![Broadcom AI Chip Revenue Hits $16.7B, Up 221% [2026]](https://tech-insider.org/wp-content/uploads/2026/10/tmp82bpcjzj.webp)
Broadcom’s AI semiconductor business just posted its biggest quarter yet. The chipmaker’s fiscal third-quarter 2026 results, released September 2, showed AI semiconductor revenue of $16.7 billion, up 221% year-over-year and 54% sequentially, according to the company’s earnings release and reporting from Benzinga and Investing.com. Total revenue hit $29.59 billion, beating the $29.36 billion Wall Street consensus, while adjusted earnings of $3.32 a share topped estimates of roughly $3.24.
Despite the beat, Broadcom shares slid in the days that followed, and Wall Street analysts split sharply on what comes next. The quarter also revealed new detail on exactly whose AI chips Broadcom is building, including OpenAI’s “JalapeÃąo” accelerator and Google and Anthropic’s Ironwood TPU v7. Here’s what the numbers mean, why the stock reaction looked paradoxical, and how Broadcom’s custom-silicon business now stacks up against Nvidia and AMD.
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 Q3 Fiscal 2026 Numbers, Explained
Broadcom’s fiscal third quarter, which closed in early August and was reported September 2, 2026, delivered revenue of $29.59 billion against a consensus estimate of $29.36 billion, according to Benzinga’s earnings coverage. Adjusted earnings per share came in at $3.32, ahead of the roughly $3.24 Wall Street expected. The headline number, though, was AI semiconductor revenue: $16.7 billion, now representing approximately 56% of total company revenue, per the same Benzinga report.
CEO Hock Tan framed the quarter around the company’s custom-chip customer base rather than any single product. “Our six XPU customers are accelerating the adoption of custom accelerators and AI semiconductor revenue more than tripled year-over-year to $16.7 billion,” Tan said, according to Broadcom’s Q3 2026 earnings call transcript. That “six XPU customers” framing is new language from Broadcom and signals the custom-chip roster has widened beyond the handful of hyperscalers the company has previously named in public.
Why AI Semiconductor Revenue Jumped 221% Year-Over-Year
Tan was direct about the scale of the jump. “Q3 AI semiconductor revenue of $16.7 billion grew 221% year-over-year, and 54% quarter-over-quarter,” he said, per the same Broadcom earnings release quoted by Futurum Group’s analysis of the results. Working backward from that 54% sequential growth figure implies Q2 fiscal 2026 AI semiconductor revenue of roughly $10.8 billion, underscoring how quickly the custom-silicon ramp has accelerated within a single fiscal year.
The growth is being driven by shipment volume, not just average selling price. Broadcom’s custom AI processor shipments rose approximately 3.5 times year-over-year in the quarter, according to a Yahoo Finance analysis of the earnings. That same analysis found custom silicon now accounts for nearly three-quarters of Broadcom’s total AI semiconductor revenue (custom XPUs were reported at approximately 73% of AI revenue on October 7, 2026), meaning most of the $16.7 billion quarterly figure came from application-specific chips rather than networking components, a notable shift from March 2026, when Broadcom said AI networking silicon still made up roughly one-third of total AI revenue, underscoring how quickly custom ASICs have outpaced networking as the segment’s primary growth driver.
Custom Silicon vs Merchant GPUs: Inside Broadcom’s ASIC Model
Broadcom doesn’t compete with Nvidia or AMD by selling general-purpose AI GPUs on the open market. Instead, it designs application-specific integrated circuits, or ASICs, tailored to a single customer’s workload, plus the networking silicon that links thousands of those chips together inside a data center. The Yahoo Finance report describes this as a fundamentally different niche than the GPU businesses run by Nvidia and AMD, since Broadcom’s chips are built to order for companies like Alphabet’s Google, Meta Platforms, OpenAI, and Anthropic rather than sold as a standardized product line.
That build-to-order model means Broadcom’s revenue growth is tied directly to how aggressively a small number of hyperscalers are willing to commit capital to custom silicon programs, rather than to broad market demand the way Nvidia’s GPU sales are. When one of those customers accelerates a rollout, as several appear to have done this quarter, the revenue swing shows up fast and large, which helps explain the 54% sequential jump.
