# Broadcom

> Source: https://aiwiki.ai/wiki/broadcom
> Updated: 2026-06-22
> Categories: AI Companies, AI Hardware
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

**Broadcom Inc.** is an American semiconductor and infrastructure software company, traded on Nasdaq as AVGO, that has become the second most important supplier of AI compute after [Nvidia](/wiki/nvidia) by co-designing custom AI accelerators for the world's largest cloud companies and by supplying the Ethernet switch silicon that wires together their AI clusters. It co-designs the custom chips behind [Google](/wiki/google)'s [Tensor Processing Units](/wiki/tpu), [Meta](/wiki/meta)'s [MTIA](/wiki/mtia) chips, and [OpenAI](/wiki/openai)'s first in-house AI processors, and its Tomahawk and Jericho Ethernet switches stitch together many of the world's largest AI datacenters.[1][2] In its second quarter of fiscal 2026, ended May 3, 2026, Broadcom reported $10.8 billion of AI semiconductor revenue, up 143 percent year over year, and guided the following quarter to roughly $16 billion.[1]

Run since 2006 by president and CEO [Hock Tan](/wiki/hock_tan), the company pairs its chip business with an infrastructure software arm anchored by [VMware](/wiki/vmware). Broadcom generated $63.9 billion of revenue in fiscal 2025 (ended November 2, 2025), about $20 billion of it from AI semiconductors.[3] Its market capitalization passed $1 trillion for the first time in December 2024 and briefly exceeded $2 trillion in April 2026.[4][5]

## When and how was Broadcom founded?

Broadcom's corporate lineage runs through Hewlett-Packard rather than the original Broadcom Corporation whose name it bears. The business began in 1961 as HP Associates, Hewlett-Packard's semiconductor operation, passed to Agilent Technologies in HP's 1999 spinoff, and became a standalone company in 2005 when the private equity firms KKR and Silver Lake bought Agilent's Semiconductor Products Group for about $2.6 billion and named it Avago Technologies.[4] Hock Tan, a Malaysian-born executive trained at MIT and Harvard Business School, became CEO in 2006 and listed Avago on Nasdaq in August 2009.[4]

Tan built the company through serial acquisition: buying franchises with entrenched market positions, cutting costs sharply, and redeploying the cash flow into the next deal. The defining transaction closed in early 2016, when Avago paid $37 billion, $17 billion in cash and $20 billion in stock, for Broadcom Corporation, the communications chip maker founded in 1991 by Henry Samueli and Henry Nicholas, and adopted the better-known Broadcom name.[4]

| Closed | Target | Price | Business |
|---|---|---|---|
| 2014 | LSI Corporation | $6.6 billion | Storage and networking chips |
| 2016 | Broadcom Corporation | $37 billion | Communications semiconductors |
| 2017 | Brocade | $5.9 billion | Storage networking |
| 2018 | CA Technologies | $18.9 billion | Enterprise software |
| 2019 | Symantec enterprise security | $10.7 billion | Security software |
| 2023 | VMware | $69 billion | Virtualization and cloud software |

In November 2017 Broadcom launched a hostile bid for [Qualcomm](/wiki/qualcomm) worth well over $100 billion, which would have been the largest technology acquisition ever. President Donald Trump blocked the takeover by executive order in March 2018 on national security grounds, while Broadcom was still legally domiciled in Singapore; the company completed its redomiciliation to the United States shortly afterward and was renamed Broadcom Inc.[4] After the failed bid, Tan pivoted toward infrastructure software, a strategy that culminated in the acquisition of VMware for $69 billion including assumed debt, announced in May 2022 and closed on November 22, 2023.[4]

## What are Broadcom's custom AI accelerators?

Broadcom's accelerator business, which it brands XPUs, follows a co-design model. The customer defines the chip's architecture and owns the workload; Broadcom supplies the surrounding platform, including high-speed SerDes interfaces, die-to-die interconnect and advanced packaging, [high-bandwidth memory](/wiki/high_bandwidth_memory) integration, and the path to volume manufacturing at [TSMC](/wiki/tsmc).[2] The model lets hyperscalers field silicon tuned to their own models at lower unit cost than general-purpose GPUs, while Broadcom locks in multi-year, multi-billion-dollar programs.

