COMPANY PROFILE / AI & MACHINE LEARNING
Qualcomm
A company known for on-device AI in smartphones has stepped into AI inference for data centers
Qualcomm's main AI territory is on-device AI: the Hexagon NPU inside Snapdragon chips that runs AI in phones, cars and IoT devices. In October 2025 the same company announced rack-scale AI inference products for data centers (AI200 and AI250), and in June 2026 it added a CPU and a next-generation accelerator under the Dragonfly name. In September 2026, alongside a custom-silicon collaboration with Amazon, it issued an Amazon affiliate a warrant (a right to buy shares) for up to 25 million shares. However, Data Center revenue is not shown separately in the 10-K or 10-Q, and company-wide revenue is still mainly smartphone chips and licensing. This profile sets out what is fact and what is still plan in the AI and machine learning part of the business. Everything is checked as of 15 October 2026.
- Qualcomm 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. Qualcomm in 30 seconds, and where it stands in AI and machine learning
Dragonfly AI200 / AI250Dragonfly is for data centers. The company says it aims to launch AI200 in 2026
Most of Qualcomm's revenue is smartphone chips (handsets were US$27.8 billion of FY2025 QCT revenue), and Data Center revenue is not shown separately Sourced. What the 10-Q does let us read is that Data Center segment equipment and services revenue rose against a year earlier (up US$88 million in the third quarter and US$182 million in the first nine months), and it says the increase was mainly due to the acquired Alphawave. So revenue from its own racks such as AI200 cannot yet be confirmed as a financial figure. Meanwhile, the warrant issued on 3 September 2026 to an Amazon affiliate vests as purchases proceed, up to a maximum of US$60 billion in payments, according to the 8-K. This is a ceiling used for vesting, not a confirmed order Sourced.
2. The logo and how the name is written
A note on names: the legal entity is QUALCOMM Incorporated, written Qualcomm in the logo and in running text. According to the 10-K, the chip business (QCT) and other product businesses, and R&D, are run by its subsidiary Qualcomm Technologies, Inc. (QTI), while the patent licensing business (QTL) is run by the parent. In this article "Qualcomm" means the whole group unless stated otherwise. Snapdragon (devices), Dragonwing (industrial and edge) and Dragonfly (data center) are product brands.
Official site: https://www.qualcomm.com/3. Where the company sits in the AI and machine learning field
4. Scale (limited to the AI-related business)
Qualcomm's revenue centres on smartphone chips (QCT) and patent licensing (QTL), and no revenue figure limited to AI is disclosed. Its fiscal year ends on the last Sunday in September: FY2025 ran from 1 October 2024 to 28 September 2025, and FY2026 ends on 27 September 2026. Figures are as filed in the 10-K and 10-Q Sourced.
| Metric (US$ million) | FY2024 (to 29 Sep 2024) | FY2025 (to 28 Sep 2025) |
|---|---|---|
| Revenue (company-wide) | 38,962 | 44,284 |
| QCT (chip business) | 33,196 | 38,367 |
| Handsets | 24,863 | 27,793 |
| Automotive | 2,910 | 3,957 |
| IoT | 5,423 | 6,617 |
| QTL (licensing) | 5,572 | 5,582 |
| Research and development (company-wide) | 8,893 | 9,042 |
| FY2026 so far (US$ million) | Third quarter (to 28 Jun 2026) | Same quarter a year earlier (to 29 Jun 2025) | First nine months |
|---|---|---|---|
| Revenue (company-wide) | 9,947 | 10,365 | 32,798 |
| QCT (chip business) | 8,504 | 8,993 | 28,193 |
| Handsets | 5,086 | 6,328 | 18,934 |
| Automotive | 1,588 | 984 | 4,015 |
| IoT | 1,830 | 1,681 | 5,244 |
| Research and development (company-wide) | 2,607 | 2,226 | 7,523 |
The Data Center business is a non-reportable segment and its total revenue is not shown. The 10-Q says Data Center equipment and services revenue rose against a year earlier by US$88 million in the third quarter and US$182 million in the first nine months, mainly because of the Alphawave acquisition. The company's stated targets are as follows (company targets and outlooks).