Broadcom’s Six XPU Customers: OpenAI, Meta, Google and Anthropic
Broadcom has historically been reluctant to name its custom-silicon customers directly, but the Q3 call and surrounding coverage confirmed several by name. Tan’s “six XPU customers” comment, combined with the Yahoo Finance report naming Google, Meta, OpenAI, and Anthropic as custom-silicon partners, gives the clearest public picture yet of who is actually buying Broadcom-designed chips.
Morningstar’s coverage of the earnings call added a forward-looking detail: management confirmed that OpenAI and Anthropic are expected to become Broadcom’s two largest customers starting in 2027, according to the Morningstar analysis of the call. That would mark a shift away from Google, long considered Broadcom’s anchor TPU customer, toward the newer wave of AI-lab-driven custom silicon programs.
Inside the Chips: JalapeÃąo, MTIA and Ironwood TPU v7
The quarter’s most specific disclosure concerned OpenAI’s first custom accelerator, one Broadcom and OpenAI had already unveiled publicly on June 24, 2026, months before this shipment update. “In Q3, we also shipped JalapeÃąo, OpenAI’s first generation custom accelerator, which outperforms Grace Blackwell GPU for inference workloads,” Tan said, according to the earnings call transcript. The comparison to Nvidia’s Grace Blackwell platform is notable because it’s a direct, management-stated performance claim about an inference workload rather than a marketing benchmark from a third party, though it has not been independently verified in public benchmarks.
Tan also addressed Meta’s program directly: “Shipments of JalapeÃąo for OpenAI will continue, and for Meta, we expect production shipments of their custom MTIA accelerator optimized for inference and recommendation at scale,” he said on the same call. On the Google and Anthropic side, Tan said, “During the quarter, we delivered Ironwood TPU v7, version 7 in high volume to both Anthropic and Google,” confirming that Anthropic has joined Google as a direct recipient of Broadcom-built TPU silicon rather than simply renting TPU capacity through Google Cloud.
Guidance: A $21.7 Billion Q4 and $58 Billion Full-Year AI Target
Broadcom guided fiscal Q4 2026 AI semiconductor revenue to approximately $21.7 billion, which the Yahoo Finance analysis calculates as roughly 236% year-over-year growth. For the full fiscal year, management is now targeting approximately $58 billion in AI semiconductor revenue, up from an earlier expectation of around $56 billion, and described internally as 186% higher than the prior fiscal year, according to Investing.com’s coverage of the guidance.
Morningstar’s reporting on the call noted management went further still, pointing to bullish guidance through 2028 that includes AI revenue quadrupling from 2026 levels and non-GAAP earnings per share nearly tripling to $30, while Investing.com reported projections of $115 billion in fiscal 2027 AI semiconductor revenue and $230 billion in fiscal 2028. Those multi-year targets, if realized, would make AI semiconductors Broadcom’s dominant business line well beyond its traditional networking and broadband chip segments.
Why Broadcom Stock Fell Despite Record Results
Morningstar’s headline on the move, “Broadcom Stock Fell 3% After Earnings Despite Immense AI Chip Demand,” captures the disconnect: a company growing AI revenue 221% year-over-year still saw its stock punished because the market had priced in even steeper near-term guidance. It’s a pattern that has repeated across AI-chip earnings all year, where the bar for “beat” has risen faster than most companies’ actual results.
Wall Street’s Split Reaction to the Earnings Beat
Analyst reaction to the print was unusually divided for a quarter this strong. Most firms raised price targets, but Bank of America went the other direction, a split that reflects genuine disagreement about how to value a business this dependent on a handful of customers.
| Firm / Analyst | Rating | Prior Target | New Target | Date |
|---|---|---|---|---|
| Cantor Fitzgerald (C.J. Muse) | Overweight | $525 | $600 | Sept 3, 2026 |
| BMO Capital (Harsh Kumar) | Outperform | $455 | $575 | Sept 3, 2026 |
| Rosenblatt (Sajal Dogra) | Buy | $500 | $600 | Sept 3, 2026 |
| Bank of America (Vivek Arya) | Buy | $530 | $460 | Sept 3, 2026 |
| Goldman Sachs | Not specified | $525 | $540 | Sept 3, 2026 |
| Bernstein | Buy (reiterated) | $550 | $575 | Sept 3â24, 2026 |
Figures per Coinpaper’s reporting on the Bank of America note and MarketBeat’s September 24 coverage of Bernstein’s reiteration. Bank of America’s Vivek Arya kept a Buy rating while cutting his target, citing near-term guidance concerns even as the firm’s own note pointed to a roughly $230 billion long-range AI revenue forecast for Broadcom’s addressable market. By late September, MarketBeat put the broader analyst consensus at a “Moderate Buy” with an average target of $527.20, while an earlier snapshot from financefeeds.com put the street average at $511.88 across a panel of 25 Buy ratings and 3 Hold ratings.
