AI Chips · Brand guide

Amazon Trainium: The Complete Guide

Amazon's own AI chips, designed by its Annapurna Labs team and rented through AWS. What Trainium and Inferentia are, how they work, the prices AWS publishes, the gigawatt deals with Anthropic and OpenAI, and the facts buried in Amazon's filings. Three reading levels, every number sourced.

Stuck at any point? Ask an AI about this page →
Amazon Trainium 16 sections · 3 levels 58 linked sources Checked September 2026
How to read this page. Every brand in our AI Chips section uses the same 16 sections in the same order. Each section starts Simple, then goes Deeper, then Expert: stop wherever you have what you need. The small numbers are sources: click one to open the original document. Where a company has not published something, we say so rather than estimate.

1.At a glance

SimpleStart here

Trainium is Amazon's own chip for training AI models, and Inferentia its chip for running them 1 2. Amazon does not sell the chips; it rents them through its cloud, AWS 3. They are designed by Annapurna Labs, a chip company Amazon bought in 2015 4 5. The newest chip, Trainium3, went on sale in December 2025 6.

What it isAmazon-designed chips for AI: Trainium (training and inference) and Inferentia (inference) 1 2
Latest chipTrainium3, on a 3 nm process, in Trn3 UltraServers 6
Who designs itAnnapurna Labs, part of AWS 5
Biggest usersAnthropic, OpenAI, and Amazon's own Bedrock service 7 8
DeeperThe detail

Amazon's chips have become a large business. In July 2026 Andy Jassy said Amazon's chips business had passed an annual revenue run rate of more than $25 billion 7; that figure covers Trainium, Graviton processors and Nitro chips together 3. He said Trainium2 is about 30% better on price-performance than comparable GPUs and "has largely sold out" 3.

ExpertFor specialists

Amazon's June 2026 quarter: net sales $200.6 billion (up 20%), AWS sales $42.2 billion (up 37%, its fastest growth in 18 quarters) and AWS operating income $16.6 billion 7. Trainium commitments stood at more than $225 billion in April 2026 9. Amazon planned about $200 billion of capital spending for 2026 in February 8; on the July call it raised that to about $220 billion, citing higher memory costs 10.

2.Company card

SimpleStart here

Trainium is made by Amazon, through its cloud business Amazon Web Services (AWS). Amazon was founded in 1994 11; AWS's first big service, S3 storage, launched on 14 March 2006 12.

Legal nameAmazon.com, Inc. 13
Headquarters410 Terry Avenue North, Seattle, Washington 98109-5210, USA 13 (map)
Chief ExecutiveAndy Jassy 11
AWS Chief ExecutiveMatt Garman 11
EmployeesAbout 1,576,000 full- and part-time (31 December 2025) 13
StockAMZN, Nasdaq Global Select Market 13
Investor siteir.aboutamazon.com
DeeperThe detail

The chips come from Annapurna Labs, an Israeli start-up Amazon bought in 2015 4. It designs AWS's Nitro, Graviton and AI chips, from a lab on the outskirts of Austin, Texas 5; Amazon's jobs site also lists Annapurna teams in Haifa and Tel Aviv, Israel, and Boston 14. The purchase price was not disclosed by Amazon; press reports put it at about $350 million 15.

ExpertFor specialists

AWS sold $128.7 billion in 2025, up 20%, with $45.6 billion of operating income 16. Amazon's total 2025 net sales were $716.9 billion 16. Jeff Bezos, the founder, is Executive Chair 11.

