Broadcom told Wall Street on June 3, 2026 that its artificial intelligence chip business generated $10.8 billion in a single quarter, up 143% from a year earlier. The number beat every estimate on the Street. Investors sold the stock anyway, sending shares down 12.59% the next day and wiping out roughly $280 billion in market value in one session.
That contradiction, a record quarter met with a sharp selloff, is the story of Broadcom’s fiscal 2026 so far. The company has become the clearest proof that hyperscalers no longer want to buy every AI chip from Nvidia. Google, Meta, ByteDance, OpenAI and Anthropic are now among the customers reported to be designing custom silicon with Broadcom’s help, and bookings for that work climbed past $30 billion in the June quarter alone, building on the $73 billion, 18-month AI semiconductor backlog Broadcom had already disclosed for its fiscal fourth quarter of 2025, as CRN reported in December 2025. Here is what actually happened in the numbers, why traders reacted the way they did, and how Broadcom’s custom-chip business stacks up against Nvidia and AMD heading into fiscal 2027.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Broadcom’s Q2 Numbers: $10.8 Billion and Counting
Broadcom’s fiscal second quarter of 2026, the three months ended May 3, 2026, closed with total revenue of $22.187 billion, up 48% year over year. AI semiconductor revenue, the line item Wall Street now watches most closely, came in at $10.8 billion, up 143% from the same quarter last year. Non-GAAP earnings per share landed at $2.44, while GAAP EPS was $1.91, according to CNBC’s coverage of the June 3 earnings report.
CEO Hock Tan framed the quarter as a story about custom chips and networking gear, not general-purpose processors. “Q2 semiconductor revenue from AI of $10.8 billion grew 143% year-over-year, above our forecast, driven by increasing demand for custom AI accelerators and AI networking,” Tan said, according to Yahoo Finance’s earnings writeup.
Cash generation kept pace with the growth. Broadcom produced $10.493 billion in operating cash flow and $10.262 billion in free cash flow, equal to 46% of quarterly revenue. For fiscal Q3 2026, the company had guided to roughly $29.4 billion in total revenue and about $16.0 billion in AI semiconductor revenue, a forecast Reuters also flagged in its June 2026 write-up of the print. That guidance turned out to be conservative: when Broadcom reported fiscal Q3 2026 results for the quarter ended in August 2026, AI semiconductor revenue actually came in at $16.7 billion, up 221% year over year, according to Wall Street Insights and Futunn’s coverage of the September 2026 print. Broadcom also guided fiscal Q4 2026 AI semiconductor revenue to about $21.7 billion, implying growth of roughly 236% year over year, per the same September 2026 reporting.
Inside the $30 Billion AI Chip Backlog
The more telling figure sat below the headline revenue line. Broadcom said bookings for AI semiconductors ran “over $30 billion against the $10.8 billion we shipped,” Tan told analysts, according to a transcript published by The Motley Fool. That single-quarter figure builds on top of the $73 billion AI semiconductor backlog, covering roughly 18 months of future demand, that Broadcom had already disclosed for its fiscal fourth quarter of 2025, according to CRN’s December 2025 reporting. That is roughly three dollars of new orders for every dollar shipped in the quarter, a ratio that suggests Broadcom’s constraint right now is manufacturing capacity, not customer demand.
A backlog that size does more than pad next quarter’s number. It gives Broadcom, and the customers waiting on chips, a multi-year production schedule to plan around, since custom silicon orders typically lock in wafer capacity, advanced packaging slots and high-bandwidth memory more than a year ahead of delivery.
What Counts as “AI Semiconductor Revenue”
Broadcom’s AI number is narrower than it sounds. It combines two product lines: custom AI accelerators, often called XPUs, built to a specific customer’s chip design, and AI networking silicon, including the Tomahawk switch and Jericho router chips that move data between thousands of accelerators inside a data center. It does not include Broadcom’s broader semiconductor portfolio, such as broadband, wireless and storage chips, or its VMware software segment, which is reported separately.
Why Broadcom Stock Fell on a Beat
A 143% jump in AI revenue would normally send a stock higher, not lower. Broadcom shares instead closed down 12.59% at $418.91 on June 4, 2026, the trading day after the report. The move erased an estimated $280 billion in market capitalization in a single session, even as the stock’s overall market cap was later reported above $2.1 trillion following a rebound later that same week.
The reaction was less about the quarter and more about expectations. Broadcom shares had already run up sharply into the print, and traders had priced in guidance that raised the full-year AI outlook. Instead, management held its fiscal 2027 AI semiconductor revenue target at “in excess of $100 billion” rather than lifting it, which some investors read as a pause after a year of upward revisions — a pause that proved temporary once Broadcom raised that same fiscal 2027 target to $115 billion in a roadmap update reported by Investing.com in September 2026. A backlog growing faster than shipments can also unsettle short-term traders, even when it signals a healthy multi-year pipeline, because it pushes revenue recognition further into the future than a straight beat-and-raise quarter would.
