Broadcom Isn't a Chip Company. It's a Machine That Buys Cash Flows.
Everyone reads Broadcom as a chipmaker riding the AI wave. It isn't — $21.5 billion of its $51.6 billion in FY2024 revenue is software it bought, not silicon it invented. The real product is the deal.
Comes with a free Profit-Engine Map template.
In November 2023, after eighteen months of regulators in three jurisdictions picking the deal apart clause by clause, Broadcom finally closed its purchase of VMware.4 The headlines called it a chip giant grabbing a software prize, the AI era's biggest land grab. They had the genre wrong. Broadcom wasn't buying VMware to enter software — it had already done that years earlier. It was running the same play it had run on CA Technologies and on Symantec's enterprise unit before that: find a piece of technology that thousands of large companies cannot rip out, buy it, cut the fat, and turn its revenue into something that renews automatically. The chip is what people see. The deal is the actual product.
The official story is that Broadcom is a semiconductor company that opportunistically diversified into software. The truer story is the reverse of how it's told: the entity called Broadcom today is the former Avago — a 2005 carve-out of Hewlett-Packard's old semiconductor business by KKR and Silver Lake — which bought the original Broadcom Corporation for about $37 billion in 2016 and simply took its name.8 The company everyone thinks of as Broadcom is the acquirer wearing the acquiree's clothes. That detail isn't trivia. It tells you what kind of company this is.
The thing Broadcom actually sells is the deal
Here is the thesis a smart friend could repeat at dinner: Broadcom is not a chip company that stumbled into software. It is a capital-allocation engine that serially buys mission-critical, hard-to-replace technology franchises — silicon and software alike — strips out cost, converts perpetual licenses into subscriptions, and pours the free cash flow into the next acquisition. The AI chip boom is real money, but it is a recent accelerant bolted onto a machine that was already running. Look at where the revenue comes from and the machine shows itself: in FY2024, Broadcom booked a record $51.6 billion, with Semiconductor Solutions at $30.1 billion and Infrastructure Software at $21.5 billion — 42% of the company from software it bought rather than chips it designed.1
Notice how the 10-K describes its own revenue. It doesn't lead with 'chips' and 'software.' It splits the company into Products at $30.4 billion and Subscriptions and services at $21.2 billion.2 That second line is the tell. When you buy CA Technologies' mainframe tools or VMware's virtualization stack, you are not buying a product once — you are renting it forever, because turning it off means turning off the systems your whole business runs on. The mainframe doesn't get unplugged. Neither does the contract.
The playbook, run four times in plain sight
The pattern is not hidden — it is stated in the filings. When Broadcom announced the $18.9 billion CA Technologies deal in July 2018, the press release literally framed it as continuing 'Broadcom's Focus on Acquiring Established Mission Critical Technology Businesses,' and projected adjusted EBITDA margins above 55%.7 That is the whole strategy printed on the box. Find a business customers can't leave. Pay cash. Engineer the margin upward. A year later Broadcom bought Symantec's enterprise security business for $10.7 billion, deepening the same software footprint.6 Then came VMware. Each target was chosen for the same trait: switching costs so high the customer is effectively captive.
And about that VMware number — popular coverage almost always says '$61 billion.' That was the equity value at the May 2022 signing price. Broadcom also assumed $8 billion of VMware net debt, which is why the filings and the deal's own counsel describe a roughly $69 billion transaction.34 The gap between the two figures is the difference between reading a headline and reading the 8-K. The machine runs on the second one.
| A typical chipmaker | Broadcom's machine | |
|---|---|---|
| Source of growth | Design wins, new silicon | Acquired franchises, deal by deal |
| What the customer can do | Switch suppliers next cycle | Almost never rip it out |
| Revenue shape | Sold once per product | Converted to recurring subscription |
| Where cash goes | Reinvested in R&D | Cut costs, then fund the next deal |
CA was explicitly underwritten to adjusted EBITDA margins above 55%7, and the result shows up at the top line: $51.6 billion of FY2024 revenue, 42% of it the recurring software the company bought rather than invented.1 Each deal funds the next, which is why the same move recurs from CA to Symantec to VMware. The loop compounds because the targets, by design, can't churn.
But isn't AI the real story now?