3.The story

SimpleStart here
  • 2015 Amazon buys Annapurna Labs 4.
  • 2018 Announces Inferentia, its first AI chip 17.
  • 2020 Announces Trainium 18.
  • 2022 Trainium instances (Trn1) go on sale 19.
  • 2023 Inferentia2 on sale; Trainium2 announced 20 21.
  • 2024 Trainium2 on sale 22.
  • 2025 Project Rainier for Anthropic switches on; Trainium3 goes on sale 23 6.
  • 2026 OpenAI deal for about 2 gigawatts of Trainium; Anthropic deal grows to up to 5 gigawatts 24 25.
DeeperThe detail

Amazon started with inference. Inferentia was announced in November 2018 17 and its Inf1 instances launched in December 2019 with up to 16 chips 26. Trn1, the first Trainium instances, went on sale in October 2022 with up to 50% lower training cost than comparable GPU instances, on AWS's claim 19. Trainium2 was announced in November 2023 21 and went on sale in December 2024, when Trainium3 was also first announced 22.

ExpertFor specialists

Project Rainier, the Trainium2 cluster AWS built for Anthropic, was activated in October 2025 with nearly 500,000 chips, "70% larger than any prior AWS AI platform"; Amazon said Anthropic would have more than a million Trainium2 chips by the end of 2025 23. By February 2026 AWS had landed 1.4 million Trainium2 chips, all subscribed 8.

4.The lineup

SimpleStart here
NameWhat it isOn sale
Inferentia2 (Inf2)Chip for running models; up to 12 per serverApr 2023 20
Trainium (Trn1)First-generation training chip; 16 per serverOct 2022 19
Trainium2 (Trn2)Second generation; 16 per server, 64 per UltraServerDec 2024 22
Trainium3 (Trn3)Third generation, 3 nm; up to 144 per UltraServerDec 2025 6
Trainium4Announced; supports NVIDIA NVLink FusionDeliveries expected 2027 27 24
DeeperThe detail

Despite the name, Trainium now does most of the inference too: Amazon says Trainium2 powers the majority of inference on Amazon Bedrock, its service for running AI models 8. Jassy wrote in April 2026 that Trainium3, which "just started shipping at the start of 2026", is 30 to 40% more price-performant than Trainium2 and "nearly fully-subscribed" 3.

ExpertFor specialists

Trainium4 is promised to deliver at least six times Trainium3's FP4 performance, three times its FP8 performance and four times its memory bandwidth 27. In April 2026 Jassy said it was "still about 18 months from broad availability", with a significant share already reserved 3. AWS has not published its process node or memory size 27.

5.How it’s made

SimpleStart here

Amazon designs its chips, but does not say who manufactures them 5.

  1. Design: Annapurna Labs, in Austin, Texas and Israel 5 14.
  2. Chip factory: not disclosed by Amazon. Press reports name TSMC for Trainium2 and Trainium3 28 15.
  3. Memory: Trainium3 carries 144 GB of HBM3e memory 6; Amazon does not name the supplier.
  4. Servers: AWS builds Trn3 UltraServers of up to 144 chips, joined by its own NeuronSwitch fabric 6 27.
DeeperThe detail

Trainium3 is "the first 3nm AI chip from AWS" 27. Other partners are named only in the press: CommonWealth Magazine calls Taiwan's Alchip Trainium's main design-services provider and says servers are assembled mainly by Wiwynn 29. Marvell has an official five-year agreement to supply AWS with "custom AI products" and networking chips, but its release does not name Trainium 30.

ExpertFor specialists

The Austin lab itself is small: press who toured it in March 2026 described it as about the size of two large conference rooms, with a separate test data centre in a co-location facility, and said Trainium3 is liquid-cooled 15. AWS has not published street addresses for its chip labs 5.

6.Where it’s made: factories and addresses

SimpleStart here

Amazon does not disclose which factory makes Trainium, so there is no factory address to give. What it does publish:

SiteLocationWhat happens there
Annapurna Labs chip labOutskirts of Austin, Texas, USA (no street address published) 5Chip design and testing 5
Annapurna Labs teamsHaifa and Tel Aviv, Israel; Boston, USA 14Chip engineering 14
Project RainierIncludes St. Joseph County, Indiana, USA; spread across several US data centres 23Anthropic's Trainium2 cluster 23
AWS Asia Pacific (Mumbai)Mumbai, IndiaInferentia2 instances; Trainium (Trn1) by Capacity Block 31 32
DeeperThe detail

Amazon has committed $11 billion to St. Joseph County, Indiana, and in November 2025 added $15 billion for Northern Indiana with 2.4 gigawatts of data centre capacity 33. Press reports say Project Rainier also includes sites in Mississippi and Pennsylvania and draws about 1 gigawatt; Amazon has not confirmed those details 34.