Google is the oldest and largest customer. Broadcom has co-designed every generation of Google's TPUs in a partnership dating to about 2014, spanning seven generations through the Ironwood (TPU v7) chips Google unveiled in April 2025.[2] In April 2026 Broadcom disclosed a long-term agreement to develop and supply future TPU generations, together with a supply assurance covering networking and other components for Google's next-generation AI racks through 2031.[6][7] As part of the same expansion, [Anthropic](/wiki/anthropic) lined up roughly 3.5 gigawatts of next-generation TPU compute capacity beginning in 2027.[6][7]

Meta's MTIA (Meta Training and Inference Accelerator) family is likewise developed with Broadcom and runs ranking and recommendation inference at scale across Facebook and Instagram. In March 2026 Meta unveiled four new MTIA generations for rollout through 2027 on a roughly six-month cadence, and in April 2026 the two companies announced an extended partnership running through 2029 to support multi-gigawatt MTIA deployments, with Meta committing to more than one gigawatt of custom silicon in the first phase; Tan simultaneously gave up the Meta board seat he had held since 2024, moving to an advisory role on Meta's silicon roadmap.[2][8][25] [ByteDance](/wiki/bytedance) is the third widely identified custom-accelerator customer.[2]

Announcing fiscal 2024 results in December 2024, Tan told investors that Broadcom's three then-current XPU customers each planned to deploy clusters of up to one million accelerators by 2027, a serviceable market he sized at $60 billion to $90 billion for fiscal 2027 alone.[9] The roster has since grown. In September 2025 Broadcom disclosed a production order exceeding $10 billion from a fourth, unnamed customer; press speculation initially centered on OpenAI until semiconductor division president Charlie Kawwas confirmed the mystery buyer was not OpenAI, and analysts floated [Amazon Web Services](/wiki/amazon_web_services), [Oracle](/wiki/oracle), and [xAI](/wiki/xai) as candidates.[10] When Broadcom reported fiscal 2025 results on December 11, 2025, Tan revealed the customer was Anthropic, which had ordered Ironwood TPU rack systems supplied through Broadcom and then added a further $11 billion order during the fourth quarter, bringing its total commitment to roughly $21 billion for delivery through late 2026 and beyond.[22] By mid-2026 Broadcom described six major custom-compute customers, a roster that company statements and industry reporting identify as Google, Meta, ByteDance, OpenAI, Anthropic, and [Fujitsu](/wiki/fujitsu), whose 2 nanometer MONAKA server processor ships on Broadcom's 3.5D XDSiP packaging platform; [Apple](/wiki/apple) and a [SoftBank](/wiki/softbank)-[Arm](/wiki/arm) project have been cited as potential additions.[2][23][24]

## What is the OpenAI and Broadcom partnership?

Reuters reported in October 2024 that OpenAI was designing its first custom inference chip with Broadcom and TSMC.[11] The relationship became official on October 13, 2025, when the two companies announced a strategic collaboration to develop and deploy 10 gigawatts of OpenAI-designed AI accelerators and rack systems.[12] Under the agreement, OpenAI designs the accelerators and systems, embedding lessons from frontier model development directly into the hardware, while Broadcom co-develops them and supplies the networking stack; the racks are scaled entirely with Ethernet rather than [InfiniBand](/wiki/infiniband).[12] Deployments across OpenAI facilities and partner data centers are slated to begin in the second half of 2026 and to be completed by the end of 2029.[12]