| Item | What the company says | Status |
|---|---|---|
| Revenue outlook for Q4 FY2026 | US$9.7 to 10.5 billion (QCT US$8.4 to 9.0 billion, QTL US$1.2 to 1.4 billion). The fiscal year ends on 27 September 2026 | Not yet confirmed |
| Growth in non-handset revenue | It expects year-on-year growth in non-handset revenue, including Data Center, to accelerate from 24% in FY2026 to more than 60% in FY2027 | Not yet confirmed |
| Non-handset revenue target | US$40 billion in FY2029 (described as nearly double the target shared in November 2024) | Not yet confirmed |
"Non-handset" includes automotive, IoT and Data Center, among others, and no Data Center-only target amount could be confirmed in the results release. The "more than 60%" growth is the company's expectation, not a result Not yet confirmed. Third-quarter revenue also fell 4% on the year, and the 10-Q says memory supply constraints and price rises will adversely affect demand from some handset makers. That demand for memory for AI computing can affect the company's core device business is one feature of this field Sourced.
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, 10-Q, 8-Ks and company releases.
| Type of resource | Details | Status |
|---|---|---|
| People | About 52,000 workers (28 September 2025, including temporary), in over 200 locations in 38 countries; the great majority in engineering or technical roles. No headcount for AI or Data Center is disclosed | Sourced |
| Money: R&D | US$9,042 million in FY2025 (20% of revenue) against US$8,893 million a year earlier; US$7,523 million in the first nine months of FY2026 (23% of revenue). Data Center R&D is not shown separately | Sourced |
| Money: capital expenditure | About US$1.2 billion in FY2025. US$1,578 million in the first nine months of FY2026 (US$785 million a year earlier). The 10-K expects it to rise to support testing of its integrated circuits | Sourced |
| Acquisition: Alphawave | Completed on 18 December 2025 at a purchase price of about US$2.3 billion (about US$1.8 billion in shares and US$301 million in cash). A maker of high-speed wired connectivity technology, IP and custom silicon. Goodwill is allocated to the Data Center segment | Sourced |
| Acquisition: Modular | Completed on 28 July 2026. An open software platform company for AI. About 18 million shares issued, valued at about US$3.1 billion at the closing price (company statement). Some shares carry a four-year service condition | Sourced |
| Acquisitions: others | In the first nine months of FY2026 it acquired seven other businesses for US$1.1 billion in total (US$76 million of goodwill allocated to the Data Center segment) | Sourced |
| Warrant for a customer | On 3 September 2026 it issued an Amazon affiliate a warrant for up to 25 million shares at US$161.26 per share. Rights vest with purchases; 3.75 million shares vested at issuance (based on initial purchase commitments) | Sourced |
| Universities and public bodies | No AI-related joint research with universities or receipt of public funding (grants) was found in the 10-K, 10-Q or releases. Development services for U.S. government agencies (QGOV) are described as a separate business | 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 |
|---|---|---|---|---|
| Dragonfly AI200 (rack) | 56 cards per rack. 768 GB of LPDDR5X per card, 43 TB per rack. Direct liquid or air cooling; 140 kW rack (product page) | Aimed for launch (the company said it expects availability in 2026). The product page shows it as active | Announced 28 October 2025 | Reference 8, 11 https://www.qualcomm.com/news/releases/2025/10/qualcomm-unveils-ai200-and-ai250-redefining-rack-scale-data-cent |
| Dragonfly AI250 (rack) | Second generation, with High Bandwidth Compute (HBC) generation 1. The company states effective bandwidth of 133 TB/s per card (18x AI200) | Expected availability in 2027. HBC commercial sampling expected in mid-2027 | Announced 28 October 2025; updated 25 June 2026 | Reference 8, 7, 12 https://www.qualcomm.com/news/releases/2025/10/qualcomm-unveils-ai200-and-ai250-redefining-rack-scale-data-cent |
