COMPANY PROFILE / MATERIALS INFORMATICS
Applied Materials
An equipment maker that deposits and shapes materials has begun to use atomic-level computation and AI as tools for materials development
Applied Materials is one of the world's leading makers of semiconductor manufacturing equipment, based in the United States. From the standpoint of materials informatics (MI, the use of data and computation to narrow down candidate materials and conditions), it is not a company that sells AI for materials as a product. It is an industrial user that uses computation and AI to develop new materials for chips. The public record shows three threads: speeding up atomic-level materials calculation on GPUs (with NVIDIA), AIx, a platform for analyzing process data with AI (2021), and joining the AI Materials Foundry, an international framework for finding materials with AI, as a founding member. This profile keeps the whole-company business brief and concentrates on the use of AI and computation in materials development.
- Applied Materials in 30 seconds, and where it stands in materials informatics
- The logo and how the name is written
- Where the company sits in the materials informatics field
- Scale of the company (the whole company in brief)
- Resources committed to materials informatics
- What it has done so far
- What comes next
- What the company is aiming for in materials informatics
- Map of partnerships in materials informatics
- How AI and computation are used in materials development
- The risks this company carries
- Glossary, references and claim-to-source audit
1. Applied Materials in 30 seconds, and where it stands in materials informatics
AIx
AI Materials FoundryCalculation serves its own R&D; the Foundry is an outside framework
5 days to 2 hoursThe company's claim: 64 CPU cores against one GPU (March 2026)
Applied Materials' strength lies in the equipment that actually deposits and processes materials on wafers. That is what the company means when it calls itself a "materials engineering" company. For Applied, AI and computation are therefore not a way to discover and sell materials, but a way to speed up the development of equipment and processes. Its 10-K says it is "increasingly incorporating AI capabilities into the development of technologies, our business operations and our products and services", but none of the material checked states how many people work full time on MI, how much is invested in it, or what it has delivered in numbers Sourced. The only numbers the company gives are shorter computing times (company claims).
2. The logo and how the name is written
A note on names: the legal entity is Applied Materials, Inc. In Japanese it is written "Applied Materials" in katakana on the company's Japanese site. This article covers the U.S. listed company. "Applied" is also how the company refers to itself. Because the logo image on the official site was refused to automated requests, this article uses the image on Wikimedia Commons (registered as a public-domain text logo, with a note that it is a trademark; checked 23 October 2026). We have not confirmed that it matches the logo currently on the official site.
Logo source: https://commons.wikimedia.org/wiki/File:Applied_Materials_Inc._Logo.svg
Official site: https://www.appliedmaterials.com/us/en.html
3. Where the company sits in the materials informatics field
MI is a flow in which past data and computation narrow down "promising candidate materials and conditions", and experiments then confirm them (see our explainer on materials informatics for an overview). Companies involved in MI fall into four types, depending on which part of that flow they own.
Applied Materials is an industrial user. Unlike a battery or chemical company, however, it does not sell the material itself; it sells the equipment and processes that turn a material into a thin film built into a chip. In terms of the MI flow, it holds within one company the steps of evaluating candidate materials by calculation (prediction and design), fixing conditions, and confirming on its own equipment and in experiments (experiment). The company's account is that differences of an angstrom in structure, interfaces or defects greatly affect performance, so it matters to evaluate materials with physics-based calculation before deposition.
4. Scale of the company (the whole company in brief)
Applied is a large company, so only the scale of the whole company is shown, briefly Sourced. No figure for MI alone is disclosed.
