COMPANY PROFILE / AI & MACHINE LEARNING
AMD
A chip company with both server CPUs (EPYC) and AI GPUs (Instinct), pairing large customer deals with equity rights
AMD supplies both the EPYC server CPU and the Instinct AI accelerator to the data center. Revenue for 2025 (the fiscal year ended 27 December 2025) was US$34.6 billion, of which the Data Center segment was US$16.6 billion. In the April to June 2026 quarter, Data Center revenue reached US$6.7 billion, up 107% on the year. OpenAI and Meta have each said they intend to deploy up to 6 gigawatts of AMD GPUs, and Anthropic up to 2 gigawatts; filings and releases show that AMD has paired these customers with warrants (rights to buy AMD shares) or an equity investment. This profile sets out what AMD sells, where it is putting its money, what stage each large deal has reached, and how export controls have affected it, within its filings and official releases. Everything is checked as of 15 October 2026.
- AMD 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. AMD in 30 seconds, and where it stands in AI and machine learning
EPYC, HeliosHelios is a design for a rack carrying 72 GPUs
AMD's large AI deals come with warrants or equity investment attached for the customer Sourced. To OpenAI (October 2025) and Meta (February 2026) it issued warrants to buy up to 160 million shares each at US$0.01 per share. Rights vest only as conditions are met, including purchase milestones for Instinct GPUs and stock-price targets. As of 27 June 2026, none had vested and there was no effect on the financial statements, according to the 10-Q. In July 2026, alongside a plan for Anthropic to deploy up to 2 gigawatts, AMD announced that it had committed to invest up to US$5 billion in Anthropic. AMD is therefore becoming both a seller and a party with a stake in its buyers' equity and rights. Separately, the 10-K states that no customer accounted for 10% or more of revenue in 2025 or 2024 Sourced.
2. The logo and how the name is written
A note on names: the legal entity is Advanced Micro Devices, Inc., usually called AMD. The logo is the white version distributed on the official site, so it is placed on a dark background. Instinct (AI accelerators), EPYC (server CPUs), Pensando (networking) and Helios (a rack design) are product names, not company names.
Official site: https://www.amd.com/en.html3. Where the company sits in the AI and machine learning field
4. Scale (limited to the AI-related business)
AMD's fiscal year ends on the Saturday nearest the end of December, and 2025 ran from 29 December 2024 to 27 December 2025. The Data Center segment includes AI accelerators (Instinct) and server CPUs (EPYC), among others, and the split (Instinct revenue alone) is not disclosed (as far as this article could confirm). Figures are as filed in the 10-K and 10-Q Sourced.
| Metric (US$ million) | 2024 (to 28 Dec 2024) | 2025 (to 27 Dec 2025) |
|---|---|---|
| Revenue | 25,785 | 34,639 |
| Data Center segment revenue | 12,579 | 16,635 |
| Client and Gaming | 9,649 | 14,550 |
| Embedded | 3,557 | 3,454 |
| Data Center segment operating income | 3,482 | 3,603 |
| Research and development | 6,456 | 8,091 |
| Operating income (company-wide) | 1,900 | 3,694 |
| 2026 so far (US$ million) | Second quarter (to 27 Jun 2026) | Same quarter a year earlier (to 28 Jun 2025) | First half (six months) |
|---|---|---|---|
| Revenue | 11,536 | 7,685 | 21,789 |
| Data Center segment revenue | 6,718 | 3,240 | 12,493 |
| Data Center segment operating income (loss) | 2,103 | (155) | 3,702 |
| Research and development | 2,528 | 1,894 | 4,925 |
The year-earlier quarter (the second quarter of 2025) includes about US$800 million of inventory and related charges on Instinct MI308 caused by U.S. export controls. The Data Center operating loss of US$155 million includes that effect, so keep it in mind when comparing with the same period a year earlier Sourced. The 10-Q attributes the 107% rise in Data Center revenue mainly to demand for EPYC and Instinct MI350, but it does not give the revenue split between the two. The company's outlook for third-quarter 2026 revenue is about US$13 billion, plus or minus US$300 million. It is a company outlook and unconfirmed 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 31,000 employees (27 December 2025). No breakdown of AI-specific headcount is disclosed | Sourced |
| Money: R&D | US$8,091 million in 2025 (about 23% of revenue: our calculation) against US$6,456 million a year earlier; US$4,925 million in the first half of 2026 | Sourced |