| Dragonfly AI300 | HBC generation 2. Aims for 54x AI200's effective bandwidth (company design target) | Planned (commercial sampling in 2028) | 25 June 2026 | Reference 7 https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent |
| Dragonfly C1000 (CPU) | Oryon cores and a chiplet design with more than 250 cores. PCIe Gen 7 and more | Planned (commercial availability expected in 2028) | 25 June 2026 | Reference 7 https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent |
| Collaboration with HUMAIN | Target of 200 megawatts of AI200 and AI250 racks in Saudi Arabia from 2026 | Target. In August 2026 Adobe's move onto HUMAIN's platform was announced (scale not confirmed) | 28 October 2025 / 31 August 2026 | Reference 9, 10 https://www.qualcomm.com/news/releases/2025/10/humain-and-qualcomm-to-deploy-ai-infrastructure-in-saudi-arabia- |
| Collaboration with Meta | A multi-year, multi-generation collaboration in which Qualcomm plans to supply data center CPUs (Dragonfly C1000) for Meta's next-generation server fleet | Planned (CPU commercial availability expected in 2028) | 25 June 2026 | Reference 7 https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent |
| Collaboration with Amazon | A multi-generation collaboration on custom silicon for AI inference, and optical connectivity up to 1.6T. A warrant that vests with purchases (ceiling of US$60 billion in payments) | Collaboration announced. Purchase volumes not fixed | 3 and 8 September 2026 | Reference 6, 4 https://www.qualcomm.com/news/releases/2026/09/qualcomm-announces-multi-generational-product-collaboration-with |
"Planned", "target" and "expected" do not mean the outcome is certain. The US$60 billion warrant ceiling is a cap on purchases counted for vesting, not a confirmed purchase or revenue amount. A megawatt is the electrical scale of a deployment, not a count of cards or an amount of revenue.
6. What it has done so far
| When | What happened | Why it matters |
|---|---|---|
| Long-standing | Snapdragon carries the AI Engine (CPU, GPU and NPU). The Hexagon NPU runs on-device AI inference at very low power (10-K) | The base of its on-device AI. The 10-K says it intends to use its Oryon CPU cores in data center products too |
| 9 June 2025 | Announced the Alphawave acquisition (10-K) | A move to bring in high-speed connectivity technology |
| 28 October 2025 | Announced AI200 and AI250. Announced the collaboration with HUMAIN targeting 200 megawatts | A declaration of serious entry into data center AI inference |
| 18 December 2025 | Completed the Alphawave acquisition (about US$2.3 billion) | Added connectivity, IP and custom silicon to the Data Center segment |
| 21 to 25 June 2026 | Agreed to acquire Modular (21st). At Investor Day announced Dragonfly C1000, HBC and AI300, and the collaboration with Meta (25th) | Added a CPU and a next-generation accelerator, and set out an annual-cadence roadmap |
| 28 July 2026 | Completed the Modular acquisition | Brought in an AI software foundation |
| 31 August 2026 | Announced that Adobe will run AI workloads on HUMAIN's platform using Dragonfly accelerators | A first deployment example (scale not confirmed) |
| 3 and 8 September 2026 | Issued the warrant to an Amazon affiliate (3rd). Announced the collaboration with Amazon (8th) | A large collaboration carrying equity rights tied to purchases |
| 14 October 2026 | A company statement on the litigation with Arm (a jury returned a verdict in its favour) | A dispute over the licence underlying the Oryon CPUs. Finality not confirmed |
7. What comes next
The company presents these as expectations and targets, not as achieved facts Not yet confirmed. The roadmap is described as an annual cadence of new products, but the actual timing can change.
None of the following could be confirmed from the company's 10-K, 10-Q, 8-Ks or releases Not yet confirmed. Total Data Center revenue and operating results / AI200 unit shipments and the scale per customer / the confirmed purchase amount from Amazon (the US$60 billion warrant ceiling is not a confirmed amount) / progress on HUMAIN's 200-megawatt deployment / Dragonfly prices. Customer-by-customer revenue or shipment figures sometimes appear in press coverage, but this article does not use them.
