COMPANY PROFILE / AI & MACHINE LEARNING
Broadcom
A chip company that builds customer-specified AI chips (XPUs) and the Ethernet fabric around them, and has begun to take part in financing AI racks
Broadcom is not an AI pure play: it has two businesses, semiconductors and infrastructure software (VMware). The AI-related part is the custom AI accelerators (XPUs) that customers, namely very large cloud companies and frontier AI developers, design to their own specifications, plus the Ethernet switches, NICs and optical connectivity that link data centers together. For the third quarter of fiscal 2026 (ended 2 August 2026), the company's management said AI semiconductor revenue was US$16.7 billion, up 221% on the year. That number appears in management's remarks in earnings releases, not in a reported category of the 10-K. From June 2026 a new arrangement, the AI XPV platform, also began: financial partners take on the purchase and leasing of AI racks, and Broadcom backstops the customer's lease obligations. This profile sets out what Broadcom sells, where its sales are concentrated and what the new funding arrangements mean, within its filings and official releases. Everything is checked as of 15 October 2026.
- Broadcom in 30 seconds, and where it stands in AI and machine learning
- The logo and how the name is written
- Where the company sits in the AI and machine learning field
- Scale (limited to the AI-related business)
- Resources committed to AI and machine learning
- What it has done so far
- What comes next
- What the company is aiming for in AI and machine learning
- Map of partnerships in AI and machine learning
- Technology, products and services
- The risks this company carries
- Glossary / References / Claim-to-source audit
1. Broadcom in 30 seconds, and where it stands in AI and machine learning
Tomahawk 6Also Ethernet NICs, PHYs, optical components and PCIe switches
Broadcom's "AI revenue" is a figure given quarter by quarter in management's remarks in earnings releases, not in a fixed category of the 10-K or 10-Q Sourced. The 10-K and 10-Q say that custom AI accelerators and AI networking drove the rise in semiconductor solutions revenue, but they do not give an AI revenue amount or definition. In the third quarter of FY2026, direct sales to a single distributor were 50% of company revenue, and the top five end customers reached about 55% Sourced. This profile therefore keeps track of where each number is stated.
2. The logo and how the name is written
A note on names: the legal entity is Broadcom Inc., usually called Broadcom. The technology and products of the earlier Broadcom Corporation have been folded in, and the 10-K traces the company's history back to AT&T/Bell Labs, Lucent and Hewlett-Packard. Tomahawk (Ethernet switches), Thor (NICs), Jericho (router silicon) and Davisson (co-packaged optics) are product names, not company names.
Official site: https://www.broadcom.com/3. Where the company sits in the AI and machine learning field
4. Scale (limited to the AI-related business)
Broadcom also has large non-AI businesses (wireless, broadband, industrial and infrastructure software such as VMware), so this section centres on the semiconductor solutions segment. Its fiscal year ends on the Sunday closest to 31 October: FY2025 ran from 4 November 2024 to 2 November 2025, and FY2026 ends on 1 November 2026. Figures are as filed in the 10-K and 10-Q Sourced.
| Metric (US$ million) | FY2024 (to 3 Nov 2024) | FY2025 (to 2 Nov 2025) |
|---|---|---|
| Revenue (company-wide) | 51,574 | 63,887 |
| Semiconductor solutions | 30,096 | 36,858 |
| Infrastructure software | 21,478 | 27,029 |
| Semiconductor solutions operating income | 16,759 | 21,232 |
| Research and development (company-wide) | 9,310 | 10,977 |
| Operating income (company-wide) | 13,463 | 25,484 |
| FY2026 so far (US$ million) | Third quarter (to 2 Aug 2026) | Same quarter a year earlier (to 3 Aug 2025) | First three quarters |
|---|---|---|---|
| Revenue (company-wide) | 29,591 | 15,952 | 71,089 |
| Semiconductor solutions | 20,839 | 9,166 | 48,363 |
| Infrastructure software | 8,752 | 6,786 | 22,726 |
| Research and development (company-wide) | 2,895 | 3,050 | 8,855 |
| Operating income (company-wide) | 15,955 | 5,887 | 35,306 |
The 10-K attributes the rise in semiconductor solutions revenue to "strong demand for our networking solutions, primarily custom AI accelerators and AI networking products". AI-related revenue is described by management in earnings releases as follows (management remarks, not a category of the 10-K or 10-Q).