| Measure (US$ million) | Fiscal 2024 | Fiscal 2025 | Fiscal 2026, nine months (to 26 July) |
|---|---|---|---|
| Revenue | 27,176 | 28,368 | 24,037 (21,568 a year earlier) |
| Operating income | 7,867 | 8,289 | n/a |
| Research, development and engineering (RD&E) | 3,233 | 3,570 | 3,055 (2,653 a year earlier) |
| Other measures | What it is | As of |
|---|---|---|
| Full-time employees | About 36,500 (in 25 countries; Asia-Pacific 46%, North America 42%, Europe and Middle East 12%) | 26 October 2025 |
| Business | Two reportable segments: Semiconductor Systems and Applied Global Services (services) | Fiscal 2025 |
| RD&E as a share of revenue | About 12.6% (our calculation: 3,570 divided by 28,368) | Fiscal 2025 |
| Where development takes place | Product development and engineering mainly in the U.S., India and Israel | 10-K for fiscal 2025 |
The figures above are for the whole company, and most revenue comes from semiconductor manufacturing equipment and its services. How much is spent on AI or materials calculation, and how many people work on it full time, is not disclosed Not yet confirmed. The rise in RD&E in fiscal 2026 (US$3,055 million over nine months, about 15% more than a year earlier) is not described as being only for AI or computation. When reading about this field, keep in mind that the scale of the company and the scale of its commitment to MI are two different things.
5. Resources committed to materials informatics
The table shows, for people, money, outside investment and links with universities or public bodies, what could and could not be confirmed from primary sources. "Not stated" is also a finding.
| Type of resource | What is there | Status |
|---|---|---|
| People | About 36,500 full-time employees (whole company). CuspAI's official page carries a statement by Applied Materials' Chief AI Officer. Headcount dedicated to AI or materials calculation is not disclosed | Sourced / Not disclosed |
| Money (R&D) | RD&E was US$3,570 million in fiscal 2025 (whole company). The share for AI and computation is not disclosed | Not disclosed |
| Money (facility) | The EPIC Center: about US$5 billion of capital investment (growing as customer projects start). A building of about 600,000 square feet with nearly four acres of cleanroom. It is a site for developing materials, equipment and process together, not dedicated to AI | Sourced |
| Compute | Uses NVIDIA GPUs (the blog names the B200) and libraries. The size of its own GPU fleet is not disclosed | Sourced / Not disclosed |
| Corporate ties | NVIDIA (speeding up simulation of materials, process and fab; blogs of February and March 2026) and CuspAI's AI Materials Foundry (listed as a founding member) | Sourced |
| Investments, acquisitions, joint ventures | No investment, acquisition or joint venture relating to MI appears in the material checked | Could not be confirmed |
| Ties with universities, institutes and public funding | Arizona State University, RPI, Stanford, UC Berkeley and UCLA are research partners of the EPIC Center. That is a framework for developing semiconductor materials, process and equipment together, and nothing limits it to AI or MI. The 10-K says in general terms that it receives government grants and tax credits, but does not tie them to MI | Sourced (link to MI not confirmed) |
Status of the main models, products and facilities
This covers what official material confirms. The speed figures are claims by the company (and NVIDIA), and no independent verification could be found.
| Name | Kind | Main specifics (as disclosed) | Form of release | Stage and date | Source |
|---|---|---|---|---|---|
| Ginestra | Atomic-scale materials simulation software | Calculates electronic properties and interactions of materials. Up to 10x faster than CPU-only runs with NVIDIA infrastructure and cuDSS | Could not be confirmed (not clear whether it is a customer product) | In use (blog of 18 February 2026) | Official blog |
| ACE+ | Multiphysics simulation | Calculates fluid flow, heat, plasma and electromagnetics for virtual chamber prototyping and optimization. Topography simulations up to 35x faster | Could not be confirmed | In use (same blog) | Official blog |
| Faster DFT calculation (cuEST) | Adoption of a method | Evaluating candidate materials: 5 days on 64 CPU cores down to 2 hours on one NVIDIA B200 GPU. Reaction discovery (about 15,000 searches): 1.5 weeks on 400 CPU cores down to 5 hours | Uses NVIDIA's library | In use (blog of 16 March 2026) | Official blog |
| AIx | AI platform for process data | Made up of ChamberAI (sensors and machine learning), AppliedPRO (recipe optimizer), digital twins and compute. Goal: halve development time and improve process windows by one third | Runs on Applied's own process and inspection equipment | Announced on 5 April 2021, described as already in use | News release |
| EPIC Center | Co-development facility | About US$5 billion (expected to grow), about 600,000 square feet. Five founding partners and five research universities | Used by partner companies and universities | Unveiled on 12 October 2026 | News release |
The March 2026 blog takes as its example evaluating candidate high-k insulating films such as HfO2 (hafnium oxide) at the atomic level before deposition Sourced. The figures say that evaluating candidate materials, which took 5 days on 64 CPU cores, now takes 2 hours on one GPU, and that reaction discovery went from 1.5 weeks on 400 cores to 5 hours on one GPU. Converted here, that is about 60x and about 50x, and the company states "55x" Our calculation. Three cautions apply. First, this is a shorter computing time, not a figure for prediction accuracy or for materials discovered. Second, the blog says the "required accuracy" is preserved, but gives no error against measurement. Third, the comparison is between 64 to 400 CPU cores and a single GPU, which are different machines, not a like-for-like test. Faster calculation does let you try more candidates, but whether a candidate can actually be deposited and give the film quality you want is decided by experiments on the equipment.