ExpertFor specialists

Trainium can now leave AWS's own buildings: "AWS AI Factories" place Trainium or NVIDIA systems in customers' data centres, starting with an AI zone of up to 150,000 chips for HUMAIN in Saudi Arabia 35. AWS added 3.9 gigawatts of power capacity in 2025 and expects to double its total power by the end of 2027 3.

7.How it works

SimpleStart here

Each Trainium chip has several NeuronCores. Inside each core, a large grid of maths circuits (a 128 by 128 "systolic array") does the matrix multiplication that dominates AI, while smaller engines handle everything else 36. Chips link to each other with AWS's NeuronLink, and servers join through AWS's EFA network 23.

DeeperThe detail

A NeuronCore has four engines: a tensor engine for matrix maths, a vector engine for sums across data, a scalar engine for functions such as GELU, and a "GpSimd" engine of eight programmable processors for custom code 36. Each core has its own software-managed on-chip memory (24 MiB on the first Trainium) 36. Trainium2 has eight NeuronCores per chip and adds support for "structured sparsity", skipping zeros for up to four times the speed 37 38.

ExpertFor specialists

Trainium3's cores (NeuronCore-v4) run the new MXFP8 and MXFP4 number formats at four times their BF16 speed, speed up the exponential maths used in attention fourfold, and add hardware support for routing in mixture-of-experts models 39 1. Its UltraServer uses NeuronSwitch-v1, an all-to-all fabric with twice the bandwidth within each UltraServer 27. Software is the Neuron SDK: PyTorch, JAX, vLLM and Hugging Face work on it, and NKI lets engineers program the chip directly 1. Neuron 2.27 (December 2025) added native PyTorch ("TorchNeuron") and an open-source NKI compiler, both in private beta 40.

8.Specs explained

SimpleStart here

AWS's own figures, per chip. "Dense" and "sparse" are explained in our How AI chips work guide.

ChipAI computeMemoryMemory speedProcess
Inferentia2Up to 190 TFLOPS FP1632 GB HBM—Not published 2
Trainium (1)3.4 PFLOPS BF16 per 16-chip server32 GiB HBM820 GB/sNot published 41 36
Trainium21.3 PFLOPS FP8 dense (5.2 sparse)96 GiB HBM2.9 TB/sNot published 37
Trainium32.52 PFLOPS FP8144 GB HBM3e4.9 TB/s3 nm 6

AWS's Neuron documentation gives slightly different memory speeds: 3 TB/s for Trainium2 and 4.7 TB/s for Trainium3 38 39.

DeeperThe detail
SystemChipsAI compute (FP8)Memory
Trn2 instance16 Trainium220.8 PFLOPS dense1.5 TiB 37
Trn2 UltraServer64 Trainium283 PFLOPS dense6 TiB 37
Trn3 UltraServerUp to 144 Trainium3362 PFLOPS20.7 TB 6

AWS says a Trn3 UltraServer has up to 4.4 times the performance and four times the performance per watt of Trainium2 6. Those are AWS's claims.

ExpertFor specialists

AWS does not say whether Trainium3's 2.52 petaflops is dense or sparse 6; its Neuron guide lists 315 teraflops of MXFP8 per core across eight cores, which is 2.52 petaflops, so it appears to be dense 39. Its BF16 rate per core (79 teraflops) is the same as Trainium2's: the gain comes from the new 8-bit and 4-bit formats 38 39. EC2 UltraClusters 3.0 can link up to a million Trainium chips 27.