| Earnings release (date) | Remark on AI semiconductor revenue | Year-on-year | Note |
|---|---|---|---|
| Q4 FY2025 (11 December 2025) | No amount stated | Up 74% | First-quarter outlook of US$8.2 billion (company outlook) |
| Q1 FY2026 (4 March 2026) | US$8.4 billion | Up 106% | Second-quarter outlook of US$10.7 billion |
| Q2 FY2026 (3 June 2026) | US$10.8 billion | Up 143% | Third-quarter outlook of US$16.0 billion |
| Q3 FY2026 (2 September 2026) | US$16.7 billion | Up 221% (up 54% on the quarter) | Fourth-quarter outlook of US$21.7 billion (up 236%) |
The "AI semiconductor revenue" above is a management remark, and the release text, 10-K and 10-Q give no definition of which products it includes that we could confirm Sourced. The figures are also rounded. For reference, third-quarter AI semiconductor revenue of US$16.7 billion is about 80% of that quarter's semiconductor solutions revenue (US$20.8 billion) and about 56% of company revenue (US$29.6 billion) (our calculation, approximate, from rounded figures) Our calculation. Every outlook is the company's own and unconfirmed (the fourth quarter of FY2026 ends on 1 November 2026): it says revenue of about US$34.8 billion and AI semiconductor revenue of about US$21.7 billion Not yet confirmed.
5. Resources committed to AI and machine learning
The table below sets out the four resource points (people, money, outside investments and acquisitions, and ties to universities and public bodies) and whether each could be confirmed in the 10-K, the 10-Q, Form 8-Ks and company releases.
| Type of resource | Details | Status |
|---|---|---|
| People | About 33,000 employees (2 November 2025), about 57% of them in R&D roles. By region: North America 49%, Asia 36%, Europe, Middle East and Africa 15%. No AI-specific headcount is disclosed | Sourced |
| Money: R&D | US$10,977 million in FY2025 (17% of revenue) against US$9,310 million a year earlier; US$8,855 million in the first three quarters of FY2026 | Sourced |
| Money: capital expenditure | US$623 million of purchases of property and equipment in FY2025 (US$548 million the year before). The 10-K expects it to be higher in FY2026 (an expectation, not a result) | Sourced |
| Money: purchase commitments | Legally binding unconditional purchase commitments (primarily inventory; those with a remaining term of one year or less are excluded) were about US$126.8 billion at 2 August 2026, against about US$0.13 billion at the end of FY2025. About US$52.7 billion falls in FY2027 and about US$73.0 billion in FY2028 | Sourced |
| Money: debt | Outstanding indebtedness of US$67.1 billion at the end of FY2025 (10-K). The 10-Q says it repurchased US$5.6 billion of senior notes in the third quarter of FY2026 | Sourced |
| Outside investments and acquisitions | No AI-related business acquisition was found for FY2025 (acquisitions in the cash flow statement are nil). It acquired VMware (2023) and others earlier, but not for AI specifically | Not confirmed |
| Guarantees for a customer's debt | In the third quarter of FY2026 a financial partner took on the purchase of AI racks and the related leases to a customer, and Broadcom backstops the customer's lease obligations over five years. Maximum exposure once all racks are deployed is about US$29 billion. The customer may issue Broadcom convertible notes of up to US$42 billion, but none had been issued at 2 August 2026 | Sourced |
| The AI XPV platform | A framework with financial partners (Apollo and Blackstone's Credit & Insurance business as initial anchor investors) to enable more than 20 gigawatts of compute through 2028. The first tranche is US$35 billion, supporting Anthropic's expansion of more than 1 gigawatt at Fluidstack-based sites from mid-2026 (9 June 2026) | Sourced |
| Universities and public bodies | No joint research with universities or receipt of public funding (grants) was found in the 10-K, 10-Q or releases. Involvement in industry bodies (OCP, the Ultra Ethernet Consortium) is mentioned in company releases | Not confirmed |
The table below lists the main products and agreements, as far as company materials confirm them (checked 15 October 2026).