Wall Street’s Price Targets After the Dust Settled
Sell-side analysts read the quarter differently than the stock’s initial move suggested. Several raised their price targets within days of the report, treating the selloff as a buying opportunity rather than a warning sign.
| Bank | Analyst | New Price Target | Prior Target | Note |
|---|---|---|---|---|
| Bank of America | Vivek Arya | $530 | $450 | Raised after the June 3 print |
| Jefferies | Blayne Curtis | $550 | Not disclosed | Called the pullback a buying opportunity |
| UBS | Timothy Arcuri | $490 | Not disclosed | Pre-earnings note |
The spread between targets, from $490 to $550, gives a rough sense of where Wall Street expects Broadcom stock to trade if the AI backlog converts into shipped revenue on schedule. Both Bank of America and Jefferies made their calls explicitly in response to the post-earnings drop, arguing the fundamentals hadn’t changed even though the stock price had.
What Is an XPU, and Why Hyperscalers Want One
An XPU is Broadcom’s shorthand for a custom AI accelerator designed to a single customer’s specifications, rather than a general-purpose GPU sold to anyone. Google’s TPU line is the best-known example built this way, with Broadcom acting as the design and manufacturing partner behind the silicon.
The appeal for a hyperscaler is straightforward economics. A chip built for one company’s specific model architecture can strip out features it doesn’t need and tune the parts it uses constantly, which can lower the cost of each unit of compute at the scale hyperscalers operate. The tradeoff is flexibility. Nvidia’s GPUs run on CUDA, a software stack with over a decade of developer tooling behind it, while a custom XPU generally locks a customer into a narrower software path built around Broadcom’s design. That’s a real cost, but for a company running the same handful of models across a million-chip fleet, it’s often a cost worth paying.
The Customer List: Google, Meta, ByteDance, OpenAI and Anthropic
Broadcom doesn’t break out AI revenue by customer, and Tan has repeatedly declined to confirm names on earnings calls. Even so, reporting around the June 3 quarter tied Broadcom’s custom-chip pipeline to at least five hyperscale customers: Google, Meta, ByteDance, OpenAI and Anthropic. Coverage of the quarter specifically referenced a large-scale compute agreement tied to Anthropic, adding a fifth major name to a customer list that had previously centered on Google’s TPU program.
Treat that list as a floor, not a ceiling. Broadcom’s own disclosures describe “several” additional prospective customers in its pipeline without naming them, and the company has historically waited for its customers to confirm chip partnerships publicly before commenting itself.
Broadcom vs Nvidia vs AMD: The AI Chip Numbers Side by Side
Broadcom’s AI semiconductor revenue is still a fraction of Nvidia’s data center business, but it is growing faster in percentage terms than either of its two largest rivals in AI silicon. Nvidia’s own fiscal Q1 2027 release put its Data Center revenue at $75.2 billion, while AMD’s investor relations disclosures put its Data Center segment at $5.8 billion for the same general period.
| Company | Segment | Quarter | Revenue | YoY Growth | Reported |
|---|---|---|---|---|---|
| Nvidia | Data Center | Fiscal Q1 2027 (ended late April 2026) | $75.2 billion | +92% | May 20, 2026 |
| Broadcom | AI Semiconductor | Fiscal Q2 2026 (ended May 3, 2026) | $10.8 billion | +143% | June 3, 2026 |
| AMD | Data Center | Q1 2026 (ended March 2026) | $5.8 billion | +57% | May 5, 2026 |
The comparison isn’t perfectly apples to apples. Nvidia’s and AMD’s “Data Center” segments include broader server hardware and networking beyond pure AI silicon, while Broadcom’s figure is a narrower, specifically AI-tagged slice of its total semiconductor business. Even accounting for that, the growth-rate gap is wide enough to matter: Broadcom’s AI revenue is compounding at a rate that would double the business roughly every seven months if sustained, versus a doubling period closer to nine months for Nvidia and over 15 months for AMD’s data center unit. For a detailed spec and pricing breakdown of where AMD’s own AI accelerators stand against Nvidia’s lineup, see our AMD MI350P vs Nvidia H200 comparison.
Custom Silicon vs Merchant GPUs: A Market Splitting in Two
Broadcom isn’t the only company chasing hyperscalers away from off-the-shelf GPUs. Marvell has built a similar custom-silicon business with its own hyperscale customers, and Qualcomm entered the data center CPU market this year with its Dragonfly C1000 chip, landing Meta as a customer in the process. Our coverage of the Dragonfly C1000 launch has more on that deal.
