Cramer Ties Broadcom's $30 Billion AI Bet to Anthropic
Broadcom's record custom-silicon guide rests on a customer no one can own. Cramer's "Anthropic or bust" line landed as AVGO fell 4.01% against a rising market.

Jim Cramer said on CNBC's Squawk on the Street Thursday morning that Broadcom's $30 billion AI bet now hinges on Anthropic, a private company investors cannot buy, telling viewers "It's Anthropic or bust… They need Anthropic to have a blockbuster IPO," as AVGO traded at 352.50, down 4.01% on the day at 16:24 GMT.
Jim Cramer used his Thursday morning slot on CNBC's Squawk on the Street to make a point that cuts against the way most investors have been framing Broadcom (AVGO): the largest custom-silicon commitment the semiconductor industry has ever guided to is underwritten, in his telling, by a company no public investor can own a share of.
"This is now Anthropic. It's Anthropic or bust… They need Anthropic to have a blockbuster IPO," Cramer said, as reported by 24/7 Wall St. The remark reframes a $30 billion AI order book from a demonstration of Broadcom's engineering moat into something closer to a single-counterparty credit question.
The market took the point. Broadcom traded at 352.50 as of 16:24 GMT on Thursday, down 4.01% from the prior close of 367.24 and inside a session range of 342.33 to 356.66. That happened on a day when the broad tape was firmly higher: the S&P 500 tracker (SPY) was up 0.98% at $772.63, the Nasdaq 100 tracker (QQQ) up 1.12% at $717.18, and the Dow 30 tracker (DIA) up 1.15% at $536.73. Broadcom underperformed the S&P 500 proxy by roughly five percentage points on the day — an unusually wide gap for a mega-cap semiconductor name on a risk-on session.
What "custom silicon" actually commits a buyer to
Custom AI accelerators — application-specific integrated circuits, or ASICs — are chips designed for one customer's workload rather than sold off the shelf. Broadcom co-designs them, owns the networking and packaging intellectual property around them, and books the manufacturing. The economics are attractive precisely because they are exclusive: the customer cannot easily switch suppliers mid-program, and the supplier gets multi-year visibility.
That exclusivity runs both ways. A general-purpose GPU that one buyer cancels can be resold to another. A custom part designed around one customer's model architecture cannot. So when a guide of the size Cramer is describing rests disproportionately on one name, the revenue is not diversified backlog in the ordinary sense — it is a bet on that customer's ability to keep funding capital expenditure at scale, quarter after quarter, for years.
This is the structural point underneath Cramer's soundbite. The $30 billion figure is not in dispute; what is in dispute is how much of it is contractually firm versus dependent on the buyer's own balance sheet holding up. Broadcom has not published a customer-by-customer breakdown, and the lead provides no such split. Investors are, in effect, being asked to underwrite a private company's funding trajectory without the disclosure they would get from a listed counterparty.
Why a private customer changes the risk math
Anthropic is private. There is no 10-Q, no quarterly cash burn disclosure, no covenant package visible to Broadcom shareholders, and no equity for a Broadcom holder to buy as a hedge against the exposure. If you believe the AI buildout continues, the natural expression is to own the supplier; but the supplier's revenue quality is now a function of information the market does not have.
Cramer's specific framing — that Anthropic needs "a blockbuster IPO" — points at the financing channel rather than the technology. An initial public offering would do three things at once for the chain of dependency:
- Convert a private funding cycle, which can stall without warning, into permanent public capital with an observable price.
- Give Broadcom investors a live, daily read on the health of the customer sitting behind a large share of the AI order book.
- Create a currency the customer could use to fund compute commitments without repeatedly returning to private rounds.
Absent that, every private round becomes a de facto Broadcom event. The chip company's revenue visibility would be hostage to a financing calendar it does not control and cannot disclose.
The concentration problem the whole AI supply chain shares
Broadcom is not alone in this. The AI hardware complex has spent two years re-rating on order books whose end demand traces back to a small handful of model developers, several of them private and several of them funded by the same strategic investors who also sell them chips and cloud capacity. Circularity of that kind is fine while capital is abundant. It is unforgiving when it is not.
The chip company's revenue visibility would be hostage to a financing calendar it does not control and cannot disclose.
What makes Thursday's move notable is the divergence. Broad indexes rose across the board while Broadcom fell — the market was not selling AI exposure generally, it was repricing this particular structure. Whether that holds beyond a single session is the first thing to watch.
What to track from here
Three things determine whether Cramer's warning ages well:
- Disclosure. Any move by Broadcom to quantify how much of the AI backlog is prepaid, take-or-pay, or cancellable would materially change how the $30 billion figure should be discounted.
- Financing news at the customer. Confirmed IPO preparation, or a large private round on strong terms, resolves the funding question in Broadcom's favor. Silence does not.
- Relative performance. Broadcom's 4.01% decline into a rising market is the market's first vote. If the stock keeps lagging the Nasdaq 100 on up days, the concentration discount is becoming permanent rather than a one-day reaction.
None of this touches Broadcom's engineering position, which is the reason it won the business in the first place. But a backlog is only as good as the buyer behind it, and Cramer's contribution Thursday was to insist that shareholders name the buyer out loud.
Frequently asked questions
What exactly did Jim Cramer say about Broadcom?
Speaking on CNBC's Squawk on the Street on Thursday morning, Cramer said of Broadcom's AI business: "This is now Anthropic. It's Anthropic or bust… They need Anthropic to have a blockbuster IPO." His argument is that the customer underwriting Broadcom's record custom-silicon commitment is a private company, so public shareholders cannot see or hedge that exposure directly.
How did Broadcom stock react?
Broadcom traded at 352.50 as of 16:24 GMT on Thursday, September 3, 2026, down 4.01% from the previous close of 367.24, with a session range of 342.33 to 356.66. The decline came on a broadly positive day for equities, with the S&P 500 tracker up 0.98% and the Nasdaq 100 tracker up 1.12%.
Why does it matter that Anthropic is private?
A private customer files no quarterly reports, so Broadcom shareholders cannot see the buyer's cash position, burn rate or funding runway. They also cannot buy or short the customer's equity to hedge the exposure. That makes a large share of Broadcom's AI order book dependent on information the public market does not have access to.
What is a custom AI ASIC?
An application-specific integrated circuit is a chip designed for one customer's particular workload rather than sold as a general-purpose product. Broadcom co-designs these accelerators and supplies the surrounding networking technology. The arrangement locks in multi-year revenue, but because the part is bespoke it cannot easily be resold if the original buyer cancels.
Would an Anthropic IPO resolve the concern?
It would address the financing side of it. A public listing converts private funding cycles into permanent capital, gives Broadcom investors a daily market read on the customer's health, and creates equity currency the buyer could use to fund compute purchases. It would not, by itself, tell investors how much of Broadcom's backlog is contractually firm.
Is this concentration risk unique to Broadcom?
No. Much of the AI hardware supply chain traces end demand back to a small group of model developers, several of them private and several funded by the same strategic investors who also sell them chips and cloud capacity. That circularity is manageable while capital is plentiful and becomes a problem when funding conditions tighten.
Sources
Photo: Gustavo Fring · Pexels Licence — source