| Product or agreement | Scale / content | Stage | Date | Reference |
|---|---|---|---|---|
| Tomahawk 6 (Ethernet switch) | 102.4 Tbps; 100G and 200G SerDes; 512 XPUs in scale-up and more than 100,000 XPUs in a two-tier scale-out network (Broadcom's stated values) | Shipping in production volume (company statement) | 12 March 2026 | Reference 14, 15 https://investors.broadcom.com/news-releases/news-release-details/broadcom-now-shipping-worlds-first-1024-tbps-switch-production |
| Tomahawk 6 - Davisson (co-packaged optics switch) | 102.4 Tbps; sixteen 6.4 Tbps optical engines; 70% lower optical interconnect power than pluggable optics (Broadcom's stated value) | Announced as shipping; the release also says samples are going to early-access customers | 8 October 2025; third generation shown at OCP in October 2026 | Reference 15, 16 https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024 |
| Custom AI accelerators for OpenAI (10 gigawatts) | OpenAI designs the accelerators and systems, developed and deployed with Broadcom. Rack deployment targeted to start in the second half of 2026 and finish by the end of 2029 | Term sheet signed; long-standing co-development and supply agreements | 13 October 2025 | Reference 10 https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-announce-strategic-collaboration-deploy-10 |
| Jalapeño (OpenAI's first AI chip) | Engineering samples running in the lab. The first chip of a multi-generation platform; initial deployment planned by the end of 2026 | Samples running (company statement); deployment is a plan | 24 June 2026 | Reference 11 https://www.globenewswire.com/news-release/2026/06/24/3316887/19933/en/openai-and-broadcom-unveil-llm-optimized-intelligence-processor.html |
| Custom Tensor Processing Units (TPUs) with Google | A long-term agreement to develop and supply custom TPUs for future generations, and a supply assurance agreement for networking and other components for Google's next-generation AI racks (up to 2031) | Agreement | 6 April 2026 | Reference 7 https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm |
| Custom silicon (MTIA) with Meta | The initial commitment exceeds 1 gigawatt, the first phase of a multi-gigawatt rollout; plans extend through 2029 | Agreement (rollout planned) | 14 April 2026 | Reference 12 https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-extended-partnership-meta-deploy-technology |
| Compute capacity for Anthropic | From 2027 Anthropic will access about 3.5 gigawatts of TPU-based compute through Broadcom. Consumption depends on Anthropic's continued commercial success | Expanded collaboration (8-K); in discussions with operational and financial partners | 6 April 2026 | Reference 7 https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm |
| Custom ASICs with Apple | Development and supply for multiple generations of Apple products through 2031. AI use is not stated | Agreement | 6 July 2026 | Reference 9 https://www.sec.gov/Archives/edgar/data/1730168/000119312526295589/d84378d8k.htm |
A gigawatt is the electrical scale of a data center, not a count of chips or an amount of revenue. "Term sheet" and "plan" do not mean quantities or amounts are fixed.