The fair objection is that all of this is yesterday's Broadcom — that the company is now an AI infrastructure giant, and the M&A history is a footnote to the chips going into the world's largest data centers. The numbers give the objection real teeth: AI semiconductor revenue hit $12.2 billion in FY2024, up 220% year-on-year.1 That is not a rounding error; it is one of the fastest-growing lines in the company. So why insist the deal machine is the main event?
Because the AI revenue sits inside the same structure the acquisitions built, and behaves the same way. Custom AI accelerators are mission-critical, deeply embedded, and expensive to design out — the same switching-cost logic that made CA and VMware attractive, now applied to silicon a hyperscaler co-designs and can't casually replace. The AI surge didn't replace the playbook; it gave the playbook its best possible product. The honest version of the counter is that the accelerant is now large enough to be misread as the engine. It isn't. Strip the AI growth away and you still have a $40-billion company built on recurring revenue from franchises it bought. Strip the machine away and the AI chips have nowhere to compound.
The most durable acquisition isn't the fastest-growing one — it's the one the customer cannot leave. Broadcom's targets share a single trait: rip them out and something essential stops working. Mainframe tooling, virtualization, enterprise security, custom accelerators — none of them are exciting, all of them are load-bearing. That's the whole filter. Buy the business with the highest switching cost, convert its revenue to a subscription so the cash recurs, cut the cost the prior owner left on the table, and use the free cash flow to do it again. The discipline is in the boredom: Broadcom doesn't chase the cool franchise, it chases the unleavable one — and prices the deal to a margin it states out loud before it signs.
“Broadcom's Focus on Acquiring Established Mission Critical Technology Businesses.”7
So how does Broadcom actually make money? Not the way a chip company does, and not the way a software company does — it makes money the way a great acquirer does, by being right about which businesses customers can't quit, and then never paying a cent more than a stated margin will justify. The chips and the software and the AI accelerators are the inventory. The machine is the thing that decides what to buy and what to charge, and then turns the answer into recurring cash. Broadcom's genius was never a particular product. It was figuring out that the most valuable thing you can own is something nobody can afford to turn off — and that you can buy it, again and again, with the cash from the last thing nobody could turn off.
Profit-Engine Map
A one-page map that pulls a business apart into the hook that gets the customer in the door and the engine that quietly earns the margin. Use it to see where the real profit lives, how the two halves are wired together, and what breaks if the link is cut. Blank to dissect your own P&L; filled as the worked example of a business whose advertised product is not where it makes its money.
The worked example unlocks with a subscription. See plans →
Sources
Where this comes from — the filings, records, and reporting behind it.
- 1Broadcom FY2024 total revenue was a record $51.6 billion, up 44% year-over-year; Semiconductor Solutions contributed $30.1 billion (58%) and Infrastructure Software $21.5 billion (42%); AI semiconductor revenue was $12.2 billion, up 220% year-on-year.
- 2Broadcom FY2024 10-K (fiscal year ended November 3, 2024): total net revenue $51,574 million, of which Products $30,359 million and Subscriptions and services $21,215 million; audited by PricewaterhouseCoopers LLP, signed December 20, 2024.
- 3Broadcom announced the acquisition of VMware on May 26, 2022, at an equity value of approximately $61 billion in cash and stock; Broadcom would additionally assume $8 billion of VMware net debt, bringing total enterprise value to approximately $69 billion.
- 4The VMware acquisition closed on November 22, 2023, after 18 months of regulatory reviews including a second request by the US FTC and Phase II investigations by the UK CMA and European Commission; the total transaction value including assumed debt was approximately $69 billion.
- 5Broadcom completed its acquisition of CA Technologies on November 5, 2018, for $18.9 billion in cash ($44.50 per share), entering the enterprise/mainframe software market.
- 6Broadcom announced the acquisition of Symantec's enterprise security business for $10.7 billion in cash on August 8, 2019, to expand its infrastructure software footprint.
- 7Broadcom announced the CA Technologies acquisition on July 11, 2018 for $18.9 billion; the deal was explicitly positioned as continuing 'Broadcom's Focus on Acquiring Established Mission Critical Technology Businesses' and was expected to drive adjusted EBITDA margins above 55%.
- 8The entity now called Broadcom Inc. originated as Avago Technologies, formed in 2005 when KKR and Silver Lake Partners acquired Agilent Technologies' semiconductor group. Avago acquired the original Broadcom Corporation for approximately $37 billion in 2016 and adopted the Broadcom name while retaining the AVGO ticker.