| Money: capital expenditure | About US$1.0 billion of purchases of property and equipment in 2025 (cash flow statement) | Sourced |
| Money: purchase, lease and other commitments | Unconditional commitments rose from about US$12.2 billion (December 2025) to about US$30.3 billion (June 2026). Maximum exposure under data center lease guarantees is US$4.1 billion. After quarter-end it entered leases not yet started (US$9.5 billion in total, expected to begin in 2027 to 2028) and investment commitments of up to US$5.0 billion | Sourced |
| Acquisitions: ZT Systems | Acquired in March 2025 (US$3.2 billion in cash and 8.3 million shares). It kept the design business and sold the manufacturing business to Sanmina in October 2025 (US$2.4 billion in cash and 1,151,052 Sanmina shares, plus up to US$450 million of contingent consideration through 2028) | Sourced |
| Acquisitions: others | Acquired Silo AI, a Finnish AI lab, in August 2024 (US$665 million in cash). In 2025 it also bought software, photonics and inference technology businesses (10-K). In the second quarter of 2026 it acquired MEXT, a memory-optimisation technology (results release) | Sourced |
| Warrants and investment for customers | Warrants for up to 160 million shares each issued to OpenAI and Meta. In July 2026 it announced a commitment to invest up to US$5 billion in Anthropic (to be made in future) | Sourced |
| The World Labs merger agreement | Merger agreement signed on 26 September 2026. The purchase price is about US$8.2 billion, payable in AMD shares. The text of the 8-K (28 September) does not give the closing timetable or conditions | Not confirmed (closing) |
| Universities and public bodies | On 27 October 2025 the U.S. Department of Energy (DOE) announced two AMD-accelerated supercomputers at Oak Ridge National Laboratory: Lux (to be deployed in early 2026) and Discovery (to arrive in 2028). The DOE cites more than US$1 billion of public-private investment | Sourced |
| Public funding received | No record of AMD receiving public funding (such as grants) was found in the 10-K or 10-Q | 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 |
|---|---|---|---|---|
| Instinct MI355X (MI350 series) | 288 GB HBM3E, TSMC 3nm. The 10-K says large cloud and other customers deployed MI350X in 2025 | Shipping | Launch date 12 June 2025 (product page) | Reference 8, 1 https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html |
| Instinct MI455X (MI400 series) | 432 GB HBM4. The core GPU of Helios | Launched (launch date on the product page) | 23 July 2026 (product page) | Reference 7, 3 https://www.amd.com/en/products/accelerators/instinct/mi400/mi455x.html |
| Helios (rack design) | 72 MI455X GPUs, EPYC "Venice", Pensando "Vulcano". AMD provides the design and does not make or sell finished racks | Shipments expected to start (second half of 2026) | 10-Q (August 2026) and product page | Reference 2, 9 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| OpenAI agreement | Up to 6 gigawatts; the first gigawatt on MI450 series. Warrant for up to 160 million shares | Contract (deployment is the customer's intention); warrant not vested | October 2025 | Reference 1, 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm |
| Meta agreement | Up to 6 gigawatts; the first gigawatt on a custom MI450-based GPU with Venice CPUs. Shipments to start in the second half of 2026 | Contract (shipments scheduled) | 24 February 2026 | Reference 4, 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000045/pressreleasedatedfebruary2.htm |
| Anthropic partnership | Up to 2 gigawatts of MI450 series in Helios; the first gigawatt from the first half of 2027. AMD committed to invest up to US$5 billion | Agreement (deployment scheduled; investment in future) | 22 July 2026 | Reference 5 https://ir.amd.com/news-events/press-releases/detail/1292/amd-and-anthropic-announce-strategic-partnership-to-deploy-up-to-2-gigawatts-of-amd-instinct-mi450-series-gpus |
| DOE supercomputers Lux and Discovery | Lux to be deployed in early 2026; Discovery to arrive in 2028 | Planned | 27 October 2025 | Reference 10 https://www.energy.gov/articles/energy-department-announces-new-public-private-partnership-model-two-supercomputers |
"Deployment is the customer's intention" reflects the 10-Q wording that each customer "intends to deploy" up to a number of gigawatts. These are not contracts for confirmed purchase quantities. A gigawatt is the electrical scale of a data center, not a count of GPUs or an amount of revenue.