6. What it has done so far
| When | What happened | Why it matters |
|---|---|---|
| November 2023 | Completed the VMware acquisition (per the 10-K) | Greatly expanded infrastructure software (not AI-specific) |
| 8 October 2025 | Announced Tomahawk 6 - Davisson (co-packaged optics) | A product that rethinks optical connectivity around power |
| 13 October 2025 | Announced a 10-gigawatt custom AI accelerator collaboration with OpenAI | A rack-inclusive deployment plan, at the term-sheet stage |
| 11 December 2025 | Fourth-quarter FY2025 results. Management said AI semiconductor revenue rose 74% on the year | AI growth drove the increase in company revenue |
| 12 March 2026 | Announced Tomahawk 6 in production volume | The company says it went from samples to production in under three quarters |
| 6 April 2026 | Disclosed in an 8-K the long-term agreement with Google (through 2031) and the expanded collaboration for Anthropic | A large agreement disclosed with named customers |
| 14 April 2026 | Announced the extended collaboration with Meta (MTIA, more than 1 gigawatt) | Broadcom claims the industry's first 2nm AI compute accelerator |
| 9 June 2026 | Set up the AI XPV platform with Apollo and Blackstone (first tranche US$35 billion) | A chip company is taking part in customers' funding arrangements |
| 24 June 2026 | Unveiled Jalapeño with OpenAI (described as nine months from design to tape-out) | Engineering samples of a customer-designed chip running |
| 6 July 2026 | Disclosed in an 8-K the long-term agreements with Apple (through 2031) | Longer commitments on custom ASICs, including non-AI uses |
| 2 September 2026 | Third-quarter FY2026 results. AI semiconductor revenue of US$16.7 billion | Fourth-quarter outlook of US$21.7 billion |
7. What comes next
The company presents these as outlooks and targets, not as achieved facts Not yet confirmed. The 10-Q says new business models such as selling and leasing AI racks could affect demand and margins, and that customers underestimating data center power or water capacity could affect deliveries.
None of the following could be confirmed from the company's 10-K, 10-Q, 8-Ks or releases Not yet confirmed. The definition and product breakdown of AI semiconductor revenue / AI revenue by customer (the 10-Q describes the 50% distributor and the 55% top five end customers anonymously) / prices and unit shipments of XPUs and networking products / contract values with OpenAI, Meta and Anthropic / the size and timing of later tranches of the AI XPV platform. Customer-by-customer amounts that appear in press coverage are not used here.
8. What the company is aiming for in AI and machine learning
This section includes points the company does not state directly. They are inferences from what the primary sources say, given with their grounds.
| What can be read from it | Primary-source basis | Status |
|---|---|---|
| Broadcom is getting involved in customers' funding and deployment, beyond supplying chips | The AI XPV platform (financial partners buy the racks and Broadcom backstops the customer's lease obligations). The 10-K lists selling or leasing AI racks and systems based on XPUs as a new business model | Inference |
| Dependence on a few large customers is deepening (a fact this article can confirm) | In the third quarter of FY2026 one distributor was 50% of company revenue and the top five end customers about 55% (about 40% in FY2025) | Sourced |
| It is building AI connectivity around Ethernet | The OpenAI release stresses combining custom accelerators with Ethernet, and the product line runs from scale-up (Tomahawk Ultra) to optical connectivity (Davisson) | Inference |
| Long-term agreements aim to lock in several generations of supply through 2029 to 2031 | Google (through 2031), Apple (through 2031), Meta (through 2029) and OpenAI (by the end of 2029). Contract values and terms are not disclosed | Inference |
| Manufacturing will remain outsourced | The 10-K says about 95% of the wafers manufactured by its contract manufacturers were made by TSMC (also about 95% in the first three quarters of FY2026) | Sourced |
The "Recent Developments" section of the 10-Q says demand for AI is strong but that customers need to access significant capital, and then explains that under the AI XPV platform financial partners provide the funding and Broadcom may give residual value guarantees where necessary Sourced. In practice a backstop on a customer's lease obligations (up to about US$29 billion) and convertible notes the customer can issue (up to US$42 billion) have been put in place. This article reads that as the company seeing its customers' financial capacity as something that can set the pace of chip sales. However, the company says it believes the probability of these contingent liabilities arising is low, and expected losses cannot be judged from the 10-Q Not yet confirmed.
9. Map of partnerships in AI and machine learning
10. Technology, products and services
| Layer | Main products | What the 10-K and company materials say |
|---|---|---|
| Custom AI accelerators (XPUs) | ASICs to a customer's specification, integrating logic, HBM, SerDes and IP cores with advanced packaging | The 10-K says ASICs are built to customers' specifications for hyperscalers, companies with AI frontier models and system integrators, and in some cases used in racks or systems |
| Ethernet switches | Tomahawk 6, Tomahawk Ultra, Jericho 4 | For scale-up, scale-out and scale-across connectivity (OCP release) |
| NICs and PHYs | Thor Ultra (800G), Thor 2 (400G), DSP-based Ethernet transceiver chips | AI Ethernet NICs. The 10-K describes the PHYs as built on a proprietary DSP architecture |
| Optical connectivity | Co-packaged optics (TH6-Davisson), optical components, lasers | Optical engines sit in the same package as the switch. Uses TSMC's COUPE technology (Broadcom release) |
| PCIe switches | For AI and non-AI uses | Connections inside servers (10-K) |
| AI racks and systems | Racks based on XPUs (for OpenAI and others) | The 10-K treats selling or leasing them as a new business model |
The table below sets out networking products using the developer's stated values from Broadcom's releases (checked 15 October 2026).