The bigger pattern is that AI compute is splitting into two distinct markets rather than one. Merchant GPUs, led by Nvidia and challenged by AMD’s Instinct line, remain the default choice for companies training frontier models or renting cloud compute on demand. Custom ASICs are pulling ahead for a narrower but very large use case: hyperscalers running the same inference workloads, at massive and predictable scale, for years at a time. Both markets are growing. They’re just growing for different reasons, and Broadcom’s backlog is evidence that the custom side of that split is no longer a niche.
From Avago to VMware: How Broadcom Got Here
Broadcom’s current form traces back to Avago Technologies’ 2016 acquisition of the original Broadcom Corporation, after which the combined company adopted the more recognizable Broadcom name. What followed was a decade of deals that looked more like a private equity roll-up than a typical chipmaker’s roadmap: enterprise software from CA Technologies in 2018, Symantec’s enterprise security business in 2019, and finally VMware, in a deal that closed on November 22, 2023 at a headline value of roughly $61 billion in cash and stock, or about $69 billion once VMware’s net debt is included.
That software business now throws off steady, high-margin cash flow that funds Broadcom’s chip investments without leaning as hard on debt or equity markets as some AI-chip peers. Meanwhile, the custom-silicon relationship with Google’s TPU program, one of the industry’s earliest big bets on application-specific AI chips, quietly built the design and manufacturing expertise Broadcom is now applying to Meta, ByteDance, OpenAI and Anthropic. The company that used to be known mainly for networking chips and Wi-Fi components is now, by revenue growth rate, one of the fastest-growing AI chip suppliers on the planet.
The Road to $100 Billion by Fiscal 2027
Broadcom’s AI semiconductor revenue has climbed every quarter for more than a year, growing from roughly $20 billion in full-year fiscal 2025 (itself up 65% year over year, a figure Broadcom disclosed in November 2025 and CIO Dive cited that December) to a fiscal 2026 pace that keeps outrunning the company’s own guidance.
| Period | AI Semiconductor Revenue | YoY Growth | Status |
|---|---|---|---|
| Fiscal 2025 (full year) | ~$20 billion (implied) | — | Actual, derived from FY26 guidance commentary |
| Q1 FY2026 (reported Mar. 4, 2026) | $8.4 billion | +106% | Actual |
| Q2 FY2026 (reported Jun. 3, 2026) | $10.8 billion | +143% | Actual |
| Q3 FY2026 (guidance) | ~$16.0 billion | >200% | Guidance |
| FY2026 (full-year guidance) | $56 billion | ~180% vs FY2025 | Guidance |
| FY2027 (target) | >$100 billion | — | Reiterated Jun. 3, 2026 |
Tan repeated the $100 billion fiscal 2027 figure across back-to-back earnings calls before eventually raising it. Back in March, after Q1 FY2026 results for the quarter that ended in early February 2026 showed AI semiconductor revenue of $8.4 billion, up 106% year over year, according to a Motley Fool transcript of the March 4, 2026 call, he told analysts, “We now have a clear path to achieve AI revenue from chips alone exceeding $100 billion,” according to CNBC’s report on that call. Three months later, after the Q2 print, he stuck with it: “We expect this momentum to continue into fiscal year 2027 and reiterate our AI semiconductor revenue guidance to be in excess of $100 billion,” per CNBC’s June 3 report. By September 2026, following the fiscal Q3 beat, Broadcom raised its full-year fiscal 2026 AI semiconductor revenue guidance to $58 billion from $56 billion and, per Investing.com’s reporting that month, laid out a roadmap targeting $115 billion in fiscal 2027 AI semiconductor revenue and $230 billion in fiscal 2028. Hitting the fiscal 2027 number would mean AI chips alone generate more annual revenue than Broadcom’s entire company did just three years earlier. Supply, particularly high-bandwidth memory, will likely decide whether that timeline holds. Our coverage of the Nvidia-SK Hynix memory supply agreement covers how tight that market already is.
What This Means for Nvidia’s Grip on AI Compute
Nvidia isn’t losing ground in absolute terms. A $75.2 billion quarter, up 92% year over year, is not the profile of a company under threat. But Broadcom’s 143% growth rate, more than 50 percentage points faster than Nvidia’s, shows where the incremental AI infrastructure dollar is increasingly headed: toward chips designed for one buyer’s workload instead of general-purpose GPUs sold to everyone.
Nvidia’s moat, its CUDA software ecosystem and its position as the default choice for training new models, remains largely intact. What’s changing is the inference side of the business, where a hyperscaler running the same model billions of times a day has both the volume and the engineering resources to justify a custom chip. That’s the wedge Broadcom, Marvell and now Qualcomm are all pushing into, and Broadcom’s backlog suggests the wedge is getting wider, not narrower.