9.Official pricing

SimpleStart here

Amazon does not sell Trainium chips, and publishes rental prices only for some of them. AWS list prices, US dollars per hour 42 43 32:

MachineChipsOn demand1-year / 3-year reserved
trn1.32xlarge16 Trainium$21.50$12.60 / $7.59 42
trn1.2xlarge1 Trainium$1.34$0.79 / $0.4744 42
inf2.48xlarge12 Inferentia2$12.98$7.79 / $5.19 43
inf2.xlarge1 Inferentia2$0.76$0.45 / $0.30 43
Trn2 and Trn3On-demand price not published by the company 44 45
DeeperThe detail

Capacity Blocks (booked in advance)

MachinePrice per hourPer chipWhere
trn2.48xlarge (16 Trainium2)$35.7608$2.235US East (Ohio) 32
trn2.3xlarge (1 Trainium2)$2.235$2.235Melbourne, São Paulo 32
trn1.32xlarge (16 Trainium)$9.532$0.596US, Mumbai and others 32

AWS says Capacity Block prices are next updated in October 2026 32. No Trainium3 Capacity Block price is listed 32.

ExpertFor specialists

For comparison, AWS lists an NVIDIA H100 at $5.19 per GPU-hour on the same Capacity Block page (see our NVIDIA guide); Trainium2 at $2.235 is less than half that, though the chips are not equal in speed. AWS claims Trn2 has 30 to 40% better price-performance than its GPU-based P5e and P5en machines 44. In India, Trn1 Capacity Blocks are sold in the Mumbai region at the same $9.532 an hour 32.

10.The software lock-in

SimpleStart here

Amazon's advantage is that it controls the chip, the servers, the network and the cloud where customers already run their businesses 5. Its biggest AI service, Bedrock, runs most of its inference on Trainium 3, so Amazon is its own largest customer.

DeeperThe detail

AWS's Neuron software is the answer to NVIDIA's CUDA: it supports PyTorch, JAX, vLLM, Hugging Face Transformers and TorchTitan, and NKI lets engineers write low-level code for the chip 1. Anthropic's engineers write kernels for Trainium together with Annapurna Labs and contribute to Neuron 46. The latest version, Neuron 2.32, supports every generation from Inferentia to Trainium3 47.

ExpertFor specialists

Amazon is also opening Trainium to NVIDIA's ecosystem: Trainium4 will support NVIDIA's NVLink Fusion connection 27. Jassy wrote that the chips business would be worth a run rate of about $50 billion if Amazon sold its chips at market prices to AWS and others, and that it is "quite possible" Amazon will sell racks to third parties 3.

11.Who buys it and why

SimpleStart here

Anthropic is the anchor customer. Amazon has invested $8 billion in it and AWS is its "primary training partner" 48. In April 2026 the deal grew: Anthropic committed more than $100 billion to AWS over ten years, for up to 5 gigawatts of capacity, and already uses more than a million Trainium2 chips 25.

DeeperThe detail
  • OpenAI (February 2026): Amazon agreed to invest $50 billion ($15 billion now, $35 billion on conditions); OpenAI will use about 2 gigawatts of Trainium3 and Trainium4 capacity, within an AWS agreement expanded by $100 billion over eight years 24.
  • Named Trn2 customers: Databricks, poolside, Itaú Unibanco, Ricoh, Hugging Face, Datadog and others 44.
  • Trainium3 customers: Anthropic, Karakuri, Metagenomi, NetoAI, Ricoh, Splash Music and Decart, several reporting up to 50% lower costs 27.
  • Apple: said at re:Invent 2024 it was "in the early stages of evaluating Trainium 2" 49.
  • Amazon itself: Rufus, its shopping assistant, used more than 80,000 Inferentia and Trainium chips on Prime Day 2024 50.
ExpertFor specialists

Amazon's July 2026 release named "multi-year, multi-gigawatt commitments" from Anthropic and OpenAI, plus Trainium customers including TwelveLabs, Poolside, Uber and Pinterest 7. Anthropic also uses Google's TPUs and NVIDIA's GPUs, and signed for up to a million Google TPUs in October 2025 51.