| Product (Broadcom stated values) | Main figures and features | Date and stage |
|---|---|---|
| Tomahawk 6 | 102.4 Tbps switching capacity; 100G and 200G SerDes; 512 200G or 1,024 100G SerDes on one chip; 512 XPUs in scale-up and more than 100,000 XPUs in a two-tier scale-out network (200 Gbps per link) | In production volume since 12 March 2026 |
| Tomahawk 6 - Davisson | Third-generation co-packaged optics: sixteen 6.4 Tbps optical engines, 200 Gbps per link. Claims 70% lower optical interconnect power than traditional pluggable optics (more than 3.5x lower). IEEE 802.3 compliant | 8 October 2025. Samples to early-access customers |
| MTIA for Meta | Broadcom claims the industry's first 2nm AI compute accelerator. The initial commitment exceeds 1 gigawatt | 14 April 2026. Rollout is planned |
Source: Broadcom press releases. Values are the developer's own, and the measurement conditions and comparisons are as given in each release. This article makes no comparison with other companies' products and no claim of superiority. "Industry first" and "70% lower" are Broadcom's claims, and independent verification was not checked here. The 8 October 2025 release says in its opening that the product is "now shipping" but says under availability that samples are going to early-access customers, so the two statements are not consistent.
What is inside a custom chip: the 10-K describes its custom silicon platform as integrating "embedded logic, high-bandwidth memory, serializer/deserializer technology, IP cores and processor cores using advanced packaging technologies" Sourced. This suggests that how compute, memory and input/output are combined to a customer's specification is the heart of the design (see our explainers on HBM, chiplets, CoWoS and heterogeneous integration).
Manufacturing concentration: the 10-K and 10-Q say about 95% of the wafers made by contract manufacturers were produced by TSMC Sourced. The 10-Q notes TSMC could prioritise other customers or raise prices. On materials, the 10-Q lists memory chips, silicon, gallium arsenide and indium phosphide wafers, copper lead frames, precious and rare earth metals and others as inputs it depends on.
Connecting with light: Broadcom describes TH6-Davisson as putting the optical engines in the same package as the switch chip, integrated with TSMC's COUPE technology Sourced. The aim is to reduce signal loss and the variability of making and testing pluggable transceivers, and to lower power. The "70% lower power" figure is Broadcom's claim, and the basis of comparison could not be checked here (see our explainers on co-packaged optics, optical transceivers, and optical interconnects and SerDes).
Seen in units of power: Broadcom talks about XPU deployments in gigawatts, and describes the AI XPV platform as "more than 20 gigawatts". For AI compute, the power a data center can accept appears to be becoming the unit of deployment rather than the number of chips (see our explainer on AI data center power and cooling).