[Sam Altman](/wiki/sam_altman) called the partnership "a critical step in building the infrastructure needed to unlock AI's potential."[12] Financial terms were not disclosed, and unlike OpenAI's contemporaneous compute deals with Nvidia and [AMD](/wiki/amd), the Broadcom agreement included no equity or investment component; Broadcom shares nevertheless rose about 9 percent on the announcement.[13] The first racks are built around 3 nanometer and 2 nanometer class chip designs.[2][12] On the June 2026 earnings call, Tan said initial production remained on track for late 2026, with a contractual commitment to deploy 1.3 gigawatts for OpenAI in 2027 as the first tranche of the 10 gigawatt program.[24]

## How does Broadcom's AI networking work?

Broadcom is the dominant merchant vendor of Ethernet switch silicon, selling to equipment makers such as Arista and Cisco as well as directly to hyperscalers, and it has positioned open Ethernet as the alternative to Nvidia's proprietary InfiniBand and NVLink fabrics for AI clusters. It was a founding member of the [Ultra Ethernet Consortium](/wiki/ultra_ethernet_consortium) in 2023, and between 2025 and 2026 it refreshed its entire networking line for AI workloads, covering scale-out (between racks), scale-up (within racks), and scale-across (between data centers) traffic.[14][15][16]

| Chip | Introduced | Role |
|---|---|---|
| Tomahawk 5 | 2022 | 51.2 Tbps scale-out switch used in early large GPU clusters |
| Jericho3-AI | 2023 | High-radix fabric router for AI clusters |
| Tomahawk 6 | June 2025 | First 102.4 Tbps Ethernet switch; supports clusters of up to one million XPUs [14] |
| Tomahawk Ultra | July 2025 | Low-latency 51.2 Tbps switch for scale-up networking, positioned against NVLink [15] |
| Jericho4 | 2025 | Fabric router connecting more than one million XPUs across multiple data centers [16] |
| Tomahawk 6 Davisson | October 2025 | First 102.4 Tbps switch with co-packaged optics [17] |
| Thor Ultra | October 2025 | First 800G AI Ethernet network interface card [15] |

Networking is a large and high-margin share of the AI franchise: it accounted for nearly 40 percent of Broadcom's AI revenue in the quarter ended May 3, 2026, a share Tan expected to settle nearer 30 percent as accelerator shipments ramp.[18][24] The OpenAI racks, Google's TPU pods, and many third-party clusters all rely on Broadcom Ethernet switching, so the company earns AI revenue even in deployments built around rival accelerators.[12][6] A 200 terabit per second successor to Tomahawk 6 taped out in mid-2026.[24]

## VMware and infrastructure software

After closing the VMware purchase in November 2023, Broadcom moved quickly to convert perpetual licenses into subscription bundles centered on VMware Cloud Foundation. The shift, accompanied by steep price increases, drew complaints from large customers and European cloud groups, but it transformed the segment's economics.[4] Infrastructure software revenue reached $7.2 billion in the quarter ended May 3, 2026, versus less than $2 billion per quarter for Broadcom's software business before the deal.[1] The acquisition also gave Broadcom its current headquarters: the company relocated to VMware's Palo Alto campus.[4]

## How big is Broadcom's AI revenue?

| Period | Total revenue | AI semiconductor revenue | AI growth (YoY) |
|---|---|---|---|
| FY2024 | $51.6 billion | $12.2 billion | +220% [3] |
| FY2025 | $63.9 billion | $20.0 billion | +65% [3] |
| Q1 FY2026 (ended Feb 1, 2026) | $19.3 billion | $8.4 billion | +106% [19] |
| Q2 FY2026 (ended May 3, 2026) | $22.2 billion | $10.8 billion | +143% [1] |
| Q3 FY2026 (guidance) | ~$29.4 billion | ~$16.0 billion | over +200% [1] |