Market Impact Beyond Broadcom
A $30 billion chip backlog doesn’t stay contained to one company’s balance sheet. It ripples through the entire AI hardware supply chain. Broadcom’s custom chips still need advanced packaging and leading-edge manufacturing capacity, adding to demand already stretched thin by Nvidia and AMD orders. Our recent coverage of TSMC’s capex plans shows how much of that capacity crunch is already showing up in foundry spending.
The backlog also adds to data center construction and leasing demand, since custom AI accelerators still need somewhere to run. That connects directly to the wave of hyperscaler leasing activity we covered in our report on $850 billion in data center leases, and to the power-grid strain regulators are now trying to manage, detailed in our piece on the FERC order giving grid operators 60 days to address AI data center power demand. Every custom chip Broadcom ships is, eventually, a rack that needs power, cooling and floor space somewhere.
Five Predictions Through Fiscal 2027
- Broadcom hits or beats its $16.0 billion Q3 guidance. That figure already implies growth of more than 200% year over year, and the company has beaten its own AI revenue forecast in each of the last two quarters, with a backlog three times the size of quarterly shipments giving it more room to do it again.
- Supply, not demand, becomes the binding constraint. With bookings running well ahead of shipments, packaging capacity and high-bandwidth memory availability, not customer orders, will most likely determine how fast Broadcom can convert its backlog into revenue.
- At least one more hyperscale customer goes public. Broadcom’s pipeline commentary about undisclosed prospects suggests the current five-name customer list (Google, Meta, ByteDance, OpenAI, Anthropic) grows before fiscal 2027 is over.
- Nvidia’s growth rate slows relative to custom silicon, even as its dollar lead stays enormous. Expect the percentage-growth gap between Nvidia and the custom-ASIC players to persist, even though Nvidia’s data center revenue will likely remain several multiples larger than Broadcom’s AI segment through fiscal 2027.
- Price targets keep climbing if Q3 guidance holds. With targets already spanning $490 to $550 after one guidance reiteration, a beat-and-raise quarter in September would likely push the high end of that range past $600.
Frequently Asked Questions
What was Broadcom’s AI semiconductor revenue in fiscal Q2 2026?
Broadcom reported $10.8 billion in AI semiconductor revenue for fiscal Q2 2026 (the quarter ended May 3, 2026), up 143% year over year, on total company revenue of $22.187 billion. The results were announced June 3, 2026.
What is Broadcom’s guidance for fiscal Q3 2026 AI semiconductor revenue?
Broadcom had guided to about $16.0 billion in AI semiconductor revenue for fiscal Q3 2026, representing growth of more than 200% year over year. When the quarter, which ended in August 2026, was actually reported in September 2026, AI semiconductor revenue came in even higher, at $16.7 billion, up 221% year over year, according to Wall Street Insights and Futunn’s coverage of the print.
Why did Broadcom stock fall after beating earnings estimates?
Shares had run up heavily heading into the report, and management reiterated rather than raised its fiscal 2027 AI revenue target of over $100 billion. That was enough to trigger profit-taking, sending the stock down 12.59% to $418.91 the next trading day, even though the quarter itself beat forecasts.
What is Broadcom’s $30 billion AI chip backlog?
It refers to confirmed bookings for AI semiconductors that Broadcom has not yet shipped. Hock Tan disclosed the figure on the June 3 earnings call, noting bookings topped $30 billion against $10.8 billion actually shipped in the quarter.
Who are Broadcom’s custom AI chip customers?
Broadcom doesn’t officially confirm customer names, but reporting around the June 2026 quarter ties its custom-chip business to Google, Meta, ByteDance, OpenAI and Anthropic, including a large-scale compute agreement connected to Anthropic.
How does Broadcom’s AI chip revenue compare to Nvidia’s?
Nvidia’s Data Center segment generated $75.2 billion in its fiscal Q1 2027, roughly seven times Broadcom’s $10.8 billion AI semiconductor figure for a similar period. Broadcom’s revenue is growing faster in percentage terms (143% versus 92%), but from a much smaller base.
What is Broadcom’s fiscal 2027 AI revenue target?
Broadcom initially guided to AI semiconductor revenue “in excess of $100 billion” in fiscal 2027, a target CEO Hock Tan reiterated on the June 3, 2026 earnings call after first laying it out in similar terms during the March 2026 Q1 call. By September 2026, that target had been raised: Investing.com reported a Broadcom roadmap pointing to $115 billion in fiscal 2027 AI semiconductor revenue and $230 billion in fiscal 2028, up from a newly raised $58 billion full-year fiscal 2026 guide.
What is an XPU and how is it different from a GPU?
An XPU is Broadcom’s term for a custom AI accelerator designed for one customer’s specific workload, unlike Nvidia’s general-purpose GPUs, which are built to run a wide range of software for many different buyers. Google’s TPU line is the best-known chip family built through this kind of partnership.