11. The risks this company carries
| Risk | Details | Status |
|---|---|---|
| Customer concentration | In the third quarter of FY2026 one distributor was 50% of revenue (46% in the first three quarters, against 32% and 30% a year earlier). The top five end customers were about 55% (about 50% in the first three quarters, against about 40% a year earlier). The 10-Q says losing any of the top five or a large drop in their demand could have a material effect | Sourced |
| New funding structures | The backstop on a customer's lease obligations is up to about US$29 billion, and the customer may issue convertible notes of up to US$42 billion. It may provide residual value guarantees. The 10-Q says it believes the probability of these arising is low (the company's view) | Sourced |
| Surge in purchase commitments | Unconditional purchase commitments are about US$126.8 billion (about US$0.13 billion at the end of FY2025). If demand falls short of expectations, inventory and costs could be affected | Sourced |
| Manufacturing concentration | About 95% of wafers come from TSMC. The 10-K and 10-Q note the possibility of price rises or changed priorities | Sourced |
| Timing of design wins | Bringing several large design wins such as custom AI accelerators into production at once could strain resources (10-Q). New business models such as selling and leasing AI racks could affect demand and margins | Sourced |
| Regulation and export controls | The 10-Q says U.S. export rules (the EAR) and additions to restricted-entity lists have affected, and may affect, revenue, supply chain and its ability to manufacture or sell, described in general terms. 17% of FY2025 revenue was shipped or delivered to China (including Hong Kong). No charge tied to a specific product control was found in what this article read | Sourced |
| Investigations and inquiries | The 10-Q says it has been involved in inquiries from regulators in Korea, Japan and the EU into certain contracting and business practices. No outcome or amount is stated. This article does not predict the result | Sourced |
| Debt | Outstanding indebtedness of US$67.1 billion at the end of FY2025. An 8-K discloses cash tender offers for debt in June 2026 | Sourced |
12. Glossary
- XPU
- A general term for custom AI processors, including those that are not GPUs. Broadcom supports their design and manufacture.
- ASIC
- A chip made for a specific purpose to a customer's specification.
- Ethernet
- A communication method widely used in data centers. Broadcom promotes it for AI connectivity too.
- Scale-up / scale-out
- Linking chips inside a rack is scale-up; linking racks and sites is scale-out or scale-across.
- SerDes
- Input/output circuitry that sends data at high speed between chips.
- Co-packaged optics (CPO)
- Putting optical transmit-and-receive parts in the same package as the switch chip.
- Backstop
- A guarantee under which Broadcom would cover part of a customer's payments if the customer defaulted.
- Gigawatt (GW)
- A unit for the electrical scale of a data center. Not a count of chips or an amount of revenue.
- Term sheet
- A document setting out main terms that is not a final legal agreement.
- 10-K / 10-Q / 8-K
- The annual, quarterly and current reports filed with the U.S. SEC.
13. References
- Broadcom Form 10-K for the fiscal year ended November 2, 2025 (filed with the U.S. SEC on December 18, 2025). Business, segment revenue, customer concentration, headcount, R&D, capital expenditure, contract manufacturing and risk factors. https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm
- Broadcom Form 10-Q for the quarterly period ended August 2, 2026 (filed with the U.S. SEC on September 10, 2026). Revenue for the third quarter and first three quarters of fiscal 2026, customer concentration, the AI XPV platform, the guarantee relating to AI racks, commitments and risk factors. https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm
- Broadcom Broadcom Inc. Announces Third Quarter Fiscal Year 2026 Financial Results and Quarterly Dividend (press release furnished with Form 8-K, September 2, 2026). Third-quarter AI semiconductor revenue and the fourth-quarter outlook. https://www.sec.gov/Archives/edgar/data/1730168/000173016826000076/avgo-08022026x8kxex99.htm
- Broadcom Broadcom Inc. Announces Second Quarter Fiscal Year 2026 Financial Results and Quarterly Dividend (furnished with Form 8-K, June 3, 2026). https://www.sec.gov/Archives/edgar/data/1730168/000173016826000051/avgo-05032026x8kxex99.htm
- Broadcom Broadcom Inc. Announces First Quarter Fiscal Year 2026 Financial Results and Quarterly Dividend (furnished with Form 8-K, March 4, 2026). https://www.sec.gov/Archives/edgar/data/1730168/000173016826000011/avgo-02012026x8kxex99.htm
