Valuing AI Stocks in a 5% World With $112 Oil: Cash vs Debt

POET Technologies (POET) is $5.31 as we write this, down 6% on a day the whole AI tape is getting taken out behind the woodshed again, and we'll get to the why in a second. First, the number that didn't move: $796.3 million of cash and short-term investments on June 30. No bank debt. No bonds. No converts. Nothing sitting ahead of the common. At $5.31 that works out to $4.40 a share of net cash and 91 cents a share for the business. On Tuesday we ran the five-line formula at $5.38 and got 98 cents. Since then oil ripped to $112, CoreWeave's bonds sank to 79 cents on the dollar, and the optical group is getting smashed for the third time in three weeks. The formula didn't change. What changed is how much it matters.

Why the market is down today

A Houthi drone went after Saudi Aramco's Shaybah oil field on Wednesday. The interceptor caught it, the debris hit a contractor's vehicle and killed two people, and a second strike hit the airport in Dammam. Brent ran through $111 to its highest print since May, WTI pushed $98, and every inflation trade that had cooled off last week came right back on. The Qs $QQQ are down 2.5% as we write this and the semis iShares Semiconductor ETF (SOXX) are down more than 6%: Nvidia Nvidia (NVDA) off 4%, AMD Advanced Micro Devices (AMD) and Marvell Marvell Technology (MRVL) off 8%, Intel $INTC off 7%, and the Meta & Tesla prints did nothing to calm anyone's nerves about the size of the AI bill.

Then the twist. Treasury yields fell. The 10-year dropped about 9 basis points to 5.05% and the 30-year to 5.36%, because money ran into bonds as it ran out of stocks. Don't read that as relief. A 5-handle 10-year next to $112 oil is a growth scare stacked on top of an inflation scare, and the Fed meets next week.

If Hormuz is a decade, not a quarter

We wrote the origin story last week. The war in Iran wasn't a pebble in a pond. It was an asteroid hitting the Strait of Hormuz, and the fallout runs through oil, then CPI, then the Fed, then the bond market, then every AI multiple on your screen. Here's the part nobody wants to say out loud: no one can tell you when it ends. A ceasefire in April fell apart in July. Tankers got hit in October. Now drones are flying into the Empty Quarter and there's chatter about Washington and Israel going back into Iran after the midterms.

So run the AI trade on a different assumption. Not a bad quarter. A regime. Years. In that world the damage doesn't happen once, it happens every year. Oil stays high, so inflation stays sticky, so rates stay high. Anyone who needs to borrow pays more every time the paper rolls. Anyone who needs to sell stock sells it lower than last time. And the only question about any AI stock becomes the boring one. What did you actually buy? How much of the price is a business, how much is cash, and how much is somebody else's debt you inherited at the door?

Capital didn't leave. It got picky.

A bank credit desk put out a line this week that's getting passed around, and we'll give you our version. The money isn't gone. There's a mountain of it. It just stopped being dumb. For two years the AI trade paid the same multiple to the company that could fund its own growth and to the company that needed the financing window to stay open to fund it. That's over. The market is sorting the whole stack into two piles: the ones who can write their own checks, and the ones whose growth plan is a bet that somebody else keeps writing them. Broadcom's Hock Tan said as much out loud last week, that capital access could slow the buildout, and the big photonics names lost about a quarter of their value in a week. The private side got the same memo. Nscale is being told by its own IPO buyers that $30 billion is really $16 to $19 billion after Microsoft walked away from its Essex data center, Firmus scrapped its IPO outright, and Blue Owl reportedly couldn't find outside money for a $4 billion data center. Pile two is getting repriced. Pile one is getting ignored, and that's its own opportunity.

CoreWeave is the canary

CoreWeave $CRWV is pile two in its purest form. It borrows to buy Nvidia chips, rents the compute out, and needs the next loan to build the next data center. At the end of Q2 it carried about $52 billion of debt and lease obligations against a $104 billion backlog. The backlog is real. The interest bill is realer. This month the bond market decided it wasn't sure the first pays for the second. Its unsecured debt yields around 15%, some of its paper trades near 79 cents on the dollar, the cost of insuring that debt hit a record, and S&P has it at B+, which is junk. The stock closed at $91.72 on October 6. It's $56.74 as we write this, down 38% since then and down 8% on the day. Goldman upgraded it Monday. Nvidia, which owns about a tenth of it, said it doesn't expect to bail it out. The three co-founders have reportedly sold $3.3 billion of stock since the IPO. Q3 lands in mid-November.

A canary doesn't tell you the mine is collapsing. It tells you the air changed. Nobody in optical borrows like CoreWeave. But the rate that just hit its coupons is the same rate that hits everybody's multiple, and the lesson travels down the stack: the further your cash flows sit from today, and the more of them someone else has to finance, the more the bond market owns you.

The basket, sorted by what you're actually paying for

Here's Tuesday's formula run on the names POET gets lumped in with. Balance sheets are the latest filed. Prices are Thursday's tape as we write this, so round them.

Coherent $COHR. About $2.0 billion of cash against $3.2 billion of debt. That's net debt of roughly $6 a share, so at $208 you're paying about $214 of business for a $208 stock. Cash covers five cents on the dollar. Coherent earns it, with $7 billion of revenue last fiscal year, but you inherit the debt at the door. Down 12% as we write this.

Lumentum $LITE. $2.7 billion of cash, $1.6 billion of convertible notes, and $2 billion of convertible preferred sitting ahead of the common. The business is about 101% of the price. Sold out through early 2029, booked solid, and still down 10% Thursday. That's what a discount-rate shock looks like when it lands on the best name in the group.

