Broadcom Promised $230 Billion of AI Revenue by 2028. The Fine Print Says the Seller Is Also the Bank.

Published: September 7, 2026 | Reading Time: ~10 minutes | Channel: Business

---

Here's the stat that should stop you mid-scroll: Broadcom's AI chip business pulled in $16.7 billion in a single quarter — more than the company's entire revenue in the same quarter a year earlier, which was $15.95 billion.¹ One product line just out-earned what the whole corporation booked twelve months ago. CEO Hock Tan then told Wall Street that AI revenue will hit $58 billion this fiscal year, $115 billion in 2027, and $230 billion in 2028 — a number that would be roughly four times this year's already-absurd total.²

The headlines, inevitably, screamed "blowout." Nine straight quarters of EPS beats.¹ A $1.8 trillion market cap.³ Guidance that made the LSEG consensus look timid.

Now the part the headlines buried: Broadcom's revenue forecast for the current quarter missed Wall Street's number — $34.8 billion guided versus $35.03 billion expected.³ The stock is up about 6% in 2026 while the S&P 500 is up 12%.³ And Broadcom's own CFO, Amie Thuener, admitted on the earnings call that the company is working to help its biggest AI customers — labs that famously don't turn profits — "bridge the gap between their current cash flow and the significant upfront investments required for their businesses," possibly through residual value guarantees booked as contingent liabilities.³

Read those two halves together and you get the real story. This isn't just a chip company reporting record earnings. It's a chip company that has started financing its own order book. We've seen this movie before. The last time a supplier became the bank for its customers, the credits rolled around March 2001.

---

The Quarter That Ate Its Own History

Let's do the numbers first, because the numbers are genuinely extraordinary — and I say that as someone who spends most of his week professionally suspicious of extraordinary numbers.

On September 2, Broadcom reported fiscal Q3 2026 results (quarter ended August 2) that beat on both lines: adjusted EPS of $3.32 against $3.24 expected, revenue of $29.59 billion against roughly $29.4 billion expected.³ Revenue grew 86% year-over-year. GAAP net income more than tripled to $13.09 billion.¹ The AI semiconductor segment — custom accelerators ("XPUs") and the networking gear that stitches them into data-center-spanning clusters — grew 221% year-over-year and 54% quarter-over-quarter to $16.7 billion, beating the $15.2 billion StreetAccount estimate.³

Metric (GAAP unless noted)	Q3 FY2025	Q3 FY2026	Change
Net revenue	$15.95B	$29.59B	+86%
Operating income	$5.89B	$15.96B	+171%
Net income	$4.14B	$13.09B	+216%
Diluted EPS	$0.85	$2.68	+215%
Non-GAAP EPS	$1.69	$3.32	+96%
Cash from operations	$7.17B	$14.20B	+98%
Free cash flow	$7.02B	$13.67B (46% of revenue)	+95%

Source: Broadcom Q3 FY2026 press release, Sept 2, 2026¹

Here's my favorite number in the whole release, and it's not the revenue: free cash flow was 46% of revenue.⁴ Broadcom is converting nearly half of every sales dollar into cash while revenue doubles. Broadcom spent just $0.5 billion on capex in the quarter¹ — this is the exact opposite of the "AI capex bonfire" story you read about its customers. The company ended the quarter with $24.0 billion in cash, up from $19.6 billion the previous quarter.¹ Whatever else you want to say about Hock Tan, the man runs a cash machine.

---

By the Numbers: The $58B → $115B → $230B Ladder

Now to the promise. On the earnings call, Tan laid out a staircase that has no historical precedent at this scale:

Fiscal Year	AI Revenue (guided/projected)	Growth
FY2026	$58 billion	+186% YoY (raised from $56B)
FY2027	$115 billion	~+98%
FY2028	$230 billion	~+100%

Source: Broadcom Q3 FY2026 earnings call, transcribed by Benzinga²

Two consecutive doublings. For context, $230 billion of AI revenue in fiscal 2028 would be more than most Fortune 50 companies generate in total revenue. And Tan didn't stop at revenue: he told the street Broadcom is on track to produce over $30 in EPS in fiscal 2028, versus the LSEG consensus of $25.86 — a 16% beat of what the professional number-crunchers had modeled.³

The customer math behind the ladder, from the call itself:²

• Google: A new long-term agreement to develop and supply future TPU generations — "multi tens of billions of dollars of TPUs annually over the next several years." Broadcom is already shipping TPU v8i, a generation ahead of MediaTek's competing V8T, which only entered development this quarter.
• Anthropic: 1 gigawatt of Ironwood TPUs deploying in 2026, 5 gigawatts of TPU v8i in 2027, and "clear line of sight" to another 10 gigawatts in 2028. Tan says Anthropic becomes Broadcom's largest XPU customer in 2027.
• OpenAI: The custom "Jalapeno" chip — which Tan claims outperforms Nvidia's Grace Blackwell for inference — is on track for a 1.3 gigawatt deployment in 2027, with line of sight to over 5 gigawatts of Jalapeno-plus-successor by 2028. A second-generation chip is "approaching tapeout," and a third generation is in development.
• Meta: Production shipments of the MTIA inference accelerator beginning now.