Reporting fiscal 2025 results in December 2025, Broadcom disclosed an AI order backlog of $73 billion deliverable over roughly 18 months, nearly half of its total bookings, with Tan describing booking activity as unlike anything the company had seen before.[20] In March 2026 Tan told investors that Broadcom had "line of sight to achieve AI revenue from chips in excess of $100 billion in 2027," and on the June 2026 earnings call the company reaffirmed its full fiscal 2026 AI outlook of about $56 billion, up around 180 percent year over year, while reiterating the 2027 target; AI bookings exceeded $30 billion in the latest quarter, and Tan said demand visibility ran through 2028.[2][18][24] Q2 FY2026 adjusted EBITDA was $15.2 billion, or 69 percent of revenue, with GAAP net income of $9.3 billion.[1]

The AI ramp has driven extraordinary market-value milestones. In December 2024, after Tan first laid out the XPU opportunity, Broadcom shares surged 24 percent in a single session and the company became the 12th in the world to surpass a $1 trillion valuation.[4] On April 22, 2026, buoyed by the Google and Anthropic agreements, its market value briefly topped $2 trillion; in early June 2026 it stood near $1.8 trillion, placing Broadcom among the world's ten most valuable companies.[5][21]

The business carries distinctive risks: revenue is concentrated in a handful of hyperscale customers whose orders are lumpy, custom accelerators carry lower gross margins than Broadcom's merchant networking silicon, and Nvidia's full-stack platform and [Marvell](/wiki/marvell), the other major custom AI silicon house, compete for the same designs.[18] Even so, Broadcom's combination of custom compute, open Ethernet networking, and sticky infrastructure software has made it the most direct beneficiary after Nvidia of hyperscale AI capital spending.