6. What it has done so far
| When | What happened | What it means |
|---|---|---|
| 5 April 2021 | Announces AIx: big data and AI to speed up developing and ramping processes on its equipment | The starting point of AI use on process data (the "process" side of MI) |
| 12 December 2025 | Files its Form 10-K for fiscal 2025, describing the building-in of AI capabilities and the risks | AI use appears in statutory disclosure |
| 18 February 2026 | Official blog: uses NVIDIA infrastructure for Ginestra, ACE+ and digital twins (from atoms to the fab) | Announces faster computation |
| 16 March 2026 | Official blog: speeds up DFT calculation on GPUs with cuEST (the HfO2 example) | Faster evaluation of candidate materials |
| 2026 (date not confirmed) | Joins CuspAI's AI Materials Foundry as a founding member | Participation in an outside framework for AI-driven materials discovery |
| 12 October 2026 | Unveils the EPIC Center (about US$5 billion of capital investment, five founding partners and five research universities) | A site for developing materials, equipment and process in parallel |
7. What comes next
Everything below is a company target or policy, not an achieved fact Not yet confirmed.
None of the following appears in the material checked Not yet confirmed.
Names and records of new materials found with AI / prediction accuracy of its calculations (against measurements, with dataset and conditions) / how far the AIx goal was met / AI and MI headcount and spending / what it is doing within the Foundry. Figures taken from news reports are not used.
8. What the company is aiming for in materials informatics
This chapter includes points the company does not state directly. They are inferences from what the primary sources say, set out with their grounds.
| What can be read | Primary source behind it | Status |
|---|---|---|
| AI and computation are tools to shorten the development cycle of materials, equipment and process | The EPIC release describes developing materials, equipment and process in parallel to remove handoffs, and the February 2026 blog speaks of "compressing R&D cycles" | Inference |
| It leans toward using an outside foundation rather than building materials-AI models entirely itself | The 10-K and 10-Q say it uses third-party AI providers. Its calculation runs on NVIDIA libraries, and it joined CuspAI's framework for materials discovery | Inference |
| The target is materials for semiconductors (dielectrics, interfaces, deposition reactions) | The March 2026 blog gives as examples the HfO2 dielectric, interfaces, and reactions of gas molecules on solid surfaces | Sourced (as far as the text goes) |
| The method is mainly physics-based calculation (DFT and the like), and use of machine-learning interatomic potentials cannot be confirmed | The blog says "physics-based simulation" and does not mention machine-learning potentials | Inference |
The March 2026 blog says that "after the right molecule has been identified", the next challenge is to deposit it reliably at manufacturing scale in a vacuum chamber, and that to optimize the complex reactions in which gas molecules meet solid surfaces, Applied uses reaction-discovery techniques Sourced. Here the weight of AI and computation seems to lie less in discovering a new material than in bringing existing candidates to conditions usable in volume production (processing temperature, impurities) Not yet confirmed. That is our interpretation, though, and not something the company has declared as strategy. Joining CuspAI's Foundry, on the other hand, can be read as showing interest in "searching for new materials" too.
