# chip earnings — X 热门讨论 (2026-09-16 11:11 UTC)
## @jpinsights (JP Insights) · 09-16 08:52 · ♥36 ↻2 💬10 One of the comments regarding my conviction in Broadcom was: “That’s what people catching falling knives tell themselves to feel better.”
I obviously disagree. It’s an easy comment to make about someone buying a falling stock. I’d be more interested in hearing which assumptions in my analysis are wrong.
Here’s my case for $AVGO.
I bought around $390 and thought I was getting a good deal. It has kept falling since, which is frustrating but part of the game. I would obviously have preferred buying cheaper. I keep on adding at lower levels. My investment horizon is at least two years, though, and the latest report gave me plenty of reasons to remain confident about what Broadcom can earn over that period.
Q3 AI semiconductor revenue reached $16.7B, up 221% year over year and 54% sequentially. Management guided Q4 to $21.7B, implying another roughly 30% increase from one quarter to the next.
Broadcom also generated $13.7B in quarterly free cash flow, equivalent to 46% of revenue. That gives the company considerable room to fund development and pay down debt as the AI business expands.
Management now expects approximately $115B in AI revenue for FY2027 and $230B for FY2028. Hock Tan tied those figures to secured supply and customer deployment plans. Getting the data centers ready will influence when that revenue arrives, so I expect some uneven quarters along the way.
The economics behind custom chips make sense to me. These customers spend enormous amounts running their models. A lower cost per workload can justify billions in chip development, particularly when they expect to run those workloads at increasing scale for years.
Broadcom has spent years helping customers develop and produce those chips. Each successful generation gives the customer another reason to continue the relationship. Moving work elsewhere takes engineering resources and introduces execution risk, even when the customer has the money to do it.
Google bringing in $MRVL has understandably raised questions. I own Marvell too, and I think the opportunity is very attractive for them.
I also think people are too quick to assume that Broadcom must suffer proportionally from every dollar Marvell wins. Google’s demand can grow fast enough to support several suppliers and still leave Broadcom with a much larger business.
My reading is that Google wants more capacity and less dependence on any individual partner. It also gains negotiating power. We should account for that in Broadcom’s future margins, but I don’t see a reason to assume the relationship is falling apart.
Google and Broadcom have a long-term agreement covering future TPU generations, alongside a supply agreement for networking and other rack components through up to 2031. Google is clearly still planning around Broadcom.
I’m comfortable with Broadcom losing some percentage share if its revenue and profit continue growing at anything close to the rates we’re discussing. The size of the market makes an enormous difference here. So does the profitability of the business it keeps.
The same applies to customers building internal chip teams. Google already designs its own TPUs and works with Broadcom. A customer can take more control over the architecture while continuing to pay an external partner for the engineering and production work needed to deliver it.
Anthropic has announced multiple gigawatts of additional TPU capacity through Google and Broadcom, expected to start coming online in 2027. That gives us actual business to analyse. I have included no revenue from an unannounced Broadcom contract for an Anthropic-designed processor in my case.
Broadcom also sells the networking needed to connect these increasingly large clusters. AI networking revenue grew more than 2.5 times year over year in Q3. I like having that exposure alongside the custom processors, particularly as customers spend more to keep all that expensive compute working efficiently.
Then there’s the valuation, which is why I keep buying.
My updated model came out at roughly $30 in adjusted EPS for FY2028. At my $390 purchase price, I paid approximately 13 times that estimate. A 22× multiple would put the shares around $660 in FY2028.
I find that attractive. It gives me substantial potential upside using a multiple I’m comfortable underwriting, provided Broadcom delivers the earnings.
I also ran a delay case with roughly $23.5 in EPS and an 18× multiple, giving a value around $424. My more severe case came out around $17.2 in EPS at 15×, or roughly $258. Execution problems combined with a lower valuation could clearly hurt. Those scenarios help me judge how much I’m willing to own and what I’m willing to pay.
There are details in the accounts I’ll keep watching closely. The growing memory content in custom accelerators dilutes gross margins, and Broadcom’s involvement in customer financing can include residual-value guarantees. Those guarantees create financial exposure that deserves scrutiny as the arrangements grow.
I want the additional sales to produce enough profit and cash per share to justify that exposure. Lower gross margins can be perfectly acceptable if operating profit grows strongly and cash conversion holds up. Revenue growth becomes much less appealing if Broadcom has to absorb increasing risk to secure it.
So far, the results support my earnings expectations, and I think the potential return is well worth taking those risks. Broadcom remains one of my highest-conviction holdings.
I’ll change my estimates if the business gives me a reason to. For now, I’m still adding when I can. https://x.com/jpinsights/status/2100145719818514728