6. What it has done so far
| When | What happened | Why it matters |
|---|---|---|
| August 2024 | Acquired Silo AI (US$665 million in cash) | Brought in AI model and software talent (10-K) |
| 31 March 2025 | Acquired ZT Systems | Gave it rack-scale system design capability (10-K) |
| April 2025 | The U.S. government imposed a new licence requirement on exports of MI308 to China and other destinations. It booked about US$800 million of inventory and related charges in the second quarter | The first large case of export controls hitting AMD (10-K) |
| 12 June 2025 | Launch date of Instinct MI355X (MI350 series), per the product page | The 10-K says large cloud and other customers deployed MI350X in 2025 |
| October 2025 | Announced the OpenAI agreement for up to 6 gigawatts, with a warrant. Sold the ZT manufacturing business to Sanmina (27th) | A large contract, and a decision to stop making racks itself |
| Fourth quarter of 2025 | Began shipping MI308 under licence and reversed about US$360 million of the charges | The charge ended at about US$440 million (10-K) |
| February 2026 | Meta agreement for up to 6 gigawatts (24th). Obtained licences to ship MI325 to certain China-based customers | The 10-Q says it is not yet known whether imports into China will be allowed |
| July 2026 | Anthropic partnership for up to 2 gigawatts (22nd). Launch date of Instinct MI455X (23rd) | Preparation for Helios shipments |
| 4 August 2026 | Second-quarter results. Said Helios was beginning to ramp. Third-quarter revenue outlook of about US$13 billion | The CEO said Helios "begins to ramp" (results release) |
| 26 September 2026 | Signed a merger agreement with World Labs (about US$8.2 billion, payable in AMD shares) | A move to bring in AI model research. Closing not confirmed |
7. What comes next
The company presents these as outlooks and plans, not as achieved facts Not yet confirmed. The 10-Q says that as product transitions speed up, as with Helios, managing supply and demand becomes harder and could affect revenue and inventory.
