For the past week, the market treated CoreWeave like the AI trade’s weakest link.

Every headline was about debt, concentration risk, and a 25% drop in five sessions.

Then Goldman Sachs called it a buying opportunity, raised its price target to $139, and investors started piling back in.

And honestly? The bond market isn’t buying any of it.

CoreWeave’s own 2032 notes are trading below 80 cents on the dollar, yielding close to 16%.

That’s not a market pricing in a bottom. That’s a market pricing in stress.

Today I’m putting Goldman’s buy case against the bond market’s fear and seeing which one actually holds up.

Missed Monday’s edition on TSMC’s post-earnings selloff? 😳 TSMC Just Beat Every Number And Still Fell

South Korean AI infrastructure company Elice plunged about 32% on its market debut despite heavy IPO demand, as investors questioned whether expensive data-center spending will generate enough returns. The weakness also hit other AI infrastructure names and dragged on Korean tech stocks. The bigger takeaway is that investors are no longer rewarding anything with an AI label. Debt, valuation, and actual profitability are starting to matter much more.

Alphabet because it can benefit from both OpenAI and Anthropic without investors paying near-trillion-dollar private-market valuations. Google Cloud provides compute to both companies, while Alphabet also owns a significant stake in Anthropic and continues building its own Gemini ecosystem. In plain English, Alphabet offers diversified AI exposure with an already-profitable core business and a much more reasonable valuation.

Berkshire Hathaway invested another $330 million in Bending Spoons at $33 per share, lifting its total commitment to nearly $1 billion and sending the stock sharply higher. The catch is that public investors are now paying roughly 60% more than Berkshire’s negotiated entry price, while the shares trade around 70 times earnings. The takeaway is classic Buffett: Bending Spoons may be an interesting business, but the price you pay still matters.

What does it look like when a consumer brand starts to break out?

Popsmith generated $5.8M in revenue in 2025 and is projecting $9M in 2026, with more than 100,000 customers and $20M+ in lifetime sales.

The brand has already earned national recognition through Oprah’s Favorite Things and partnerships with Costco, Sur La Table, and Williams Sonoma, where Popsmith is rolling out nationwide this holiday season.

Now Popsmith is raising capital to accelerate its next stage of growth, and individual investors have the opportunity to become shareholders.

Explore the Popsmith investment opportunity.

CoreWeave fell 25% last week. Its worst week in eleven months.

Then Monday, Goldman Sachs upgraded the stock from Neutral to Buy and raised its target from $106 to $139.

The entire bull case rests on one number. A $104 billion revenue backlog, booked as of the second quarter, plus another $25 billion in fresh Q3 commitments.

That’s the market’s confident case, stated as strongly as the people who believe it would state it.

Here’s the complication. CoreWeave’s bondholders are not reading from the same script as its equity analysts.

The 2032 notes trade below 80 cents on the dollar. The 2030 debt yields close to 16%, after jumping roughly five points in a matter of weeks.

Bond investors don’t care if the backlog converts beautifully five years from now. They care whether the company can service its debt if it doesn’t.

So the real question isn’t whether $104 billion is real. It’s whether that promised revenue can outrun the debt before the interest bill does the damage first.

I haven’t covered CoreWeave directly before. The closest parallel I’ve written is Broadcom’s vendor-financing question two weeks ago (Pragmatic Friday: Broadcom Just Became Its Own Customer’s Bank), and the question there was the same one I’m asking today: when the supplier becomes the customer’s banker, who actually owns the risk?

Start with the steelman, because it’s a genuinely strong one.

Q2 revenue grew 112.3% year over year to $2.58 billion. That is not a company losing the AI race.

The backlog sits at $104 billion. About 40% of it is expected to convert within 24 months, according to the company.

Nvidia didn’t just sell CoreWeave chips. Nvidia bought $2 billion of its Class A stock and expanded a partnership to build more than 5 gigawatts of AI factories through 2030.

Jensen Huang put it bluntly: CoreWeave’s “cost of capital is actually decreasing.”

Operating cash flow turned positive at $679 million last quarter. The company runs 51 active data centers today, with roughly 1.5 gigawatts of live compute and 4.2 gigawatts of power under contract.

Dan Ives called last week’s drop a gift, pointing to the same $104 billion backlog Goldman cited.

Consensus now expects adjusted EBITDA near $16.9 billion in 2027, and more than $27.6 billion in 2028.

Demand is real. The largest chipmaker on earth just put its balance sheet behind it. And the backlog is long enough, in theory, to outlast a bad quarter or two.

Here’s where the steelman stops holding up.

CoreWeave’s total debt was $27.6 billion at the end of Q2, against roughly $5 billion in shareholder equity. More than five times its equity base.

Analysts project gross leverage will hit 6.3x by year-end, based on roughly $47.6 billion in projected debt against about $7.6 billion in projected adjusted EBITDA.

Interest expense alone jumped to $640 million last quarter, up from $267 million the quarter before. Guidance for the current quarter is $860 million to $940 million.

Free cash flow was negative $5.74 billion. 2026 capital spending is guided at $35 billion to $39 billion, funding data centers that haven’t generated a dollar of backlog revenue yet.