- Broadcom Broadcom Inc. Announces Fourth Quarter and Fiscal Year 2025 Financial Results and Quarterly Dividend (furnished with Form 8-K, December 11, 2025). https://www.sec.gov/Archives/edgar/data/1730168/000173016825000116/avgo-11022025x8kxex99.htm
- Broadcom Form 8-K (filed April 6, 2026). The long-term agreement with Google for custom Tensor Processing Units (TPUs) and a supply assurance agreement, and the expanded collaboration with Anthropic. https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm
- Broadcom Form 8-K (filed June 18, 2026). Results of cash tender offers for debt securities. https://www.sec.gov/Archives/edgar/data/1730168/000119312526275077/d149683d8k.htm
- Broadcom Form 8-K (filed July 6, 2026). The long-term agreements with Apple, extending custom ASIC development and supply through 2031. https://www.sec.gov/Archives/edgar/data/1730168/000119312526295589/d84378d8k.htm
- Broadcom / OpenAI OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators (joint press release, October 13, 2025; read in a browser). https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-announce-strategic-collaboration-deploy-10
- Broadcom / OpenAI OpenAI and Broadcom Unveil LLM-Optimized Intelligence Processor (joint press release, June 24, 2026; read in a browser). https://www.globenewswire.com/news-release/2026/06/24/3316887/19933/en/openai-and-broadcom-unveil-llm-optimized-intelligence-processor.html
- Broadcom / Meta Broadcom Announces Extended Partnership with Meta to Deploy Technology to Support Multi-Gigawatts of Meta's Custom Silicon, MTIA (joint press release, April 14, 2026; read in a browser). https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-extended-partnership-meta-deploy-technology
- Broadcom / Apollo / Blackstone Broadcom, Apollo, and Blackstone Establish Landmark Strategic Platform to Accelerate More Than 20 Gigawatts of Global AI Deployments (joint press release, June 9, 2026; the same text as hosted on Apollo's site). https://www.apollo.com/insights-news/pressreleases/2026/06/broadcom--apollo--and-blackstone-establish-landmark-strategic-pl
- Broadcom Broadcom Now Shipping World's First 102.4 Tbps Switch in Production Volume (press release, March 12, 2026; read in a browser). https://investors.broadcom.com/news-releases/news-release-details/broadcom-now-shipping-worlds-first-1024-tbps-switch-production
- Broadcom Broadcom Announces Tomahawk 6 - Davisson, the Industry's First 102.4-Tbps Ethernet Switch with Co-Packaged Optics (press release, October 8, 2025; read in a browser). https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024
- Broadcom Broadcom Redefines AI Infrastructure with Industry-Leading Networking Innovations at 2026 OCP Global Summit (press release, October 8, 2026; read in a browser). https://www.globenewswire.com/news-release/2026/10/08/3377533/19933/en/broadcom-redefines-ai-infrastructure-with-industry-leading-networking-innovations-at-2026-ocp-global-summit.html
- Broadcom Official website (source of the logo image: https://www.broadcom.com/img/broadcom-logo.svg; checked October 15, 2026). https://www.broadcom.com/
14. Claim-to-source audit
| Claim in the text | Status | Reference |
|---|---|---|
| Revenue, segment revenue and operating income, R&D and operating income for FY2024 to FY2025 | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm |
| About 33,000 employees (57% in R&D), capital expenditure, TSMC's share, FY2025 customer concentration (distributor 32%, top five about 40%) | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm |
| The description of custom silicon and Ethernet products, and the risk of selling and leasing AI racks as a new business model | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm |
| Revenue, segment revenue and R&D for the third quarter and first three quarters of FY2026 | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm |
| Customer concentration (distributor 50%, top five about 55%) and purchase commitments of about US$126.8 billion | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm |
| The AI XPV platform, the backstop of about US$29 billion and the customer's convertible notes of up to US$42 billion | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm |
| Third-quarter FY2026 AI semiconductor revenue of US$16.7 billion (management remark) and the fourth-quarter outlook | Sourced | Reference 3 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000076/avgo-08022026x8kxex99.htm |
| Second-quarter FY2026 AI semiconductor revenue of US$10.8 billion (management remark) | Sourced | Reference 4 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000051/avgo-05032026x8kxex99.htm |