None of the following could be confirmed from the company's 10-K, 10-Q, 8-Ks or releases Not yet confirmed. Revenue from Instinct alone / Helios unit shipments and prices / actual shipments to OpenAI and Meta (the warrants had not vested at 27 June 2026) / the timing and conditions for closing the World Labs acquisition / when the investment in Anthropic will be made. Figures for named customers' deployments or revenue 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 |
|---|---|---|
| AMD is widening from chips to providing rack-scale designs, but leaves manufacturing to others | It kept the ZT design business, sold the manufacturing business to Sanmina and named Sanmina its preferred manufacturing partner for complex AI systems. The 10-Q says it does not make or sell finished Helios racks | Inference |
| It seeks to secure large customers' purchases by combining them with warrants and equity investment | The OpenAI and Meta warrants vest only as Instinct purchase milestones and stock-price conditions are met. It has committed to invest up to US$5 billion in Anthropic | Inference |
| It makes open standards (UALink, Ultra Ethernet, an OCP rack) a point of differentiation | The Helios product page says it is designed on open standards such as OCP Open Rack Wide, UALink and UEC. ROCm is described as an open-source software stack | Inference |
| It keeps bringing AI model and software expertise in-house | Silo AI (2024), MEXT (2026) and the World Labs merger agreement (September 2026). The purpose of the merger is not stated in the 8-K text | Inference |
| Its revenue is not concentrated in a few customers | No customer accounted for 10% or more of revenue in 2025 or 2024 (10-K). For receivables, one customer was 11% (December 2025) and another 24% (December 2024) | Sourced |
The OpenAI and Meta warrants work so that the more AMD GPUs the buyer purchases, the more AMD shares it can acquire at a very low price. The 10-Q says vesting depends on GPU purchase milestones and stock-price conditions, and that a vested tranche must also meet further technical and commercial conditions before it can be exercised Sourced. This reads as a design meant to secure purchase volume and deepen the customer's involvement. At the same time, no rights had vested at 27 June 2026, so there is distance between an "intention" to deploy and actual purchases. This article reads the structure as AMD using the value of its shares to secure large-scale demand, but 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 |
|---|---|---|
| AI accelerators | Instinct MI350 series (MI355X), MI400 series (MI455X, and MI430X for HPC and sovereign AI), MI350P | The 10-K states an annual cadence for new Instinct products. MI355X launched in 2025 and MI455X on 23 July 2026 |
| Server CPUs | EPYC (5th generation; 6th generation "Venice" in 2026) | 5th generation introduced in 2025; 6th generation introduced in the results release |
| Networking | Pensando Pollara 400 AI NIC, Vulcano AI NIC, Salina DPU, UALink / UALoE | For connecting GPU clusters. Vulcano is 800 Gbps (product page) |
| Rack design | Helios (72 MI455X GPUs) | Based on the OCP Open Rack Wide standard. AMD provides the design and does not make or sell finished racks (10-Q) |
| Software | ROCm (an open-source AI and HPC software stack) | The product page cites day-0 support for PyTorch, TensorFlow, JAX and others |
| AI PCs and embedded | Ryzen AI (with NPU), Versal AI Engine | A secondary business within this field |
The table below sets generations side by side using the developer's stated values from AMD's product pages (checked 15 October 2026). All are theoretical peak values, not measured performance.
| Item (AMD stated values) | Instinct MI355X (MI350 series) | Instinct MI455X (MI400 series) |
|---|---|---|
| Launch date | 12 June 2025 | 23 July 2026 |
| Process technology | TSMC 3nm / 6nm FinFET | TSMC 2nm / 3nm FinFET |
| Peak MXFP4 (4-bit) performance | 10.1 PFLOPS | 40.3 PFLOPS |
| Peak MXFP8 / OCP FP8 performance | 5 PFLOPS (without sparsity) | 20.1 PFLOPS |
| GPU memory | 288 GB (HBM3E) | 432 GB (HBM4) |
| Peak memory bandwidth | 8 TB/s | 23.3 TB/s |
| Cooling and power | Typical board power 1,400 W (TBP) | Direct liquid cooling (DLC) |
| Item (AMD stated values) | Helios rack (72 MI455X GPUs) |
|---|---|
| Peak FP4 / FP8 performance (rack total) | 2.9 EFLOPS / 1.4 EFLOPS |
| HBM4 capacity (rack total) | 31 TB |
| Scale-up bandwidth (rack total) | 260 TB/s (UALink over Ethernet) |
| Scale-out bandwidth (rack total) | 43 TB/s (Pensando Ethernet) |
| Availability | The design is being shared with partners; volume deployments are expected in the second half of 2026 (product page) |
Source: AMD product pages. The values are theoretical peaks, and sparsity is handled differently by data type (the MI355X and MI455X tables list "with structured sparsity" figures on separate lines for FP16, BF16 and INT8, while the MXFP4 and MXFP8 lines carry no sparsity label). AMD's MI455X page includes a comparison table with NVIDIA's Vera Rubin, but the basis of comparison depends on each company's footnotes, so this article does not reproduce comparisons with other companies' products and makes no claim of superiority. Also, another passage on the same Helios page gives "19.6 TB/s memory bandwidth per GPU", which does not match the 23.3 TB/s in the specification table. This article uses the specification-table value and notes that the two statements differ.