Broadcom vs Nvidia vs AMD: Comparing the AI Silicon Race
Broadcom, Nvidia, and AMD are often lumped together as “AI chip stocks,” but their business models diverge sharply, which makes direct revenue comparisons misleading. The table below lays out what’s publicly confirmed about each company’s approach as of this reporting window.
| Company | Core AI Hardware Model | Named Platforms / Chips | Confirmed Revenue Data Point | Named AI Customers |
|---|---|---|---|---|
| Broadcom | Custom ASICs (“XPUs”) plus AI networking silicon | JalapeÃąo (OpenAI), MTIA (Meta), Ironwood TPU v7 (Google/Anthropic) | $16.7B AI semiconductor revenue in Q3 FY2026, guided to $21.7B in Q4 | OpenAI, Meta, Google, Anthropic |
| Nvidia | Merchant general-purpose AI GPUs and full-stack platforms | Vera Rubin platform (Vera CPU, Rubin GPU, NVLink 6, ConnectX-9, BlueField-4, Spectrum-6) | Not disclosed in Broadcom-comparable terms for this exact period | Broad hyperscaler and AI-lab customer base |
| AMD | GPUs plus an expanding custom-silicon practice | Instinct accelerator line; data-center CPU/GPU portfolio | Data-center revenue run rate estimated near $25B for 2026, including ~$12.5B in H1 2026 | Public cloud and enterprise data-center customers |
The AMD figures come from the same Yahoo Finance analysis that broke down Broadcom’s custom-silicon mix, which directly contrasted Broadcom’s roughly $58 billion full-year AI semiconductor target against AMD’s smaller but still fast-growing data-center run rate. Nvidia’s comparable figure for this exact period was not available in a form directly comparable to Broadcom’s AI semiconductor segment reporting, so no specific number is presented here; Nvidia’s Vera Rubin platform details come from separate semiconductor-industry trade coverage rather than a head-to-head quarterly disclosure. Readers interested in Nvidia’s own custom-chip pressure can see tech-insider.org’s coverage of Nvidia’s Hugging Face acquisition for how the company is responding to the same custom-silicon trend from the other direction.
Historical Context: How Fast Broadcom’s AI Business Has Grown
Broadcom’s AI semiconductor business has compounded quickly within a single fiscal year. The 54% sequential growth disclosed this quarter implies a prior-quarter base in the $10.8 billion range, building on an even earlier base of $8.4 billion in AI semiconductor revenue, up 106% year-over-year, that Broadcom reported back in March 2026, which lines up with Broadcom’s own earlier-fiscal-year disclosures covered previously on tech-insider.org’s report on Broadcom’s AI chip revenue climb. Hock Tan’s commentary about “six XPU customers” also marks an expansion from the smaller customer set Broadcom described in earlier quarters, a trajectory consistent with how custom-silicon demand has broadened across the AI industry through 2026.
That growth trajectory mirrors a broader pattern across the AI infrastructure buildout this year, from AWS’s order for 2 million additional Nvidia GPUs to Qualcomm and Amazon’s long-term AI data-center chip partnership. Each of these deals points to the same underlying dynamic: hyperscalers and AI labs are committing multi-year capital to secure dedicated silicon supply, and Broadcom’s custom-ASIC backlog is one of the clearest financial expressions of that spending so far.
Market Impact: What This Means for the Broader AI Supply Chain
Broadcom’s results land at a moment when AI capital spending is under more scrutiny than at any point since the current buildout began. Meta’s own AI infrastructure spending, detailed in tech-insider.org’s coverage of Zuckerberg’s AI spending commitments, is one of the demand sources feeding Broadcom’s MTIA shipments, which means Broadcom’s guidance is effectively a real-time signal of how committed hyperscalers remain to their own AI capex plans.
The same logic applies to AMD’s recent push into AI-adjacent markets, including its $8.2 billion acquisition of World Labs, and to the market-cap stakes laid out in tech-insider.org’s analysis of Nvidia’s market-cap lead over Apple. If Broadcom’s $58 billion full-year AI target holds, it would cement custom ASICs as a permanent second track in AI infrastructure spending, running alongside rather than replacing merchant GPU purchases from Nvidia and AMD.