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 |
|---|---|---|
| It positions Data Center as a growth pillar to supplement smartphone-centred earnings | The 10-K says growth depends in part on success beyond handsets, including automotive, IoT and data center. The company's target also puts non-handset revenue at US$40 billion in FY2029 | Inference |
| In data centers it enters through inference-specific racks, not as a GPU substitute | AI200, AI250 and AI300 are all described as inference products, with tokens per watt as the stated yardstick | Inference |
| It aims to make memory bandwidth and power its point of differentiation | AI200 gets capacity from LPDDR, and AI250 onward aims to raise effective bandwidth with HBC (company design targets) | Inference |
| It fills gaps in connectivity, software and custom silicon through acquisitions | Alphawave (connectivity, IP, custom silicon) and Modular (AI software platform) were completed in fiscal 2026 | Inference |
| It seeks to secure large customers' purchases by pairing them with warrants | The warrant to an Amazon affiliate vests as purchases proceed (8-K) | Inference |
The 8-K filed on 8 September 2026 says the warrant issued to the Amazon affiliate vests "in tranches tied to the execution of certain commercial arrangements, the placement of binding purchase orders and actual purchases" (up to US$60 billion in payments, with 3.75 million shares vested at issuance) Sourced. The more purchases accumulate, the more shares the buyer can acquire at a fixed price. This article reads that as Qualcomm using the value of its shares to secure large data center purchases. But the 8-K and the announcement do not reveal what is being bought (products, quantities, timing), and the company does not describe it that way 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 |
|---|---|---|
| On-device AI | Qualcomm AI Engine (Hexagon NPU, Adreno GPU, Oryon / Kryo CPUs), AI Stack, AI Hub | The 10-K describes the Hexagon NPU as a key processor designed for sustained, high-performance AI inference at very low power. It provides a basis for developers to optimise and deploy models |
| Data center accelerators | Dragonfly AI200, AI250, AI300 | Rack-scale inference products. AI200 uses LPDDR; AI250 onward uses HBC (company announcements) |
| Data center CPU | Dragonfly C1000 | Oryon cores. Agentic, general-purpose and AI head-node variants (company announcement). Planned |
| Connectivity | 224G / 448G SerDes, 800G and 1.6T optical, AOC and AEC, DSPs | Connectivity products that bring in Alphawave technology (company announcement) |
| Software | Qualcomm AI Inference Suite, Dragonfly AI Infrastructure Management Suite, Modular's platform | Model deployment and rack-level operations management (product pages and announcements) |
The table below sets out the developer's stated values for the data center racks from Qualcomm's product pages (checked 15 October 2026). They are company statements, not measured performance.
| Item (Qualcomm stated values) | Dragonfly AI200 | Dragonfly AI250 |
|---|---|---|
| Generation and memory approach | LPDDR5X | HBC generation 1 (near-memory compute) |
| Cards per rack | 56 | 56 |
| Memory capacity per card | 768 GB | 768 GB |
| Memory capacity per rack | 43 TB | 43 TB |
| Rack bandwidth | 0.414 PB/s (LPDDR5X) | 7.455 PB/s (effective bandwidth with HBC) |
| Scale-up / scale-out | PCIe 6.0 / Ethernet (RoCE) | PCIe 6.0 / Ethernet (RoCE) |
| Cooling and power | Direct liquid or air cooling; 140 kW | Direct liquid or air cooling; 140 kW |
Source: Qualcomm product pages. The "effective bandwidth" for AI250 is the company's term for what the HBC technology delivers and is defined differently from HBM bandwidth. The product pages claim "industry-leading" figures against competitors' published specifications, but the basis of comparison is in their footnotes, so this article makes no comparison with other companies' products and no claim of superiority. Also, the 28 October 2025 announcement gave rack power as 160 kW, whereas the current product pages say 140 kW, so the statement has changed.
Choice of memory: AI200 carries 768 GB of LPDDR5X (low-power memory with smartphone origins) per card, a design the company says gives large capacity Sourced. Compared with HBM (see our explainer on HBM) the bandwidth is lower, but the thinking is that power and cost per unit of capacity are kept down. The 10-Q says memory supply constraints and price rises are affecting demand in the device business, so the choice of memory bears on the whole company.
Near-memory compute (HBC): the company describes the HBC used in AI250 as bringing compute and memory closer with 3D stacking to reduce data movement Sourced. Figures such as "6x better bandwidth per watt than HBM" are design targets the company compares with competitors' published specifications, not measurements. Commercial sampling is planned for mid-2027, so it cannot yet be evaluated (see also our explainers on chiplets and heterogeneous integration).