XPU shipments grew more than 3.5x year-over-year and made up 73% of AI revenue in the quarter; AI networking revenue grew more than 2.5x.² Six XPU customers in total.² That's the supply side of the story — real silicon, real shipments, real 68% non-GAAP operating margins.² Nobody is faking this quarter.

---

Why "Blowout" Is the Wrong Word

Here's where I lose the consensus, and where the second-order effects start to matter.

First: the guide missed. Q4 revenue guidance of $34.8 billion came in below the $35.03 billion analysts expected.³ Reuters' own headline (circulated in the sector week-in-review) called it exactly as it is: "Broadcom forecasts quarterly sales below estimates amid rising custom AI chip competition."⁵ When a company misses the top line by 0.7% while claiming revenue will quadruple by 2028, the market is not being ungrateful — it's asking a fair question about the slope between here and there. Add that infrastructure software revenue of $8.75 billion also slipped under the $8.82 billion consensus,³ and the "record everything" framing gets a dent.

Second: the stock's telling you something. Broadcom shares closed at $357.90 on Friday (last trade before the Labor Day close), which is about 28% below its 52-week high of $495.00.⁶ A company that just printed $16.7 billion of quarterly AI revenue and promised $230 billion is trading at a 28% discount to its own high water mark, and it's lagging the S&P by six points year-to-date.³ Markets don't hand out discounts like that for no reason. The reason is the next point.

Third — and this is the one everyone will be talking about by October: the seller is becoming the buyer's bank. CFO Amie Thuener, on the record: Broadcom is working to empower "two of our most strategic customers, the leading AI labs, to bridge the gap between their current cash flow and the significant upfront investments required for their businesses," including **"residual value guarantees that are contingent liabilities to the labs."**³ Tan's own summary: "It makes economic sense for Broadcom to invest and enable these guys."³

Follow that logic to its conclusion. Anthropic — a lab that loses money on every frontier-model generation and pays for compute with a capital-raising treadmill — becomes Broadcom's largest XPU customer in 2027.² Broadcom's revenue growth increasingly depends on customers whose own revenue doesn't cover their purchases, so Broadcom underwrites the purchases. That is the textbook definition of vendor financing: sell the shovels, then lend the gold rushers the money to buy them. Yahoo Finance's headline writers saw it too — "Broadcom's AI chip bet raises questions over who foots the bill."⁵

If you're too young to remember why that phrase should raise your pulse: in the late 1990s, Lucent and Nortel and Cisco booked spectacular quarters partly by extending financing and guarantees to telecom customers who couldn't otherwise afford the gear. The revenue was real. The margins were real. Then the customers' business models met reality, the guarantees detonated, and billions of "solid" revenue became impairment charges. I'm not saying Broadcom is Lucent — Broadcom's 46% FCF margin and a customer list including Google and Meta make it a very different animal.¹ I'm saying the mechanism rhymes, and mechanisms are what bite you.

Fourth: concentration has a name, and in 2027 it might be Anthropic. The growth story is anchored on a handful of frontier labs — Tan himself said "the vast majority of compute demand for AI workloads today originates from this concentrated group."² When your $230 billion year requires one loss-making lab to deploy 10 gigawatts of chips in 2028, you've built a skyscraper on a tenant with no fixed lease term.

---

What This Means For You

Enough doom — there's real money to be made and real mistakes to avoid here. Five concrete moves, not vibes:

1. If you own AVGO: don't anchor to the $230 billion headline — anchor to the cash conversion. The investment case that survives every scenario is the one priced on the current business: ~$13.7 billion of quarterly free cash flow at 46% of revenue, a $0.65 dividend, and a nine-quarter beat streak.¹⁴ Size your position so that you'd be content if the 2028 promise quietly evaporates and you're left holding a cash machine, not a lottery ticket.

2. Watch the 10-K, not the press release. "Residual value guarantees" are now formally on the table as contingent liabilities.³ When Broadcom's annual filing lands (typically December), read the guarantees and commitments note with the same intensity you'd give a loan agreement you personally co-signed — because economically, if you're a shareholder, you did. If the disclosed contingent-liability number starts scaling toward double-digit billions, the "quality of revenue" story changes, and the multiple should too.

3. Track Anthropic's funding rounds as a leading indicator for AVGO. This is now a hard dependency: Broadcom's 2027 "largest customer" needs tens of billions in external capital to pay for its chips.² Anthropic raise announced → AVGO risk premium compresses. Anthropic raise delayed or downsized → that's your early warning, months before it shows in any revenue line. Same diligence for OpenAI's capex disclosures.

