What Does Broadcom (AVGO) Actually Do? A Simple Guide to AI's Hidden Giant

If you've only heard of Nvidia (NVDA), you might assume the AI chip market is entirely Nvidia's game. But behind every AI compute feast sits a company most retail investors overlook—yet virtually every tech giant depends on. That company is Broadcom (NASDAQ: AVGO).

This article breaks down, in plain language, what Broadcom does, why it suddenly matters so much for AI, why the stock has tumbled from its highs, and what to watch next.

1. What Does Broadcom Do? Two Businesses, Plain and Simple

Many people think of Broadcom as "the networking chip company." That is half right. Broadcom's business actually splits into two major parts.

The first is its semiconductor business, which itself breaks into two sub-segments. One is the legacy lineup—networking chips, broadband chips, storage controllers, and similar products that have been Broadcom's bread and butter for years, generating steady cash flow. The other is the newer, sexier part: custom AI chips (known in the industry as ASICs), which is what the market is most excited about today.

The second is its infrastructure software business, which came mainly from the 2023 acquisition of VMware for roughly $69 billion. VMware specializes in server virtualization and cloud software, and the deal transformed Broadcom from a pure-play hardware company into a hardware-plus-software conglomerate.

Adding the two together, its most recent quarter (Q2 FY2026) brought in total revenue of $22.187 billion. Semiconductors make up the bulk of that, with AI-related revenue now nearly half of the semiconductor segment. In other words, AI has become Broadcom's most important growth engine.

2. Why AI Cannot Do Without Broadcom

When people talk about AI, the instinctive reaction is "buy Nvidia GPUs." But if you ask Google, Meta, or OpenAI—the companies actually training the most advanced AI models—they will tell you they need something else in addition to general-purpose GPUs: chips custom-designed for specific tasks.

That is Broadcom's home turf.

Take Google as an example. Google began developing its own TPU (Tensor Processing Unit) in 2015—a chip purpose-built for AI workloads. Every generation of TPU has relied on Broadcom's design expertise. Over the past decade, Broadcom has co-built nine generations of TPUs for Google. This kind of long-term, deeply embedded partnership is Broadcom's deepest moat.

Beyond Google, Broadcom's custom AI chip customers also include Meta, ByteDance, and OpenAI. OpenAI's recently unveiled inference chip, Jalapeño, was co-developed with Broadcom. This 700-watt inference chip was benchmarked at Hot Chips 2026 and showed1.5–1.9× more AI work per watt than Nvidia's GB300, with 1.7–3.6× lower latency.

That is Broadcom's place in the value chain: Nvidia sells the "universal shovel," while Broadcom sells "task-specific shovels" designed for each customer. The two are not direct substitutes—they complement each other.

Industry estimates put Broadcom's share of the custom AI ASIC market at roughly 70%. In other words, for every ten custom AI chips designed globally, about seven involve Broadcom.

3. Why Has the Stock Dropped From $495 to $359?

Now that we understand the business, let's look at the price action.

Before June this year, Broadcom's stock hit an all-time high of $495, cementing its place in the trillion-dollar market-cap club. But over the past two-plus months, the stock has fallen as low as around $287, recently oscillating in the $350–$360 range. From peak to current, the drop has been more than 27%.

The question is: has Broadcom's fundamental business deteriorated?

The answer is really no. Its latest Q2 numbers were very strong: AI semiconductor revenue of $10.8 billion, up 143% year over year; new AI orders above $30 billion, roughly3× the quarter's actual shipments; and quarterly free cash flow above $10 billion. All of that tells us AI demand has not disappeared and Broadcom's business has not collapsed.

So why the savage drop? What the market is genuinely repricing is three new risks.

Risk #1: Google is cultivating a backup supplier. Google remains Broadcom's largest customer, with contracts running through 2031, and no one is replacing Broadcom in the short term. But Google is simultaneously deepening its partnership with Marvell (MRVL) on TPU-related chips, and rumors suggest it may also work with AMD on the tenth-generation TPU. That means Broadcom's "sole supplier" premium is gradually fading, and its pricing power is being diluted.