Cooling and power: the rack is in the 140 kW class and assumes liquid cooling (product pages). Putting tokens per watt forward as the yardstick fits the trend of power becoming the unit of deployment Sourced (see our explainer on AI data center power and cooling).
What it shares with on-device AI: Qualcomm says it will bring NPU and CPU designs built for "high performance and low power" in devices into the data center (10-K). How far a design made for devices holds up at data center scale cannot be judged here (see our explainer on edge AI and on-device AI).
11. The risks this company carries
| Risk | Details | Status |
|---|---|---|
| Execution in new areas | The 10-K says its investments in new areas such as automotive, IoT and data center may not succeed. Data Center revenue is not shown separately, so progress is hard to measure from outside | Sourced |
| The core smartphone business | Third-quarter FY2026 revenue fell 4% on the year. Affected by memory supply constraints and price rises, Apple using its own modems (10-Q), and large customers (Apple, Samsung, Xiaomi) using their own chips | Sourced |
| Customer concentration | In FY2025 Apple, Samsung and Xiaomi each accounted for 10% or more of revenue. In data centers too, dependence on a few large customers may arise | Sourced |
| Licence dispute with Arm | Litigation over the design of the Oryon CPUs. On 30 September 2025 the district court entered final judgment in Qualcomm's favour and Arm appealed (10-Q). A separate case (Qualcomm's suit) went to trial from 5 October 2026, and on 14 October the company stated that the jury ruled in its favour. Further proceedings such as appeals may follow. This article does not predict the outcome | Sourced |
| Integration of acquisitions and share issuance | Integrating Alphawave and Modular. Both were paid mainly in shares, and together with the warrant (up to 25 million shares) could dilute shareholders (10-Q, 8-Ks) | Sourced |
| Supply and costs | Rising costs and capacity constraints in wafers, assembly, test, advanced packaging and memory. The 10-Q says limited supply could prevent it from meeting customer demand | Sourced |
| China and trade or security tensions | The 10-K and 10-Q say a significant part of its business is concentrated in China and that U.S.-China trade and national security tensions are a risk (no description of export-control effects on data center products was found in what this article read) | Sourced |
| Roadmap delays | AI250, AI300 and C1000 are all expected in 2027 to 2028 and may not proceed as planned | Sourced |
12. Glossary
- On-device AI
- Running AI inside a device such as a smartphone or a car, rather than in the cloud.
- NPU (Hexagon)
- A processor specialised for AI calculations. Qualcomm's is called Hexagon.
- QCT / QTL
- Qualcomm's QCT is the chip business and QTL is the patent licensing business.
- Dragonfly
- Qualcomm's brand for its data center products (AI200, AI250, AI300 and C1000).
- LPDDR
- Low-power DRAM widely used in smartphones; AI200 uses it for large capacity.
- HBC
- High Bandwidth Compute: Qualcomm's near-memory compute technology that brings compute and memory close with 3D stacking.
- Warrant
- A right to buy shares at a price set in advance. Issued to an Amazon affiliate for up to 25 million shares.
- Megawatt (MW)
- A unit for the electrical scale of a deployment. Not a count of cards or an amount of revenue.
- Oryon
- A CPU core designed by Qualcomm, based on a licence for Arm's instruction set.
- 10-K / 10-Q / 8-K
- The annual, quarterly and current reports filed with the U.S. SEC.