12.Hidden in plain sight

SimpleStart here

Public facts buried in Amazon's filings and letters. Each links to the exact document.

  1. Anthropic made Amazon $53.4 billion in one quarter, as pre-tax gains, mostly on its Anthropic stake, in the June 2026 quarter 7.
  2. The chips business is racing: Trainium and Graviton alone passed a $10 billion run rate in February 2026 8; by July Amazon said its chips business had passed $25 billion 7.
  3. Trainium commitments passed $225 billion by April 2026 9.
  4. Amazon's cash flow turned negative: free cash flow over the twelve months to June 2026 was an outflow of $7.6 billion, as capital spending rose 64% 7.
  5. Anthropic's spending is AWS revenue: the filing says the commercial deal is "primarily for the provision of AWS cloud services, which includes the use of AWS chips" 52.
DeeperThe detail
  1. Amazon's Anthropic notes were valued at $42.2 billion at 31 March 2026, against $8 billion invested 52.
  2. The new $20 billion Anthropic facility pays out only as compute milestones are reached 52.
  3. Few suppliers: "We rely on a limited group of suppliers for semiconductor products" 53.
  4. Trainium2 was sold out: 1.4 million chips landed and "fully subscribed" by February 2026 8.
  5. Most of Bedrock runs on Trainium, not NVIDIA 8.
ExpertFor specialists
  1. OpenAI's $35 billion second tranche is mandatory on milestones or an IPO before 31 December 2028 52.
  2. Total commitments reached $569.3 billion at 31 March 2026, including $106.3 billion of leases not yet started 52.
  3. Trainium3's BF16 speed per core did not rise over Trainium2; the gain is in 8-bit and 4-bit maths 38 39.
  4. Capital spending in the June quarter alone was $54.2 billion 7.
  5. AWS's water use is 0.15 litres per kilowatt-hour, against 0.375 for the industry, by Amazon's figures 23.

13.Weak spots

SimpleStart here

Limits, from Amazon's own documents and reporting:

  • Few suppliers for semiconductors 53.
  • Only through AWS: Trainium is not sold as a chip 3.
  • Two customers dominate: Anthropic and OpenAI hold the multi-gigawatt commitments 7.
  • Software still maturing: native PyTorch was only in private beta in December 2025 40.
DeeperThe detail

Customers have not all been happy. Business Insider reported an internal Amazon document from July 2025 in which Cohere found Trainium1 and Trainium2 underperforming NVIDIA's H100 and access to Trainium2 "extremely limited", and Stability AI called Trainium2 "less competitive" on latency and cost; Amazon said Trainium2 was fully subscribed 54. These are press reports.

ExpertFor specialists

Anthropic, the anchor customer, spreads its work across Trainium, Google TPUs and NVIDIA GPUs 51. AWS publishes no on-demand price for Trainium2 or Trainium3, which makes outside comparison harder 44 45. Amazon's annual report also names the availability and price of "land, water, and energy" as a risk 53.

14.India angle

SimpleStart here

In June 2026 Amazon announced $48 billion of investment in India from 2026 to 2030 55. AWS plans $21 billion for cloud and AI infrastructure in India from 2026 to 2030, including a new data centre in Bharat Future City, Hyderabad, announced in July 2026 56.

DeeperThe detail

AWS runs two Indian regions: Mumbai (since 2016) and Hyderabad (since November 2022) 57. Inferentia2 instances have been available in Mumbai since December 2023 31, and Trainium (Trn1) can be booked there by Capacity Block at $9.532 an hour 32. We found no official listing of Trainium2 or Trainium3 in India.

ExpertFor specialists

Amazon's 2023 India plan was ₹1,05,600 crore (US$12.7 billion) by 2030 57; the 2026 figures are much larger 55 56. Under the IndiaAI Mission, AWS offers its AI services through an empanelled partner, SHI India; the chip types were not specified 58.

