Audit Reveals
When you strip out the programmatic sellers, the discretionary picture is starker than the headline number suggests:
Confirmed discretionary selling across AI-hardware: ~$340M+
AMAT ($150M) + NVDA Stevens ($186M) + MRVL Koopmans ($2.8M) + others pending audit
Confirmed discretionary buying: $374K
One director, one company (AVGO), one day.
The programmatic sellers — Huang, Su, Palkhiwala — are background radiation. They'd sell whether the stock was at $100 or $1,000. They tell you these executives own a lot of stock. They don't tell you what they think.
The discretionary sellers are the signal. Dickerson chose to sell at $590, then chose again at $735. No algorithm made that decision. And four trading days later, the stock was at $608.
The Chip Selloff Confirms the Timing
Between Dickerson's second sale (June 30) and the July 3 holiday close:
- AMAT: -17.3% (from $735 sale price to $608)
- MRVL: -10.3% on July 3 alone (COO sold July 1)
- Samsung: -9.1%, SK Hynix: -14.6% (KOSPI July 2)
- SOX index: worst week since June 23 circuit breaker
The catalyst? Meta signaled it's selling excess AI computing capacity — raising questions about whether chip demand has peaked this cycle. The people who build the machines that build the chips appear to have known.
What This Means
500 to 1 was the ratio. This post is the quality adjustment. Not all of that 500 carries equal weight. The 10b5-1 sellers are noise — scheduled, mechanical, uninformative. The discretionary sellers are the people who looked at their stock, looked at the market, and pressed sell.
At AMAT specifically: the CEO, the CTO, and the President all sold discretionarily within the same two-week window. No pre-arranged plan. Combined $150M. The stock is now 17% below the CEO's last exit price.
Still monitoring: MRVL CFO Meintjes' 207K-share exit filing (90-day execution window through September). NVDA Director Stevens' 10b5-1 status unconfirmed. SK Hynix Nasdaq ADR listing July 10 may generate new filing data.