Credo $CRDO. $764 million of cash and no borrowings on the balance sheet. Cash covers a couple of cents on the dollar and nobody cares, because Credo is pile one. It wrote a $750 million check for DustPhotonics out of its own pocket, still had that $764 million left, and threw off $83 million of free cash flow in the quarter. Insiders are selling. The company isn't.

Applied Optoelectronics $AAOI. $509 million of cash against $93 million of debt and $129 million of converts. Net cash of about $3 a share on a $69 stock, so cash covers eight cents on the dollar. This is the name in the group whose growth depends on financing staying available. It spent $565 million on capex in a single quarter, raised more than $1 billion through at-the-market stock sales in the first half, and opened a second $600 million ATM in August. From $233.67 at the high to $68.79 as we write this, down 11% on the day. That's not a demand story. That's the window.

Lightwave Logic $LWLG. $96 million of cash, no debt, 154 million shares. At $4.08 the market pays $3.45 of a $4.08 stock for a polymer story with a 2027 ramp. Cash covers 15 cents. Down 8% today. It reports November 12.

Aeva $AEVA. $178 million of cash and securities, plus a $125 million facility it can draw by issuing preferred shares. Burn runs $31 million a quarter, so the cash alone is under six quarters. It sold $115 million of stock at $22.25 in June and closed at $13.69 on October 8. Cash covers about a fifth of the market cap.

POET $POET. $796 million of cash. No debt. Cash covers 83 cents of every dollar of market cap at $5.31.

Line them up by what the market makes you pay for the business as a share of the stock price: Coherent 103%, Lumentum 101%, Credo 98%, Applied Opto 96%, Lightwave Logic 85%, Aeva about 80%, POET 17%. One of these is not like the others. You can likely guess which one we think is mispriced, and we're paid by them, so run the lines yourself.

POET: $4.40 of cash, 91 cents of business, $0 of debt

The formula one more time, at $5.31. Market cap: $5.31 times 173.0 million basic shares is $919 million. Net cash: $796.3 million of cash and short-term investments minus every liability on the books, $34.3 million, is $762 million. That's $4.40 a share. Enterprise value: $919 million minus $762 million is $157 million. So the business is 91 cents a share. Count the warrants, options and RSUs and the business runs to about $1.20. Still mostly cash. And most of the $34 million we subtracted is a Black-Scholes mark on warrants that never costs POET a dollar; strip it and the business gets cheaper still.

What does $157 million buy? Forty-nine granted patents and 37 applications. The Optical Interposer platform. A $50 million initial purchase order from Lumilens inside a framework that runs to $500 million, from a customer that raised $700 million while it was still in stealth. Named engagements with Mitsubishi and Lessengers. More than 150 people in six countries and a capacity build in Penang targeting up to a million units a month by the end of 2027. Volume shipments into the AI data center market are expected, not yet underway, and we'll say that every time. But the market's sticker price for all of it is $157 million. In May the same patents, the same customer and less cash went for $20.81 a share.

Burn, capex & inflation: the clock inside the cash

Cash is only a hero if it outlasts the war, so let's inflate the war. POET's deck projects about $15 million a quarter of operating burn and about $60 million of 2026 capex. Only $5.4 million of that capex had gone out the door by June 30, so the back half is the heavy half. Call it $85 million of burn and capex in the second half, less whatever the pile earns, and you land near $725 million at year-end. That's still $4.19 a share.

Now the inflation part, because nobody else is doing it. If 3.7% CPI is the regime, a $60 million-a-year operating burn is $62 million next year and $64.5 million the year after, with nobody new hired. The capex line is worse, because the stuff POET buys is the stuff every AI company is buying. The CPI line for computers and related gear is up 9.7% from a year ago, by one count we used last week, which means a $60 million equipment budget buys about 9% less than it did. Inflate the whole $120 million-a-year tab at 3.7%, give the cash nothing, and $796 million still covers about six years.

And the cash isn't earning nothing. At a 4% money-market rate, which is under where the 2-year trades, $796 million throws off about $32 million a year. Q2's other income was $4.3 million, with the $400 million in the bank for only six weeks of the quarter. So in a 5% world the pile roughly keeps pace with 3.7% inflation and pays about half the operating burn while it waits. CoreWeave pays 15% to wait. Applied Opto pays in shares. That's the whole difference between pile one and pile two, and in a years-long regime it compounds.

What would make us wrong

Cash isn't a floor. Cash-rich companies trade below cash all the time when the market decides the cash is going to get burned, and POET hasn't shipped at volume, so there's no gross margin to point at yet. Qualification, margin at volume and repeat orders are the three hurdles, and the cash only answers the first one by letting POET sit through the cycle. The 27.4 million warrants struck between $5.94 and $7.12 are supply every time the stock pushes back through $6 and $7. And the September 30 cash balance is a Q3 number, not a June number. The print is Monday, November 16 at 7:00 a.m. ET, no call, with POET's first conference call pegged to the full-year results on March 1, 2027.

What we're watching

One: Brent above $110 into the October 28 Fed keeps the hawkish branch alive. If oil holds there, the statement keeps December live and the long end stays where it is.

Two: CoreWeave's bonds decide whether the canary is dead or just coughing. Paper near 79 cents climbing back toward 90 says the air cleared. Sliding toward 70 says pile two gets repriced again, and the small caps bleed first.

Three: the September 30 cash balance on November 16 tests the deck. A number near $750 million says burn and capex ran at the deck's pace and this formula barely moves. Well under that, and we'll ask what the money bought.

Four: Coherent reports November 4, Lumentum and Applied Opto on November 5. Those are the first prints of the picky-capital era, and the capex commentary inside them tells the whole photonics stack whether the checks are still coming.

Subscribe to StockAlpha.ai and we'll re-run every one of these lines the morning of November 16 with the September 30 balance sheet.