AI's loudest voices spent the week telling everyone to slow down, and the market took them at their word. AI infrastructure names sold off into a month that traditionally is already unkind to stocks.

Jason Bodner, co-founder of quantitative research firm MoneyFlows, thinks investors are reacting to the wrong headline. The one that matters came out of NVIDIA's NASDAQ: NVDA August earnings call, where CEO Jensen Huang guided to roughly 70% revenue growth next fiscal year and then told analysts that demand is running well above that.

That gap is the whole story. AI's constraint right now is not appetite. It is the physical hardware that moves data around inside a data center, and NVIDIA named the two pinch points: memory and silicon photonics. Bodner's three picks sit on top of both.

Get Micron Technology alerts:

Slowing the Models Does Not Slow the Buildout

Anthropic CEO Dario Amodei argued in a Sept. 12 essay that AI companies should deliberately slow the pace at which they improve model capabilities. OpenAI's Sam Altman endorsed the idea hours later; Elon Musk backed it too. Meanwhile, a researcher who had worked at both Anthropic and OpenAI had resigned days earlier with a public warning about existential risk.

Bodner reads the corporate half of that very differently than the tape did.

Regulation, in his view, hands incumbents licensing regimes, capital minimums and a moat that smaller competitors cannot clear. It also makes AI clean enough for health care and government procurement. His analogy is the airline industry, which lobbied for a federal regulator and got mass adoption rather than grounded planes.

The macro backdrop gave sellers cover. The Federal Reserve raised rates on Sept. 16 rather than cutting, oil is sitting near $100 a barrel on U.S.-Iran escalation, and September has a long history of chewing up gains.

What is not deteriorating is profit. Second-quarter beat rates across the S&P 500 ran well above their five-year average, and earnings growth landed at its strongest level in years even after stripping out a couple of outsized outliers. That is a sentiment problem, not an earnings problem.

Micron Owns the Memory Shortage NVIDIA Keeps Naming

Micron Technology NASDAQ: MU is the fastest-growing producer of high-bandwidth memory (HBM), the stacked chips that sit beside a graphics processor and feed it data. Bodner's framing: a GPU is the world's fastest chef, and HBM is what keeps the ingredients on the counter instead of scattered across town.

Management has said industry demand continues to significantly exceed supply, and that tightness could persist well past next year. Micron's HBM output has been selling out ahead of production, and analysts have chased that reality with steadily higher forward estimates. Bodner's point is that a stock can be up several hundred percent and still look cheap if the forward numbers are moving faster than the price.

NVIDIA is effectively underwriting the shortage. Its supply commitments more than doubled in a single quarter, primarily tied to memory procurement. Fabs do not go up overnight, capacity is spoken for, and pricing power follows.

Coherent Sits Where Copper Runs Out

Copper wiring is fine at the chip level, but it's ugly across a data center. Resistance, heat and distance all pile up, so the AI industry is shifting from copper to optical connections. That transition is silicon photonics, and Huang has called it the bottleneck of the coming decade.

Coherent Corp. NYSE: COHR supplies the lasers, transceivers and optical materials underneath it. NVIDIA is investing $2 billion in Coherent alongside a multibillion-dollar purchase commitment and future access to its capacity. Lumentum Holdings NASDAQ: LITE got an identical commitment the same day, which tells you NVIDIA is buying insurance on a supply chain it expects to strain.

The stock has taken a hard pullback with the rest of the AI complex. Bodner's read is that the group has already absorbed most of its selling while energy, health care and gold rallied.

Tower Semiconductor Makes the Wafers Photonics Runs On

Tower Semiconductor NASDAQ: TSEM is the pick-and-shovel layer. It is a specialty foundry that manufactures wafers that other companies design on, which means an investor does not have to guess which transceiver vendor wins.

Its silicon photonics business has gone from a rounding error to the engine of the company, and management raised its long-term targets on the back of it. Tower has locked in more than a billion dollars of 2027 photonics contracts with customer prepayments already in hand. That is demand booked before the capacity exists.

What Could Break the Setup

The doomsday case is that regulation kills AI demand. Bodner does not buy it, and adoption data is on his side.

The real risk is capital intensity. Tower is committing billions of its own money to a Japan expansion, and heavy capital expenditure can outrun actual demand if photonics adoption slows. Memory has broken hearts before. These three trade on the assumption that tightness persists, and the same forces that lifted the group can unwind it together.

Watch supply commitments and customer prepayments rather than safety headlines. That is what is actually moving these three.