4. Operators and sellers, take note of the custom-silicon land grab. Six XPU customers, $230 billion of projected demand, and Broadcom openly bragging it beat MediaTek to the next TPU generation.² If you sell into AI data centers — cooling, power, substrates, lasers (Broadcom itself flagged substrate and laser capacity expansion)² — the custom-ASIC ecosystem is where hyperscaler money is actually moving. Price for volume, contract for duration.

5. If you're an investor in the broader AI complex, rotate your attention from "who has demand" to "who has financing." The market's question for 2026 has quietly shifted. Nvidia, Broadcom, TSMC — the demand side is no longer the debate. The differentiator is balance-sheet exposure: which suppliers are taking credit risk on their customers, and how much. Read the commitments notes. Every one of them.

---

⚠️ The Risks Nobody's Talking About

1. Guarantee contagion. Residual value guarantees are only free until they're not. If one major lab restructures or reprices its compute commitments, Broadcom's contingent liabilities convert into real charges against real earnings. There is no public quantification of this exposure yet — that absence of a number is itself the risk.³

2. Customer concentration at the worst possible moment. A single customer segment — frontier labs operating at structural losses — is underwriting the majority of projected growth. Anthropic alone is slated to be the largest XPU customer in 2027.² Hyperscalers (Google, Meta) are creditworthy; labs on a fundraising treadmill are a different credit entirely.

3. Competition is arriving exactly on schedule. MediaTek's rival TPU chip (V8T) is in development; Marvell projects $18 billion of data center revenue by 2028.⁵ Reuters framed the quarter's caution around "rising custom AI chip competition."⁵ Broadcom's pricing power — and those 68% operating margins — are a target, not a constant.

4. The macro window is narrowing. The Fed under Warsh has been signaling hikes against an oil backdrop that keeps re-inflating, and Friday's jobs report surprised enough to jolt yields. A higher-for-longer rate environment raises the cost of every data center, every lab's next funding round, and every dollar of that $230 billion future. Long-duration promises get marked down when discount rates rise — that's arithmetic, not opinion.

---

🎯 The Bottom Line

Broadcom's quarter is real: $16.7 billion of AI revenue, 46% free-cash-flow margins, nine straight beats, and a customer list that includes every company building frontier AI.¹⁴ The $230 billion promise is where faith is required — because the same call that promised it also admitted the seller is extending credit to the buyers. When the company shipping the chips becomes the bank financing them, celebrate the present, audit the future, and read the guarantees section of the next 10-K like your net worth depends on it. It might.

---

📚 Verified Sources

1. PRNewswire / Broadcom Inc. — Official Q3 FY2026 financial results press release (quarter ended August 2, 2026): revenue, segment data, FCF, guidance, and Tan/Thuener quotes. https://www.prnewswire.com/news-releases/broadcom-inc-announces-third-quarter-fiscal-year-2026-financial-results-and-quarterly-dividend-302868129.html
2. Benzinga — Full Broadcom Q3 FY2026 earnings call transcript: XPU customer detail, gigawatt deployments, FY26/27/28 AI revenue path, margin commentary. https://www.benzinga.com/news/26/09/61592658/broadcom-q3-2026-earnings-call-transcript
3. CNBC — "Broadcom (AVGO) Q3 earnings report 2026": LSEG consensus comparison, Q4 guidance miss, YTD stock performance, market cap, residual value guarantees. https://www.cnbc.com/2026/09/02/broadcom-avgo-q3-earnings-report-2026.html
4. 24/7 Wall St. — Earnings card: beat magnitudes, nine-quarter streak, FCF conversion, VMware private AI cloud announcement. https://247wallst.com/cards/broadcom-q3-2026-earnings-avgo-01m1hx207g9d64z38jwrpy5fdm
5. Distill Intelligence — Semiconductors & AI Chips Weekly Briefing (Sept 4, 2026), aggregating Reuters ("forecasts quarterly sales below estimates amid rising custom AI chip competition") and Yahoo Finance ("who foots the bill") headlines; Marvell 2028 projection. Bronze-tier aggregator, used for market-narrative context only. https://www.distillintelligence.com/briefings/semiconductors-ai-chips-2026-09-04
6. Nasdaq realtime quote (via market data tool), Sept 7, 2026 — AVGO last sale $357.895 (+0.21%), volume 32,863,628, 52-week range $289.96–$495.00, market closed for Labor Day. Retrieved live; US markets were closed Monday, Sept 7, 2026.

All claims verified against Gold-tier (official Broadcom press release via PRNewswire) and Silver-tier (CNBC, Benzinga transcript, 24/7 Wall St.) sources. Each source URL was scraped and confirmed accessible with content matching cited claims; one source (Seeking Alpha) was discarded after returning a 403. Market context figures cross-checked against live quote data pulled September 7, 2026. Last verified: September 7, 2026.

---

When the guy selling the shovels starts lending you the gold rush, the quarter is real — the decade is a credit decision. 🎯

This article is analysis, not investment advice. Position sizing, guarantees exposure, and customer credit are moving targets — do your own diligence.