9. Map of partnerships in materials informatics
10. How AI and computation are used in materials development
Applied Materials' official blogs describe calculation in materials development in three stages: atoms (materials), then process (inside the tool), then the fab.
| Stage | Tools | What the company says |
|---|---|---|
| 1 Material (atoms) | Ginestra, DFT | Evaluates candidate dielectrics, interfaces and material alternatives. Searches for deposition reaction paths that lower processing temperature and limit impurities. Says thousands of material combinations can be evaluated without physical trials |
| 2 Process (inside the tool) | ACE+, PhysicsNeMo | Virtual chamber prototyping and optimization of hardware and recipes. Turns ACE+ data into real-time digital twins, giving near-instant feedback when parameters change |
| 3 Fab | Omniverse, digital twins | Examines layout, throughput and process control of the whole factory virtually before implementation |
| (Process data) | AIx (ChamberAI, AppliedPRO) | Finds the few correlations that matter among thousands of process variables, using tool sensors and wafer measurement data, to optimize recipes (2021 description) |
Applied Materials' account is built in an order that checks calculation results by actually depositing on equipment. Candidate materials are evaluated with DFT, reaction paths are searched, the chamber is prototyped virtually, and finally the film is made on real equipment such as an integrated Endura system Sourced. When you use MI or calculation in your own materials development, deciding at the start which questions go to calculation and where experiments take over keeps you from over-trusting the results. The idea behind AIx, which is to narrow down influential process variables from equipment and wafer measurement data, can also apply to deposition, coating and polymerization processes outside semiconductors. Note, though, that Applied Materials' material gives no figure for how accurately its calculations predicted measurements. If you adopt such methods, a practical first step is a small test of whether the predictions hold on your own known data (see our explainer on how to read MI results and their reproducibility).
11. The risks this company carries
| Risk | What it is | Status |
|---|---|---|
| Risks that come with using AI | The 10-K and 10-Q say there is no guarantee that AI use will pay off, that algorithms and data may be flawed, and that intellectual property may leak or be infringed | Sourced |
| Reliance on outside AI providers | The 10-K and 10-Q say it uses third-party AI providers and depends on continued access to them | Sourced |
| Results are hard to check from outside | The speed figures come from the company (and NVIDIA). Calculation accuracy and any record of materials found with AI could not be confirmed | Not yet confirmed |
| Completing the EPIC Center and realizing its benefits | The 10-K and 10-Q list completing the EPIC Center and other major infrastructure on schedule and on budget, and realizing the expected benefits, as a risk. How much AI contributes is a separate question | Sourced |
| Swings in the chip market | The 10-K says the semiconductor industry has historically been cyclical and customer demand varies. That bears on the funds available for R&D | Sourced |
| Blurred scope of MI | EPIC's partnerships cover the development of materials, equipment and process as a whole, not only AI or MI. Reading EPIC's size as investment in AI or MI would misstate the picture | Inference |
12. Glossary
- Materials informatics (MI)
- Using data and computation to narrow down promising candidate materials and conditions, then confirming them by experiment.
- DFT (density functional theory)
- A standard way to calculate the electronic structure of matter. Accurate, but slow.
- HfO2 (hafnium oxide)
- A high-dielectric-constant (high-k) material used as the gate insulator in transistors.
- GAA (gate-all-around)
- A next-generation transistor structure in which the gate surrounds the channel on all sides.
- Reaction pathway search
- Searching by calculation for many routes by which source molecules react on a surface to form a film, to find the most efficient one.
- Deposition and chamber
- Deposition is making a thin film on a wafer. A chamber is the vacuum vessel in which the reaction takes place.
- Digital twin
- A model of a real tool or factory built inside a computer, in which changed conditions can be tried virtually.
- GPU / B200 / cuEST
- A GPU is a computer part that is good at parallel calculation. B200 is a model of NVIDIA GPU. cuEST is an NVIDIA library that runs electronic-structure calculations on GPUs.
- AIx
- Applied Materials' process-data analysis platform. Short for Actionable Insight Accelerator.
- EPIC Center
- Equipment and Process Innovation and Commercialization Center. The co-development facility the company has unveiled.
