Nvidia’s quiet decision to underwrite loans to struggling NeoCloud customers signals a risk that most AI investors ignore, and it makes clear why a rival chipmaker continues to win over the customers that are worth it.
I bought more Broadcom (NASDAQ:AVGO | AVGO Price Prediction) last week, and I plan to continue following the stock closely. Shares fell 9.07% in the week ending October 16 to $328.75. For me, this decline marked the AI sector I trust the most, and I added.
Why My Money Follows Custom Silicon
My thesis is based on one fact: Broadcom co-designs the chips that the largest AI manufacturers run on, and these customers commit to years of product plans. On the September call, Hock Tan said that “when you co-develop a chip optimized for your particular LLM workloads, you will outperform any GPU,” and added that customers “can do all of this for half the cost of a GPU.” When a customer gets that kind of cost advantage, it comes back for the next generation, and the one after that.
Three receipts behind every purchase
First, growth. AI semiconductor revenue reached $16.70 billion in the fiscal third quarter, up 221% year-over-year. Management guided the fourth quarter to $21.70 billion, up 236%, and forecast about $115 billion in AI revenue for fiscal 2027 and $230 billion for fiscal 2028. It also said it has already secured supply for both years.
Second, cash and income. Free cash flow reached $13.7 billion, or 46% of revenue. The $0.65 quarterly dividend reflects the 15th consecutive annual increase, and Broadcom still paid down $5.6 billion in longer-term debt during the quarter. It has also beaten EPS estimates for 10 consecutive quarters.
Third, valuation. Tan said Broadcom was “on track to exceed $30 in earnings per share in fiscal 2028.” At current prices, that equates to about 11 times the target. This multiple supports the valuation case.
Why I leave aside NVIDIA and Marvell
The name most investors choose first is Nvidia (NASDAQ:NVDA). What turns me off is a recent report that NVIDIA has contacted insurers about structures that could insure “against losses on loans to new cloud computing companies or Neo clouds” in case those borrowers default and the chips they have put in cannot be resold at a price sufficient to repay the lenders. I want my exposure to AI to be tied to clients with strong balance sheets. NVIDIA also pays just $0.28 per share in annual dividends and trades at a forward P/E of 25.
Marvell Technology (NASDAQ:MRVL) is the other custom chip designer, but it trades at a forward P/E of 62 and pays $0.24 per share per year. Broadcom pays more than that every quarter.
Risk that could harm this position
Concentration is the real danger. The company has six XPU customers, and four of them account for most of the scale. It also offers residual value guarantees through its AI financing platform with Apollo and Blackstone, and management declined to give a maximum exposure figure. Constraints related to land, electricity and data center buildings could delay deployments.
I still stand by my thesis because these clients sign long-term agreements. Broadcom plans to supply tens of billions of dollars worth of TPU to Google alone each year. Management also claims that “our demand actually exceeds this outlook” and that the company has a cash reserve of $24 billion.
What could move Broadcom next
Broadcom reports again on December 9, 2026, against forecast revenue of $34.80 billion, up 93%. I plan to monitor AI’s earnings and dividend growth. While NVIDIA works to insure its neocloud customers, Broadcom continues to sign hyperscaler contracts and increase its payments, and this contrast supports my thesis.
Contact (email protected) for any questions or corrections.
Gn bussni