Expert and Analyst Perspective
Beyond the numbers, Broadcom’s own executives offered the clearest window into how the company sees its position. Hock Tan’s repeated emphasis on named customer shipments, rather than aggregate demand language, was a notable departure from how Broadcom has historically discussed its custom-silicon business on earnings calls, where customer identities were often withheld for competitive reasons.
That shift toward specificity, confirming JalapeÃąo for OpenAI, MTIA for Meta, and Ironwood TPU v7 for Google and Anthropic, suggests Broadcom’s custom-chip relationships have matured to the point where naming customers no longer carries the same competitive risk it once did. It also gives investors a far more direct way to track Broadcom’s AI revenue against the public capex disclosures of those same four companies going forward.
What Comes Next: 5 Predictions for Broadcom’s AI Chip Business
- Q4 guidance will be the next stock-moving event. With $21.7 billion in AI semiconductor revenue already guided, any sign Broadcom is tracking ahead of or behind that number will likely move the stock more than the Q3 beat itself did.
- OpenAI and Anthropic’s share of revenue will keep climbing. Morningstar’s reporting that both are expected to become Broadcom’s two largest customers by 2027 suggests the customer mix will keep shifting away from Google as the single largest anchor account.
- Analyst price targets will keep diverging before they converge. The gap between Bank of America’s $460 target and Cantor Fitzgerald’s $600 target reflects a real disagreement about customer concentration risk that won’t resolve until at least another quarter or two of results.
- Custom silicon will keep growing faster than merchant GPU spending at the margin. The 3.5x year-over-year jump in custom AI processor shipments suggests hyperscalers are increasingly willing to commit to single-customer chip programs rather than relying solely on off-the-shelf GPUs.
- Networking silicon will become a bigger part of the AI chip story. As XPU clusters scale, the networking components that connect them, an area where Broadcom already has a strong franchise, are likely to get more investor attention relative to the compute chips themselves.
Frequently Asked Questions
What was Broadcom’s AI semiconductor revenue in Q3 fiscal 2026?
Broadcom reported AI semiconductor revenue of $16.7 billion for its fiscal third quarter of 2026, up 221% year-over-year and 54% sequentially, according to the company’s earnings release reported by Benzinga and Futurum Group.
Why did Broadcom stock fall after a record quarter?
Investors weighed the Q4 guidance against already-elevated expectations, even though the Q3 results themselves beat consensus estimates.
What is Broadcom’s JalapeÃąo chip?
JalapeÃąo is described by Broadcom CEO Hock Tan as OpenAI’s first-generation custom accelerator, which he said outperforms Nvidia’s Grace Blackwell GPU on inference workloads, according to the company’s Q3 2026 earnings call transcript. Broadcom and OpenAI first unveiled the chip on June 24, 2026, with initial deployment targeted for December 2026 and broader scaling planned afterward, according to CNBC.
Which companies does Broadcom build custom AI chips for?
Based on the Q3 2026 earnings call and related reporting, Broadcom’s confirmed custom-silicon customers include OpenAI (JalapeÃąo), Meta (MTIA), and Google and Anthropic (Ironwood TPU v7), with CEO Hock Tan referencing “six XPU customers” in total.
What is Broadcom’s full-year AI revenue guidance for fiscal 2026?
Broadcom is guiding toward approximately $58 billion in AI semiconductor revenue for fiscal 2026, up from an earlier roughly $56 billion estimate, according to Investing.com’s coverage of the earnings call.
How is Broadcom’s AI chip business different from Nvidia’s?
Broadcom designs custom, single-customer ASICs and AI networking silicon, while Nvidia sells general-purpose AI GPUs and platforms like Vera Rubin to a broad base of customers, according to Yahoo Finance’s analysis of Broadcom’s business model.
What is Ironwood TPU v7?
Ironwood TPU v7 is the seventh-generation tensor processing unit that Broadcom CEO Hock Tan said was delivered in high volume to both Google and Anthropic during the fiscal third quarter of 2026, per the earnings call transcript.
Did analysts raise or cut their Broadcom price targets after earnings?
Most did raise targets, including Cantor Fitzgerald (to $600), BMO Capital (to $575), and Rosenblatt (to $600), but Bank of America cut its target from $530 to $460 even while maintaining a Buy rating, according to Benzinga and Coinpaper.