Risk #2: The new "backstop financing" model. Here is the simple version: Broadcom helps AI customers buy AI server racks. Investment institutions put up the money to purchase those racks, then lease them to the AI customers, who pay rent over five years. Broadcom provides a backstop guarantee on those leases—if the customer defaults, Broadcom must take over the equipment or sell it off.

The big problem with this model is that it potentially shifts all the risk back onto Broadcom. In the past, selling chips at worst meant customers paying late; now Broadcom also bears the customer's five-year credit risk and equipment depreciation risk. Market estimates put the maximum potential exposure from this backstop at up to $29 billion. That is the "worst case," not a guaranteed loss, but the figure alone has spooked the bond market—Broadcom's five-year credit default swap (CDS) spread has widened by 28 basis points recently, as investors effectively buy insurance against this potential risk.

Risk #3: Valuation. At the current price, Broadcom trades at roughly 30× trailing FY2026 earnings. On the surface that does not sound outrageous, but it is built on a very optimistic assumption: that AI semiconductor revenue will surpass $100 billion in FY2027 and earnings will grow 68% year over year. If that assumption does not hold, the multiple will get compressed quickly.

4. The September 2 Earnings: Five Things to Watch

Broadcom has set its full-year AI revenue target at roughly $56 billion. The first half delivered $19 billion, leaving $37 billion to go in the second half. The Q3 earnings report on September 2 is the moment of truth for whether that target is achievable.

There are at least five things to watch.

First, Q3 AI revenue. The company's own goal is $16 billion. Hitting exactly $16 billion only counts as "meeting the bar"; getting above $16.5 billion is what counts as beating expectations.

Second, Q4 guidance. If Q3 delivers just $16 billion, Q4 would need to reach $21 billion to hit the full-year target. If Q4 guidance comes in meaningfully below $20 billion, the full-year target is in trouble.

Third, will the company reiterate the FY2027 AI revenue target of $100 billion-plus, and can that growth come from multiple customers rather than relying solely on orders fueled by backstop financing?

Fourth, clarity on the backstop financing. How large is the guarantee? What happens to equipment if customers default? If management does not address this clearly, the stock will be discounted even if revenue beats expectations.

Fifth, earnings quality. Look at operating margin and cash flow. Some gross-margin compression is normal as hardware mix increases, but if both profits and cash flow deteriorate meaningfully, that is a real problem.

In short: AI revenue shows growth, backstop financing shows credit risk, and margins show earnings quality. All three need to pass at the same time for the print to count as solid.

5. So, Is Broadcom Worth a Look?

There is no single right answer, but here are three angles to consider.

If you focus on the long-term AI trend, Broadcom remains one of the most critical players in the supply chain. The custom AI chip space will only keep expanding, and Broadcom's technology depth and customer relationships run deep. BMO Capital Markets recently initiated coverage with an "Outperform" rating and a $455 price target, citing exactly this thesis.

If you focus on valuation margin of safety, the stock is indeed cheaper after a 20%-plus pullback from the highs. But it is not yet at "buy with eyes closed" levels—the very optimistic FY2027 growth assumption still has to play out.

If you focus on near-term timing, September 2 earnings is the major catalyst. Before the print, the stock will most likely keep chopping sideways in a range. After the print, depending on how those five key points land, the stock will find its next direction.

For risk-averse investors, the safer move now is probably to wait for the earnings print and for clarity on the backstop financing details before deciding. For investors who can tolerate meaningful volatility, the current price allows for a small starter position, but position size must be strictly controlled—no all-in bets.

Ultimately, Broadcom is a company with a genuinely strong business and a genuinely big story, but the risks the market is repricing right now are also genuinely real. To understand it properly, you cannot look only at the AI halo—you also have to look at customer concentration, the financing structure, and the competitive landscape. This is a classic AI-era company: worth following for the long term, but requiring caution in the short term.