13. References
- Qualcomm Form 10-K for the fiscal year ended September 28, 2025 (filed with the U.S. SEC on November 5, 2025). Business, segment revenue, R&D, headcount, capital expenditure, the litigation with Arm, how the Data Center business is positioned, and risk factors. https://www.sec.gov/Archives/edgar/data/804328/000080432825000085/qcom-20250928.htm
- Qualcomm Form 10-Q for the quarterly period ended June 28, 2026 (filed with the U.S. SEC on July 29, 2026). Revenue for the third quarter and first nine months of fiscal 2026, the Alphawave and Modular acquisitions, the Data Center business's contribution to revenue, capital expenditure and the litigation with Arm. https://www.sec.gov/Archives/edgar/data/804328/000080432826000086/qcom-20260628.htm
- Qualcomm Qualcomm Announces Third Quarter Fiscal 2026 Results (press release furnished with Form 8-K, July 29, 2026). Fourth-quarter outlook and non-handset revenue targets. https://www.sec.gov/Archives/edgar/data/804328/000080432826000085/qcom062826erex991.htm
- Qualcomm Form 8-K (filed September 8, 2026). The warrant issued to an Amazon affiliate (September 3, 2026). https://www.sec.gov/Archives/edgar/data/804328/000110465926105718/tm2623289d1_8k.htm
- Qualcomm Form 8-K (filed June 24, 2026). The definitive agreement to acquire Modular Inc (June 21, 2026). https://www.sec.gov/Archives/edgar/data/804328/000110465926077071/tm2618522d1_8k.htm
- Qualcomm Qualcomm Announces Multi-Generational Product Collaboration with Amazon to Build Next-Generation AI Data Center Infrastructure (press release, September 8, 2026; read in a browser). https://www.qualcomm.com/news/releases/2026/09/qualcomm-announces-multi-generational-product-collaboration-with
- Qualcomm Qualcomm Unveils Comprehensive Data Center Roadmap for the Agentic AI Era with New Qualcomm Dragonfly Portfolio (press release, June 25, 2026, including the Meta collaboration; read in a browser). https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent
- Qualcomm Qualcomm Unveils AI200 and AI250 - Redefining Rack-Scale Data Center Inference Performance for the AI Era (press release, October 28, 2025; read in a browser). https://www.qualcomm.com/news/releases/2025/10/qualcomm-unveils-ai200-and-ai250-redefining-rack-scale-data-cent
- Qualcomm / HUMAIN HUMAIN and QUALCOMM to deploy AI Infrastructure in Saudi Arabia For Global Inferencing (joint press release, October 28, 2025; read in a browser). https://www.qualcomm.com/news/releases/2025/10/humain-and-qualcomm-to-deploy-ai-infrastructure-in-saudi-arabia-
- Qualcomm / HUMAIN Adobe Becomes First Global Software Company to Migrate AI Workloads onto HUMAIN Platform, Accelerated by Qualcomm (press note, August 31, 2026; read in a browser). https://www.qualcomm.com/news/releases/2026/08/adobe-becomes-first-global-software-company-to-migrate-ai-worklo
- Qualcomm Qualcomm Dragonfly AI200 (product page with specification table; checked October 15, 2026; read in a browser). https://www.qualcomm.com/data-center/products/qualcomm-dragonfly-ai200
- Qualcomm Qualcomm Dragonfly AI250 (product page with specification table; checked October 15, 2026; read in a browser). https://www.qualcomm.com/data-center/products/qualcomm-dragonfly-ai250
- Qualcomm Qualcomm Statement on Trial Verdict Win (press release, October 14, 2026; the company's statement on the litigation with Arm; read in a browser). https://www.qualcomm.com/news/releases/2026/10/qualcomm-statement-on-trial-verdict-win
- Qualcomm Official website (source of the logo image, as shown in the site's header; checked October 15, 2026). https://www.qualcomm.com/
14. Claim-to-source audit
| Claim in the text | Status | Reference |
|---|---|---|
| Revenue, segment revenue, R&D, headcount, capital expenditure and the 10% customers (Apple, Samsung, Xiaomi) for FY2024 to FY2025 | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/804328/000080432825000085/qcom-20250928.htm |
| The Data Center business being a non-reportable segment, and the description of on-device AI (Hexagon NPU) | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/804328/000080432825000085/qcom-20250928.htm |
| Revenue, segment revenue, R&D and capital expenditure for the third quarter and first nine months of FY2026 | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/804328/000080432826000086/qcom-20260628.htm |
| The rise in Data Center equipment and services revenue (US$88 million and US$182 million) and the content of the Alphawave and Modular acquisitions | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/804328/000080432826000086/qcom-20260628.htm |