Memory: from MI355X to MI455X, GPU memory moves from 288 GB of HBM3E (8 TB/s) to 432 GB of HBM4 (23.3 TB/s) (AMD stated values) Sourced. The product page says MI455X carries 12 HBM stacks. Memory capacity and bandwidth bear directly on whether a large model fits on one GPU, which is the common understanding (see our explainer on HBM).
Advanced process and chiplets: MI455X is listed as "TSMC 2nm / 3nm FinFET", which suggests several dies made on different processes combined on one substrate. The 10-K mentions a "custom-ready chiplet platform" and an Infinity Fabric switch Sourced (see our explainers on chiplets, CoWoS and heterogeneous integration).
Cooling and power: MI455X uses direct liquid cooling (DLC), and the MI355X page gives a typical board power of 1,400 W. Helios assumes a wider rack (Open Rack Wide) with more room for power and cooling than earlier racks, a power shelf with a vertical busbar, and liquid-coolant distribution Sourced (see our explainer on AI data center power and cooling).
Connection: the product page says Helios links GPUs inside the rack by running UALink over Ethernet (UALoE), and connects racks with Pensando Ethernet. This is the configuration AMD describes; real-world performance and superiority over other approaches could not be checked here.
11. The risks this company carries
| Risk | Details | Status |
|---|---|---|
| Gap between intention and purchases | OpenAI and Meta "intend to deploy up to 6 gigawatts" and Anthropic up to 2 gigawatts; none is a confirmed purchase quantity. The warrants had not vested at 27 June 2026 | Sourced |
| Warrants and investment for customers | Exercising warrants or making the investment could affect dilution of AMD shares and the valuation of the investment. The investment of up to US$5 billion is to be made in future (the 10-Q reports investment commitments of up to US$5.0 billion entered into after quarter-end) | Sourced |
| Export controls | New licence requirement in April 2025. It booked about US$800 million of charges on MI308 and later reversed about US$360 million (about US$440 million net). In February 2026 it obtained MI325 licences for China, but it is not known whether imports will be allowed, and a 25% tariff applies on U.S. import for inspection. The AI Diffusion Rule is to be rescinded and replaced (10-K). The rules can change | Sourced |
| Product transitions and supply | The 10-Q says system-level products such as Helios make supply and demand harder to manage and could affect revenue and inventory. It depends on outsourced manufacturers such as TSMC | Sourced |
| Rising commitments | Unconditional commitments rose to about US$30.3 billion. Maximum lease-guarantee exposure is US$4.1 billion. Leases not yet started total US$4.5 billion and US$9.5 billion (after quarter-end) | Sourced |
| Competition | The 10-K says it competes mainly with Intel and NVIDIA in the data center. The 10-Q mentions NVIDIA's market position and software ecosystem influencing customers, and says the September 2025 NVIDIA-Intel partnership could intensify competition (AMD's own statements) | Sourced |
| Integration of acquisitions | Integrating ZT Systems, the World Labs merger (closing not confirmed) and Silo AI. Acquisitions may not deliver the expected benefits (10-K) | Sourced |
| Customers' own designs | Some customers develop their own data center chips, which could narrow AMD's market (10-K) | Sourced |
12. Glossary
- Instinct
- AMD's brand of AI accelerators (GPUs) for the data center, in generations such as MI350 and MI400.
- EPYC
- AMD's brand of server CPUs. x86-based.