- RD&E
- Research, development and engineering. The expense line for research, development and engineering.
- Form 10-K / 10-Q
- The annual and quarterly reports filed with the U.S. SEC.
13. References
- Applied Materials, Inc. Form 10-K for the fiscal year ended 26 October 2025 (filed with the U.S. SEC on 12 December 2025). Business overview, revenue, R&D expense, headcount, risk factors on the use of AI, and the mention of the EPIC Center. https://www.sec.gov/Archives/edgar/data/6951/000162828025056742/amat-20251026.htm
- Applied Materials, Inc. Form 10-Q for the quarterly period ended 26 July 2026 (filed with the U.S. SEC on 20 August 2026). Revenue and R&D expense for the third quarter and nine months of fiscal 2026, and risk factors on the use of AI. https://www.sec.gov/Archives/edgar/data/6951/000162828026058235/amat-20260726.htm
- Applied Materials, Inc. "Applied Materials AI(x) Platform Harnesses the Power of Big Data and AI to Accelerate Semiconductor Technology Breakthroughs from Lab to Fab" (news release, 5 April 2021). Components and goals of AIx. https://ir.appliedmaterials.com/news-releases/news-release-details/applied-materials-aix-platform-harnesses-power-big-data-and-ai
- Applied Materials, Inc. "Applied Materials Collaborates With NVIDIA to Accelerate End-to-End Chip Manufacturing" (official blog, 18 February 2026). Use of NVIDIA infrastructure in Ginestra, ACE+ and digital twins, and the claimed speedups. https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-accelerates-end-to-end-chip-manufacturing-with-nvidia-ai.html
- Applied Materials, Inc. "Applied Materials Accelerates Materials Simulation for the AI Era in Collaboration with NVIDIA" (official blog, 16 March 2026). Speeding up DFT calculations (cuEST), the HfO2 example and the computing-time comparison. https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-collaborates-nvidia-ai-era.html
- Applied Materials, Inc. "Applied Materials Unveils $5 Billion EPIC Center in Silicon Valley to Accelerate Semiconductor Innovation for the AI Era" (news release, 12 October 2026). Scale, investment, partners and operating concept of the EPIC Center. https://www.appliedmaterials.com/us/en/newsroom/press-releases/101226-applied-unveils-5b-epic-center-silicon-valley-accelerate-innovation-ai-era.html
- CuspAI Official website (cusp.ai), the "Announcing the AI Materials Foundry" section. Lists Applied Materials as a founding member and carries a statement by its Chief AI Officer (checked 22 October 2026; the page gives no announcement date). https://cusp.ai/
14. Claim-to-source audit
| Claim in the text | Status | Source |
|---|---|---|
| Legal name, headquarters (Santa Clara, California), Delaware corporation (founded 1967) and Nasdaq listing | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/6951/000162828025056742/amat-20251026.htm |
| Fiscal 2025 (ended 26 October 2025): revenue of US$28,368 million (27,176 million the year before), operating income of 8,289 (7,867) and R&D (RD&E) of 3,570 (3,233) | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/6951/000162828025056742/amat-20251026.htm |
| About 36,500 employees (26 October 2025, in 25 countries); product development and engineering mainly in the U.S., India and Israel | Sourced | Reference 1 https://www.sec.gov/Archives/edgar/data/6951/000162828025056742/amat-20251026.htm |
| Fiscal 2026, nine months: revenue of 24,037 million (21,568 million a year earlier) and RD&E of 3,055 (2,653) | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/6951/000162828026058235/amat-20260726.htm |
| RD&E as about 12.6% of revenue (fiscal 2025) | Our calculation | Our own division in this article (3,570 divided by 28,368). |
| 10-K and 10-Q: AI capabilities are being built into technology development, operations and products; risks include flawed algorithms and data and reliance on outside AI providers | Sourced | References 1, 2 https://www.sec.gov/Archives/edgar/data/6951/000162828025056742/amat-20251026.htm https://www.sec.gov/Archives/edgar/data/6951/000162828026058235/amat-20260726.htm |
| AIx (5 April 2021): made up of ChamberAI, inline metrology, AppliedPRO, digital twins and compute; goal of halving development time and widening process windows by one third | Sourced | Reference 3 https://ir.appliedmaterials.com/news-releases/news-release-details/applied-materials-aix-platform-harnesses-power-big-data-and-ai |