Not a company running out of demand. A company racing to build fast enough to collect on the demand it already has.

Because that race has a clock on it, and the clock is the interest bill, not the backlog.

CEO Mike Intrator told employees, according to Bloomberg, that “there will be periods of time when the market misunderstands and acts irrationally.”

He also said the company is confident its backlog “more than supports all of our debt obligations.”

That’s a confident sentence. Whether it’s a true one is what the bond market is still pricing.

There’s a second layer here that the backlog number conveniently skips.

Nvidia bought $2 billion of CoreWeave stock the same month it expanded its GPU supply deal with the company. Nvidia itself has acknowledged that “some will call this circular financing.”

Not an accusation from a short seller. An acknowledgment from the supplier itself.

Then there’s concentration. Fitch noted back in June that 65% of CoreWeave’s first-quarter revenue came from just two customers.

The company has added names since, but Meta, OpenAI, and Microsoft still make up the bulk of the backlog that’s supposed to cover $47.6 billion of debt.

Zoom out and the macro backdrop isn’t helping. The Philadelphia Semiconductor Index just fell for eight straight sessions, its longest losing streak since October 2008, after large cloud providers signaled delayed data center projects.

On top of that, CoreWeave is facing a pending securities class action over construction delays.

None of this proves the backlog is fake. It proves the backlog is concentrated, intertwined with its own supplier, and sitting inside a sector that just had its worst stretch in eighteen years.

The chart tells this story better than any earnings call could.

CoreWeave spent most of September holding a range above $78. Then on Wednesday, October 14th, that range broke, the stock fell 15.5% in a session, and it kept falling into Friday’s close of $61.39, down nearly 12% on the day and 25% for the week.

That’s the worst five-session stretch this stock has had in eleven months.

Monday’s Goldman upgrade sparked a bounce back into the mid-$60s. But that bounce happened on a stock still down about 20% for the month, trading below both major moving averages, with an RSI near 29, deep in oversold territory.

There’s a floor around $60.57 that’s held twice before, though Friday’s session low of $59.90 briefly poked beneath it. If that level gives way for good, the next real support isn’t until the high $40s.

On the upside, $78.85 is the level that matters. A daily close back above it would be the first real signal that buyers, not just bargain hunters, are back in control.

Until then, this is a stock bouncing on an analyst call while its bonds keep telling a more cautious story.

CoreWeave guided Q3 interest expense to $860 million to $940 million, up from $640 million in Q2. If the print lands at the high end or above, that’s the bond market’s fear getting validated in real numbers. If it comes in lighter, Goldman’s case gets a lot more credible fast.

The $104 billion number only matters once it turns into cash. Watch whether the next quarterly update shows that promised 40%-within-24-months timeline actually holding, or whether it quietly starts slipping the way construction timelines in this industry often do.

A 16% yield on 2030 paper is a distress signal, not a rounding error. If that yield compresses toward the low double digits over the next few weeks, professional debt investors are coming around to Goldman’s view. If it keeps climbing, believe the bonds over the equity analysts.

This is a slow-moving risk, not a non-risk. Any filings, discovery developments, or settlement chatter around the construction-delay lawsuit could reintroduce headline risk right as the stock tries to stabilize.

Nvidia has made similar equity-plus-supply arrangements with more than one AI infrastructure company. Watch whether Nvidia’s tone on CoreWeave specifically stays as supportive on upcoming calls. A shift there would matter more than another price target.

Here’s my verdict after sitting with both sides of this all week.

Goldman isn’t wrong that the backlog is real. Dan Ives isn’t wrong that $104 billion of contracted revenue is an extraordinary asset for a company this young.

But an analyst upgrade doesn’t repay a bond. Cash flow does.

And right now, CoreWeave’s cash flow is negative $5.74 billion, its interest bill is growing faster than its EBITDA, and its biggest customers are also, awkwardly, its business partners.

I don’t think CoreWeave is a fraud. I think it’s a company that correctly identified where the demand is, then financed its way into a corner to capture it.

Markets rarely stay this split for long. Eventually, either the backlog converts fast enough to out-run the debt, or the debt forces a conversation nobody at CoreWeave wants to have.

Until Q3’s interest expense print comes in under guidance, I’m treating the bond market as the smarter read in this argument. Not the stock.

Shoppers type exactly what they want and get “0 results found” back. They don’t complain. They just vanish to a competitor.

Gladly understands what shoppers mean, typos and all, and turns dead-end searches into product recommendations.

See how Gladly brings search back to life

Turn Dead Ends Into Sales

There’s a principle at work here that goes beyond CoreWeave.

A backlog is not the same thing as cash in the bank. It’s a promise, made by other companies, that still has to survive their own budget cycles.

The more a business leans on debt to capture a promise before it arrives, the more that promise has to show up exactly on schedule.

This is the story underneath half the AI infrastructure trade right now, not just CoreWeave’s version of it.

When the financing gets aggressive enough, the quality of the demand starts to matter less than the speed of the cash.

Watch the interest payments as closely as you watch the revenue growth. The first one is where companies actually break.

If this deep dive changed how you see CRWV, send it to one friend who’s still trading the headline.