15.What’s next

SimpleStart here

Only what Amazon has announced:

  • Trainium3: ramping through 2026; nearly all supply expected to be committed by mid-2026 8.
  • Trainium4: deliveries expected to begin in 2027 24.
  • Anthropic: nearly 1 gigawatt of Trainium2 and Trainium3 by the end of 2026 25.
DeeperThe detail

On the July 2026 call Jassy said there is "a real chance" Amazon will sell Trainium outside AWS in future 10. Anthropic plans Trainium3 capacity later in 2026 and more inference capacity in Asia and Europe 25.

ExpertFor specialists

The test is whether Trainium can win customers beyond its two anchors. OpenAI's 2 gigawatts depend on Trainium3 and Trainium4 arriving on time 24, and AWS has not yet said who makes Trainium4 or on what process 27.

16.Sources

58 sources, all checked September 2026. Official = the company or organisation’s own page; Filing = a document filed with a regulator; Press = news coverage, used only where no official source exists and labelled in the text.

  1. AWS: AWS Trainium product pageOfficial
  2. AWS: AWS Inferentia product pageOfficial
  3. About Amazon: Andy Jassy's 2025 letter to shareholders (9 Apr 2026)Official
  4. Amazon Science: How silicon innovation became the secret sauce behind AWS's successOfficial
  5. About Amazon: Take a look inside the lab where AWS makes custom chipsOfficial
  6. AWS What's New: Amazon EC2 Trn3 UltraServers (2 Dec 2025)Official
  7. Amazon (SEC exhibit): Amazon.com announces second-quarter 2026 results (30 Jul 2026)Filing
  8. Amazon Investor Relations: Amazon.com announces fourth-quarter results (Feb 2026)Official
  9. About Amazon: Andy Jassy on Amazon's chips business, Q1 2026 (29 Apr 2026)Official
  10. Investing.com (press): Amazon Q2 2026 earnings call transcriptPress
  11. companiesmarketcap (10-K copy): Amazon Form 10-K fiscal 2025: company details and executive officersResearch
  12. AWS News Blog: Twenty years of Amazon S3 (2026)Official
  13. Amazon: 2025 Annual Report and Form 10-KFiling
  14. Amazon Jobs: Annapurna Labs team pageOfficial
  15. TechCrunch (press): An exclusive tour of Amazon's Trainium lab (22 Mar 2026)Press
  16. About Amazon: Amazon fourth-quarter 2025 results (5 Feb 2026)Official
  17. AWS What's New: Announcing Amazon Inferentia (28 Nov 2018)Official
  18. TechCrunch (press): AWS launches Trainium, its new custom ML training chip (1 Dec 2020)Press
  19. Amazon Press Center: AWS announces general availability of EC2 Trn1 instances (10 Oct 2022)Official
  20. AWS What's New: Amazon EC2 Inf2 instances generally available (13 Apr 2023)Official
  21. Amazon Press Center: AWS unveils next-generation AWS-designed chips (28 Nov 2023)Official
  22. Amazon Press Center: AWS Trainium2 instances now generally available (3 Dec 2024)Official
  23. About Amazon: AWS activates Project Rainier (Oct 2025)Official
  24. About Amazon: Amazon and OpenAI strategic partnership and investment (27 Feb 2026)Official
  25. Anthropic: Anthropic and Amazon expand compute partnership (20 Apr 2026)Official
  26. AWS What's New: Introducing Amazon EC2 Inf1 instances (3 Dec 2019)Official
  27. About Amazon: Trainium3 UltraServer: faster AI training at lower cost (Dec 2025)Official
  28. TrendForce (press): Trainium3 reportedly built with TSMC's 3nm (4 Dec 2024)Press
  29. CommonWealth Magazine (press): Alchip and Amazon's AI chip supply chainPress
  30. Marvell: Marvell expands strategic collaboration with AWSOfficial
  31. AWS What's New: Inf2 instances in more regions, including Mumbai (13 Dec 2023)Official
  32. AWS: Amazon EC2 Capacity Blocks for ML pricingOfficial
  33. About Amazon: Amazon to invest $15 billion in Northern Indiana data centers (24 Nov 2025)Official
  34. Semafor (press): AWS's multi-state AI data center powering Anthropic's Claude (29 Oct 2025)Press
  35. About Amazon: AWS re:Invent 2025 announcementsOfficial
  36. AWS Neuron documentation: Trainium and Inferentia2 architectureOfficial
  37. AWS News Blog: Amazon EC2 Trn2 instances and Trn2 UltraServers now availableOfficial
  38. AWS Neuron documentation: Trainium2 architectureOfficial
  39. AWS Neuron documentation: Trainium3 architectureOfficial
  40. AWS What's New: Announcing AWS Neuron 2.27 (22 Dec 2025)Official
  41. AWS News Blog: Amazon EC2 Trn1 instances are now availableOfficial
  42. AWS: Amazon EC2 Trn1 instances pageOfficial
  43. AWS: Amazon EC2 Inf2 instances pageOfficial
  44. AWS: Amazon EC2 Trn2 instances pageOfficial
  45. AWS: Amazon EC2 Trn3 instances pageOfficial
  46. Anthropic: Powering the next generation of AI development with AWS (Nov 2024)Official
  47. AWS Neuron documentation: What's new in NeuronOfficial
  48. About Amazon: Amazon invests additional $4 billion in Anthropic (Nov 2024)Official
  49. GeekWire (press): Apple makes surprise cameo at AWS re:Invent (Dec 2024)Press
  50. AWS Machine Learning Blog: Scaling Rufus with over 80,000 Inferentia and Trainium chips for Prime Day (10 Oct 2024)Official
  51. Anthropic: Expanding our use of Google Cloud TPUs and services (23 Oct 2025)Official
  52. Amazon: Form 10-Q, quarter ended 31 March 2026Filing
  53. Amazon: Form 10-K, fiscal year 2025 (SEC)Filing
  54. Financial World (press, reporting Business Insider): Startups say Amazon's Trainium chips trail NVIDIA in real workloadsPress
  55. About Amazon: Amazon's investment in India (25 Jun 2026)Official
  56. About Amazon India: AWS data centre in Hyderabad (15 Jul 2026)Official
  57. Amazon Press Center India: AWS to invest INR 1,05,600 crore (US$12.7 billion) in India by 2030 (18 May 2023)Official
  58. CRN Asia (press): AWS partners with empanelled SHI India under IndiaAI Mission (Apr 2026)Press