| Ginestra: atomic-level simulation of material behavior; up to 10x faster than CPU-only runs with NVIDIA infrastructure and cuDSS (company claim) | Sourced | Reference 4 https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-accelerates-end-to-end-chip-manufacturing-with-nvidia-ai.html |
| ACE+: multiphysics simulation of fluid flow, heat, plasma and electromagnetics; topography simulations up to 35x faster on NVIDIA infrastructure (company claim) | Sourced | Reference 4 https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-accelerates-end-to-end-chip-manufacturing-with-nvidia-ai.html |
| PhysicsNeMo turns ACE+ data into digital twins, and Omniverse is used to build fab digital twins | Sourced | Reference 4 https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-accelerates-end-to-end-chip-manufacturing-with-nvidia-ai.html |
| DFT: with cuEST, evaluating candidate materials fell from 5 days on 64 CPU cores to 2 hours on one B200 GPU, and reaction discovery (about 15,000 pathway searches) from 1.5 weeks on 400 cores to 5 hours (the company says "55x") | Sourced | Reference 5 https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-collaborates-nvidia-ai-era.html |
| The HfO2 example: candidate dielectrics, interfaces and material alternatives are evaluated by materials simulation in R&D | Sourced | Reference 5 https://www.appliedmaterials.com/us/en/newsroom/blogs/applied-materials-collaborates-nvidia-ai-era.html |
| 5 days to 2 hours is about 60x, and 1.5 weeks to 5 hours is about 50x | Our calculation | Our own conversion in this article, with 1.5 weeks taken as 252 hours. The company's figure is "55x". |
| EPIC Center: about US$5 billion of capital investment (growing as customer projects start), a roughly 600,000-square-foot building, founding partners TSMC, Intel, SK hynix, Micron and Samsung, and five research universities | Sourced | Reference 6 https://www.appliedmaterials.com/us/en/newsroom/press-releases/101226-applied-unveils-5b-epic-center-silicon-valley-accelerate-innovation-ai-era.html |
| The EPIC operating idea: develop materials, equipment and process in parallel and remove handoffs | Sourced | Reference 6 https://www.appliedmaterials.com/us/en/newsroom/press-releases/101226-applied-unveils-5b-epic-center-silicon-valley-accelerate-innovation-ai-era.html |
| Listed as a founding member of the AI Materials Foundry (CuspAI), with a statement by its Chief AI Officer | Sourced | Reference 7 https://cusp.ai/ |
| The date of joining the Foundry | Not yet confirmed | Not stated on CuspAI's page. Our own note: this article does not infer a date. |
| Third-party verification of AIx, Ginestra or ACE+, comparison of calculated accuracy with measurement, and any record of new materials found with AI | Not yet confirmed | Not stated in the material checked. Our own note: no estimate is made here. |
| MI-dedicated headcount and spending, and any MI-related investment, acquisition or joint venture | Not yet confirmed | Not stated in the material checked. Our own note: this article does not assert that none exist. |
| Whether machine-learning interatomic potentials are used | Not yet confirmed | Not stated in the material checked. Our own note: the blogs describe mainly physics-based (DFT-type) calculation. |
| The classification of the company (industrial user) | Our calculation | Our own grouping for this article. |
| The reading of the company's aims (four points in chapter 8) | Inference | Our own interpretation of statements in the official material. |
| Risk items that no filing states directly | Inference | Our own interpretation of the 10-K and 10-Q risk factors and the official blogs. |
Last updated: 22 October 2026 / Troy Technical
All figures in this article come from filings with the U.S. Securities and Exchange Commission and from official statements by Applied Materials and CuspAI. No market-research estimates or figures taken from news reports are used. Passages marked "Inference" are our reading of primary sources and are not statements by the company. The speedup figures for computation are claims by the company (and NVIDIA), and we could not confirm independent verification.