| First-quarter FY2026 AI semiconductor revenue of US$8.4 billion (management remark) | Sourced | Reference 5 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000011/avgo-02012026x8kxex99.htm |
| The rise in AI semiconductor revenue in the fourth quarter of FY2025 (up 74%) and the first-quarter outlook of US$8.2 billion | Sourced | Reference 6 https://www.sec.gov/Archives/edgar/data/1730168/000173016825000116/avgo-11022025x8kxex99.htm |
| The long-term agreement with Google and the expanded collaboration with Anthropic (about 3.5 gigawatts) | Sourced | Reference 7 https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm |
| The long-term agreements with Apple (through 2031) | Sourced | Reference 9 https://www.sec.gov/Archives/edgar/data/1730168/000119312526295589/d84378d8k.htm |
| The June 2026 tender offers for debt | Sourced | Reference 8 https://www.sec.gov/Archives/edgar/data/1730168/000119312526275077/d149683d8k.htm |
| The 10-gigawatt collaboration with OpenAI (term sheet; second half of 2026 to the end of 2029) | Sourced | Reference 10 https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-announce-strategic-collaboration-deploy-10 |
| The Jalapeño announcement (samples running; initial deployment planned by the end of 2026) | Sourced | Reference 11 https://www.globenewswire.com/news-release/2026/06/24/3316887/19933/en/openai-and-broadcom-unveil-llm-optimized-intelligence-processor.html |
| The extended collaboration with Meta (MTIA, more than 1 gigawatt, through 2029) | Sourced | Reference 12 https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-extended-partnership-meta-deploy-technology |
| Setting up the AI XPV platform (9 June 2026; US$35 billion first tranche; more than 20 gigawatts) | Sourced | Reference 13 https://www.apollo.com/insights-news/pressreleases/2026/06/broadcom--apollo--and-blackstone-establish-landmark-strategic-pl |
| Tomahawk 6 in production volume (12 March 2026) | Sourced | Reference 14 https://investors.broadcom.com/news-releases/news-release-details/broadcom-now-shipping-worlds-first-1024-tbps-switch-production |
| Tomahawk 6 - Davisson specifications and the power-reduction claim (8 October 2025) | Sourced | Reference 15 https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024 |
| Products shown at OCP 2026 (Tomahawk Ultra, Jericho 4, Thor Ultra, third-generation Davisson) | Sourced | Reference 16 https://www.globenewswire.com/news-release/2026/10/08/3377533/19933/en/broadcom-redefines-ai-infrastructure-with-industry-leading-networking-innovations-at-2026-ocp-global-summit.html |
| Source of the logo image | Sourced | Reference 17 https://www.broadcom.com/ |
| The four company types and Broadcom's placement | Our calculation | This article's own classification. Company names are examples from firms covered in this series |
| AI semiconductor revenue at about 80% of semiconductor solutions and about 56% of company revenue (Q3 FY2026) | Our calculation | Our calculation (16.7 / 20.839 and 16.7 / 29.591). AI revenue is a rounded management remark |
| The fourth-quarter FY2026 outlook, deployments for OpenAI, Meta and Anthropic, the size of the AI XPV platform and Jalapeño deployment | Not yet confirmed | Company outlooks, plans and term sheets, not settled facts. Items that could not be confirmed are stated in the text. Reference 2 https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm |
| The reading that it is getting involved in customers' funding and deployment | Inference | This article's interpretation of the 10-K and 10-Q |
| The reading that it is building AI connectivity around Ethernet | Inference | This article's interpretation of the official press releases |
| The reading that long-term agreements aim to lock in several generations of supply | Inference | This article's interpretation of the 8-Ks and press releases |
| The materials engineer's view on custom chip makeup, manufacturing, optical connection and power | Commentary | General technical commentary, not a Broadcom disclosure. Figures are the company's stated values from official releases |
Last updated 15 October 2026 / Troy Technical
Figures in this article come from Broadcom's statutory filings with the U.S. Securities and Exchange Commission (the 10-K, 10-Q, 8-Ks and earnings releases) and from official press releases issued by Broadcom and its co-announcers. No research-firm estimates or press-based figures have been used. "AI semiconductor revenue" is a management remark, not a category of the 10-K or 10-Q. Product performance values are Broadcom's stated values, and no claim of superiority over other companies' products is made. Passages marked "Inference" are this article's interpretation of primary sources, not statements made by the company.