The AI Trade vs. the Tape: Bonds, Hormuz, and the Buildout

August has been loud. The long end of the Treasury market broke to levels last seen before the financial crisis. Crude just booked its second straight weekly gain on the Iran standoff. And the AI names whipped around with all of it.

Here's the part that matters if you own the AI trade: almost none of this volatility is about AI demand. Some of it is actually caused by AI spending. That distinction is the whole recap.

The long end woke up

Start with bonds. The 30-year yield pushed above 5.33% on Tuesday, its highest since June 2007. The 10-year tagged 4.75%, a 20-month high. The drivers aren't mysterious. July's deficit was the largest monthly shortfall since March 2021, inflation is still running above target, and war-priced oil isn't helping. Treasury blinked midweek and said it would double its long-bond buybacks. Yields dropped 10 basis points on the headline. By Thursday the whole move had round-tripped.

And there's a wrinkle that belongs in an AI recap. One of the forces pushing term premiums higher is record corporate borrowing to fund AI data centers. The buildout has gotten big enough to move the Treasury market. That's not a bug in the AI thesis. That's the size of it.

Hormuz keeps a bid under oil

The other volatility source floats. The June ceasefire collapsed, Washington says its blockade of Iranian ports can run indefinitely, and Tehran says the strait stays closed. Earlier this month Iran hit two ADNOC tankers in the strait. This week the UAE said it came under Iranian missile fire and cut off all trade with Tehran, and the U.S. teed up what it's calling the toughest sanctions in history. Brent traded near $94 into Friday, up more than 5% on the week. Diesel cracks printed records. Oil at these levels feeds the inflation math, and the inflation math feeds the long end. The loop closes on itself.

When semis sold off this month, that loop was the seller. We covered it as it happened: why chip stocks were down, and why it wasn't the AI trade. Rates and tankers did the selling. The order books didn't move.

The buildout doesn't blink

Against that tape, look at what the buyers of AI infrastructure actually did in August. Nvidia $NVDA signed agreements with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to stand up financing platforms targeting more than $500 billion of third-party capital for AI factories. Jensen Huang's pitch is that compute is now an investable asset class, financed the way the market finances pipelines and towers.

Amazon $AMZN went the other direction and just wrote the check itself. The company raised its 2026 capital spending to roughly $220 billion, said AWS grew at its fastest pace in 18 quarters, and told investors it will still be capacity-constrained through 2027 against demand that already stretches into 2028. The increase was pinned largely on memory prices, the same squeeze that has Micron $MU sitting in the middle of the AI trade. Read those two together. The industry isn't debating whether to build. It's inventing new ways to pay for it.

The next bottleneck is light

Every dollar of that capex runs into the same two walls: power and interconnect. Scale-out AI clusters have to move staggering volumes of data between racks, and copper runs out of physics at the distances and speeds now being deployed. That's why the money is moving to photonics. Industry estimates put optical connectivity for AI at $25 to $30 billion today, growing around 60% a year, with forecasts reaching $80 to $100 billion by the end of the decade. The public market already trades the theme through optical names like Coherent $COHR.

The private market has done its own math too. Lumilens, an optical connectivity company that came out of stealth this month at a $5.51 billion valuation, already has a multi-billion-dollar hyperscaler agreement and product moving into live AI data centers.

The quiet way in

Down the supply chain from all of that sits POET Technologies $POET. Lumilens, a company that raised $700 million while it was still operating in stealth, turned around and placed a giant order with POET: an initial $50 million purchase order for its optical engines, under a framework that can top $500 million over five years. POET just printed its sixth consecutive quarter of sequential revenue growth, up 112% from a year ago, with $796.3 million in cash and short-term investments and a September update on the production ramp already committed. Small company, big neighborhood.

The tape will keep trading tankers and term premiums. The buildout keeps ordering parts anyway. The edge is knowing which parts come next.