- Helios
- A design that puts 72 MI455X GPUs and more into one rack. AMD provides the design and does not make finished racks.
- ROCm
- AMD's open-source software stack for AI and HPC.
- UALink
- An industry interconnect standard for linking accelerators. UALoE runs UALink over Ethernet.
- MXFP4
- One 4-bit floating-point format. Peak performance is a theoretical figure, not a measurement.
- Warrant
- A right to buy shares at a price set in advance. AMD issued up to 160 million shares' worth each to OpenAI and Meta.
- Gigawatt (GW)
- A unit for the electrical scale of a data center. Not a count of GPUs or an amount of revenue.
- TBP
- Typical board power: the usual power draw of a board.
- 10-K / 10-Q / 8-K
- The annual, quarterly and current reports filed with the U.S. SEC.
13. References
- AMD Form 10-K for the fiscal year ended December 27, 2025 (filed with the U.S. SEC on February 4, 2026). Business, segments, revenue, R&D, headcount, ZT Systems, the OpenAI agreement and warrant, export controls and risk factors. https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm
- AMD Form 10-Q for the quarterly period ended June 27, 2026 (filed with the U.S. SEC on August 5, 2026). Revenue for the second quarter and first half of 2026, the OpenAI and Meta agreements and warrants, commitments, lease guarantees, Helios and export controls (MI325). https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm
- AMD AMD Reports Second Quarter 2026 Financial Results (press release furnished with Form 8-K, August 4, 2026). Revenue by segment, third-quarter revenue outlook and highlights (Helios, Instinct MI400, the Anthropic partnership and others). https://www.sec.gov/Archives/edgar/data/2488/000000248826000121/q22026991.htm
- AMD AMD and Meta Announce Expanded Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs (press release furnished with Form 8-K, February 24, 2026). https://www.sec.gov/Archives/edgar/data/2488/000000248826000045/pressreleasedatedfebruary2.htm
- AMD AMD and Anthropic Announce Strategic Partnership to Deploy Up to 2 Gigawatts of AMD Instinct MI450 Series GPUs (press release, July 22, 2026). https://ir.amd.com/news-events/press-releases/detail/1292/amd-and-anthropic-announce-strategic-partnership-to-deploy-up-to-2-gigawatts-of-amd-instinct-mi450-series-gpus
- AMD Form 8-K (filed September 28, 2026). The September 26, 2026 merger agreement with World Labs Technologies, Inc. (purchase price of about US$8.2 billion, payable in AMD shares). https://www.sec.gov/Archives/edgar/data/2488/000000248826000182/amd-20260926.htm
- AMD AMD Instinct MI455X GPU (product page with specification table; checked October 15, 2026). https://www.amd.com/en/products/accelerators/instinct/mi400/mi455x.html
- AMD AMD Instinct MI355X GPU (product page with specification table; checked October 15, 2026). https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html
- AMD AMD Helios rackscale solution (product page with specifications and availability; checked October 15, 2026). https://www.amd.com/en/products/rackscale-solutions/helios.html
- U.S. Department of Energy Energy Department Announces New Public-Private Partnership Model, Two Supercomputers, to Accelerate American Dominance in Science and Technology (October 27, 2025). The AMD-accelerated Lux and Discovery supercomputers at Oak Ridge National Laboratory. https://www.energy.gov/articles/energy-department-announces-new-public-private-partnership-model-two-supercomputers
- AMD Official website (source of the logo image, which is the white SVG version distributed on the site; checked October 15, 2026). https://www.amd.com/en.html
14. Claim-to-source audit
| Claim in the text | Status | Reference |
|---|---|---|
| Revenue, segment revenue, R&D and operating income (2024 to 2025), about 31,000 employees | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm |
| No customer at 10% or more of revenue; receivables concentration | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm |
| The ZT Systems acquisition, the sale of its manufacturing business to Sanmina, and the Silo AI acquisition | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm |
| The OpenAI agreement and warrant, the MI308 charges (about US$800 million and about US$360 million reversed), and the export-control history | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/2488/000000248826000018/amd-20251227.htm |