On this page

1. At a glance2. Company card3. The story4. The lineup5. How it’s made6. Where it’s made: factories and addresses7. How it works8. Specs explained9. Official pricing10. The software lock-in11. Who buys it and why12. Hidden in plain sight13. Weak spots14. India angle15. What’s next16. Sources

Quick facts

HeadquartersSeattle, Washington
Founded1994
CEOAndy Jassy
Makes its own chips?Designs them (Annapurna Labs)
Main chip factoryNot disclosed by Amazon
Latest quarter AWS sales$42.2 billion

AI Chips

All chip brandsAMDAppleCerebrasGoogle TPUGroqHuawei AscendIntelMeta MTIAMicrosoft MaiaNVIDIAAI memory (HBM) and why it is scarceExport rules and the US–China chip contestHow a chip is made: sand to serverHow AI chips workIndia’s semiconductor pushHow to read a chip spec sheetToday’s AI news

In the news

Newest stories mentioning Amazon Trainium, straight from the companies’ own newsrooms and the wider press. Links open the original.

All of today’s AI news →

Ask an AI about this page

Opens your assistant with this page as the source, and a question rather than a summary. It will ask what you are building before it answers.

ChatGPTClaudeGeminiPerplexityGrok

Nothing is sent from here. The link carries only this page’s title and address.