## References

1. Broadcom Inc. "Broadcom Inc. Announces Second Quarter Fiscal Year 2026 Financial Results and Quarterly Dividend." PR Newswire, June 3, 2026. https://www.prnewswire.com/news-releases/broadcom-inc-announces-second-quarter-fiscal-year-2026-financial-results-and-quarterly-dividend-302790698.html
2. Tom's Hardware. "The custom AI ASIC state of play (May 2026): Broadcom deals, Google TPUs, Meta MTIA and beyond." May 2026. https://www.tomshardware.com/tech-industry/semiconductors/custom-ai-asics-examined-from-broadcom-to-mtia
3. Broadcom Inc. "Broadcom Inc. Announces Fourth Quarter and Fiscal Year 2025 Financial Results and Quarterly Dividend." PR Newswire, December 2025. https://www.prnewswire.com/news-releases/broadcom-inc-announces-fourth-quarter-and-fiscal-year-2025-financial-results-and-quarterly-dividend-302639606.html
4. Wikipedia. "Broadcom." https://en.wikipedia.org/wiki/Broadcom
5. Yahoo Finance (Barchart). "Broadcom Just Hit $2 Trillion Market Cap. Is AVGO Stock a Buy Now?" April 2026. https://finance.yahoo.com/markets/stocks/articles/broadcom-just-hit-2-trillion-185534906.html
6. Capacity. "Broadcom locks in long-term Google TPU deal through 2031." April 2026. https://capacityglobal.com/news/broadcom-locks-in-long-term-google-tpu-deal/
7. Yahoo Finance. "Broadcom's Long-Term Google TPU Deal Is Bigger Than It Looks for AI Infrastructure." April 2026. https://finance.yahoo.com/sectors/technology/articles/broadcom-long-term-google-tpu-155634924.html
8. CNBC. "Meta rolls out in-house AI chips weeks after massive Nvidia, AMD deals." March 11, 2026. https://www.cnbc.com/2026/03/11/meta-ai-mtia-chip-data-center.html
9. The Motley Fool. "Broadcom Sinks as CEO Becomes Defensive on AI Opportunity. Should Investors Buy the Dip?" June 7, 2025. https://www.fool.com/investing/2025/06/07/broadcom-sinks-ai-revenue-soars-buy-dip/
10. CNBC. "Broadcom's chip president says mystery $10 billion customer isn't OpenAI." October 13, 2025. https://www.cnbc.com/2025/10/13/broadcom-mystery-customer-openai.html
11. Reuters. "OpenAI builds first chip with Broadcom and TSMC, scales back foundry ambition." October 29, 2024. https://www.reuters.com/technology/openai-builds-first-chip-with-broadcom-tsmc-scales-back-foundry-ambition-2024-10-29/
12. OpenAI. "OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators." October 13, 2025. https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration/
13. CNBC. "Broadcom stock pops 9% on OpenAI custom chip deal, adding to Nvidia and AMD agreements." October 13, 2025. https://www.cnbc.com/2025/10/13/openai-partners-with-broadcom-custom-ai-chips-alongside-nvidia-amd.html
14. Broadcom Inc. "Broadcom Ships Tomahawk 6: World's First 102.4 Tbps Switch." June 2025. https://www.broadcom.com/company/news/product-releases/63146
15. Broadcom Inc. "Broadcom Advances Industry Support for Ethernet for AI Networking at the 2025 OCP Global Summit." October 2025. https://www.broadcom.com/blog/2025-ocp-global-summit-broadcom-advances-industry-support-for-ethernet-for-ai-networking
16. Broadcom Inc. "Broadcom Ships Jericho4, Enabling Distributed AI Computing Across Data Centers." 2025. https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-jericho4-enabling-distributed-ai-computing-across
17. Broadcom Inc. (GlobeNewswire). "Broadcom Announces Tomahawk 6 Davisson, the Industry's First 102.4-Tbps Ethernet Switch with Co-Packaged Optics." October 8, 2025. https://www.globenewswire.com/news-release/2025/10/08/3163429/19933/en/Broadcom-Announces-Tomahawk-6-Davisson-the-Industry-s-First-102-4-Tbps-Ethernet-Switch-with-Co-Packaged-Optics.html
18. SemiconAlpha. "Broadcom Q2 FY2026: The $100 Billion AI Trajectory Remains On Track." June 2026. https://semiconalpha.substack.com/p/broadcom-q2-fy2026-the-100-billion
19. Broadcom Inc. "Broadcom Inc. Announces First Quarter Fiscal Year 2026 Financial Results and Quarterly Dividend." PR Newswire, March 2026. https://www.prnewswire.com/news-releases/broadcom-inc-announces-first-quarter-fiscal-year-2026-financial-results-and-quarterly-dividend-302704490.html
20. Seeking Alpha. "Broadcom outlines $19.1B Q1 2026 revenue target as AI backlog surges to $73B." December 2025. https://seekingalpha.com/news/4530775-broadcom-outlines-19_1b-q1-2026-revenue-target-as-ai-backlog-surges-to-73b
21. StockAnalysis. "Broadcom (AVGO) Market Cap & Net Worth." Accessed June 9, 2026. https://stockanalysis.com/stocks/avgo/market-cap/
22. CNBC. "Broadcom reveals its mystery $10 billion customer is Anthropic." December 11, 2025. https://www.cnbc.com/2025/12/11/broadcom-reveals-its-mystery-10-billion-customer-is-anthropic.html
23. Broadcom Inc. (GlobeNewswire). "Broadcom Ships 3.5D Face-to-Face Compute SoC Powering AI Revolution." February 26, 2026. https://www.globenewswire.com/news-release/2026/02/26/3245680/19933/en/Broadcom-Ships-3-5D-Face-to-Face-Compute-SoC-Powering-AI-Revolution.html
24. The Motley Fool. "Broadcom (AVGO) Q2 2026 Earnings Call Transcript." June 3, 2026. https://www.fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/
25. CNBC. "Meta commits to one gigawatt of custom chips with Broadcom as Hock Tan agrees to leave board." April 14, 2026. https://www.cnbc.com/2026/04/14/meta-commits-to-one-gigawatt-of-custom-chips-with-broadcom-as-hock-tan-agrees-to-leave-board.html