| Revenue, Data Center segment revenue and operating income, and R&D for the second quarter and first half of 2026 | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| The OpenAI and Meta agreements (up to 6 GW) and the warrants not having vested | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| Commitments of about US$30.3 billion, lease guarantees of US$4.1 billion, and post-quarter investment commitments and leases | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| Helios shipments expected in the second half of 2026, and AMD not making or selling finished racks | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| The MI325 licence for China and the 25% tariff | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| Second-quarter results, the third-quarter revenue outlook (about US$13 billion), MI400 series, Helios adopters, the Anthropic partnership and the MEXT acquisition | Sourced | Reference 3 https://www.sec.gov/Archives/edgar/data/2488/000000248826000121/q22026991.htm |
| The Meta agreement (up to 6 GW; first gigawatt shipments from the second half of 2026) | Sourced | Reference 4 https://www.sec.gov/Archives/edgar/data/2488/000000248826000045/pressreleasedatedfebruary2.htm |
| The Anthropic partnership (up to 2 GW, first gigawatt from the first half of 2027, commitment to invest up to US$5 billion) | Sourced | Reference 5 https://ir.amd.com/news-events/press-releases/detail/1292/amd-and-anthropic-announce-strategic-partnership-to-deploy-up-to-2-gigawatts-of-amd-instinct-mi450-series-gpus |
| The merger agreement with World Labs (about US$8.2 billion, payable in AMD shares) | Sourced | Reference 6 https://www.sec.gov/Archives/edgar/data/2488/000000248826000182/amd-20260926.htm |
| MI455X specifications and launch date | Sourced | Reference 7 https://www.amd.com/en/products/accelerators/instinct/mi400/mi455x.html |
| MI355X specifications and launch date | Sourced | Reference 8 https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html |
| Helios specifications and availability, and the differing bandwidth statements on the page | Sourced | Reference 9 https://www.amd.com/en/products/rackscale-solutions/helios.html |
| The DOE supercomputers Lux and Discovery | Sourced | Reference 10 https://www.energy.gov/articles/energy-department-announces-new-public-private-partnership-model-two-supercomputers |
| Source of the logo image | Sourced | Reference 11 https://www.amd.com/en.html |
| The four company types and AMD's placement | Our calculation | This article's own classification. Company names are examples from firms covered in this series |
| Data Center at about 48% of 2025 revenue, and R&D at about 23% of revenue | Our calculation | Our calculation (16,635 / 34,639 and 8,091 / 34,639) |
| Helios shipments, deployments for OpenAI, Meta and Anthropic, the investment in Anthropic, and closing of the World Labs acquisition | Not yet confirmed | Company outlooks, plans and commitments, not settled facts. Items that could not be confirmed are stated in the text. Reference 2 https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm |
| The reading that AMD provides rack-scale designs while leaving manufacturing to others | Inference | This article's interpretation of the 10-K and 10-Q |
| The reading that it secures large customers' purchases with warrants and equity investment | Inference | This article's interpretation of the 10-Q and releases |
| The reading that open standards are its point of differentiation | Inference | This article's interpretation of the product pages |
| The reading that it is bringing in model and software expertise | Inference | This article's interpretation of the 10-K and 8-K |
| The materials engineer's view on memory, advanced process, cooling and connection | Commentary | General technical commentary, not an AMD disclosure. Figures are AMD's stated values from product pages |
Last updated 15 October 2026 / Troy Technical
Figures in this article come from AMD's statutory filings with the U.S. Securities and Exchange Commission (the 10-K, 10-Q and 8-Ks) and from AMD and U.S. Department of Energy announcements and product pages. No research-firm estimates or press-based figures have been used. Product performance values are AMD's stated theoretical peaks, 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.