| The course of the litigation with Arm (final judgment of 30 September 2025; trial scheduled for 5 October 2026) | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/804328/000080432826000086/qcom-20260628.htm |
| The Q4 revenue outlook, non-handset growth and the revenue target (US$40 billion in FY2029) | Sourced | Reference 3 https://www.sec.gov/Archives/edgar/data/804328/000080432826000085/qcom062826erex991.htm |
| The warrant to an Amazon affiliate (up to 25 million shares, US$161.26, US$60 billion payment ceiling, 3.75 million shares vested) | Sourced | Reference 4 https://www.sec.gov/Archives/edgar/data/804328/000110465926105718/tm2623289d1_8k.htm |
| The definitive agreement to acquire Modular (21 June 2026) | Sourced | Reference 5 https://www.sec.gov/Archives/edgar/data/804328/000110465926077071/tm2618522d1_8k.htm |
| The collaboration with Amazon (custom silicon for AI inference, 1.6T optical connectivity, EDA on AWS) | Sourced | Reference 6 https://www.qualcomm.com/news/releases/2026/09/qualcomm-announces-multi-generational-product-collaboration-with |
| The announcement of Dragonfly C1000, HBC and AI300, the Meta collaboration, and the planned timing | Sourced | Reference 7 https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent |
| The announcement of AI200 and AI250 (28 October 2025) and the expected launch timing | Sourced | Reference 8 https://www.qualcomm.com/news/releases/2025/10/qualcomm-unveils-ai200-and-ai250-redefining-rack-scale-data-cent |
| The collaboration with HUMAIN (200 megawatts, from 2026) | Sourced | Reference 9 https://www.qualcomm.com/news/releases/2025/10/humain-and-qualcomm-to-deploy-ai-infrastructure-in-saudi-arabia- |
| Adobe's move onto HUMAIN's platform (31 August 2026) | Sourced | Reference 10 https://www.qualcomm.com/news/releases/2026/08/adobe-becomes-first-global-software-company-to-migrate-ai-worklo |
| AI200 specifications (56 cards, 768 GB, 43 TB, 140 kW and others) | Sourced | Reference 11 https://www.qualcomm.com/data-center/products/qualcomm-dragonfly-ai200 |
| AI250 specifications (effective bandwidth, HBC generation 1) | Sourced | Reference 12 https://www.qualcomm.com/data-center/products/qualcomm-dragonfly-ai250 |
| The company statement on the litigation with Arm (14 October 2026) | Sourced | Reference 13 https://www.qualcomm.com/news/releases/2026/10/qualcomm-statement-on-trial-verdict-win |
| Source of the logo image | Sourced | Reference 14 https://www.qualcomm.com/ |
| The four company types and Qualcomm's placement | Our calculation | This article's own classification. Company names are examples from firms covered in this series |
| The warrant's exercise value of about US$4 billion as a rough guide | Our calculation | Our calculation (25 million shares x US$161.26; depends on vesting conditions) |
| AI200 launch, the timing of AI250, AI300 and C1000, non-handset growth, Amazon's purchases and the HUMAIN deployment | Not yet confirmed | Company outlooks, plans and targets, not settled facts. Items that could not be confirmed are stated in the text. Reference 7 https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent |
| The reading that it positions Data Center as a growth pillar | Inference | This article's interpretation of the 10-K and the results release |
| The reading that it enters through inference-specific racks | Inference | This article's interpretation of the official announcements |
| The reading that memory bandwidth and power are its point of differentiation | Inference | This article's interpretation of the product pages and announcements |
| The reading that it secures purchases by pairing them with warrants | Inference | This article's interpretation of the 8-K |
| The materials engineer's view on memory approach, near-memory compute, cooling and what it shares with on-device AI | Commentary | General technical commentary, not a Qualcomm disclosure. Figures are the company's stated values and design targets |
Last updated 15 October 2026 / Troy Technical
Figures in this article come from Qualcomm's statutory filings with the U.S. Securities and Exchange Commission (the 10-K, 10-Q, 8-Ks and results release) and from its official press releases and product pages. No research-firm estimates or press-based figures have been used. Data Center revenue is not disclosed, and no estimate is made. Product performance values are the company's stated values and design targets, 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.