Amazon
Last reviewed
Sources
38 citations
Review status
Source-backed
Revision
v4 ยท 6,271 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
38 citations
Review status
Source-backed
Revision
v4 ยท 6,271 words
Add missing citations, update stale details, or suggest a clearer explanation.
| Amazon.com, Inc. | |
|---|---|
| Type | Public |
| Industry | E-commerce, cloud computing, artificial intelligence, digital streaming, advertising, robotics |
| Founded | July 5, 1994 |
| Founder | Jeff Bezos |
| Headquarters | Seattle, Washington, United States |
| Key people | Andy Jassy (President and CEO) Jeff Bezos (Executive Chairman) Brian Olsavsky (Chief Financial Officer) Doug Herrington (CEO Worldwide Amazon Stores) Matt Garman (CEO Amazon Web Services) |
| Products | Amazon.com retail Amazon Web Services (AWS) Amazon Bedrock Amazon Q Alexa and Alexa+ Echo, Kindle, Fire devices Ring and Blink security cameras Prime Video, Amazon Music, Twitch Amazon Robotics Zoox robotaxi |
| Revenue | $213.4 billion (Q4 2025); $700+ billion annualized run rate (2026) |
| AWS revenue | $142 billion annualized run rate (Q4 2025) |
| Employees | ~1.55 million worldwide (2026) |
| Website | amazon.com |
Amazon.com, Inc. is an American multinational technology company headquartered in Seattle, Washington, and one of the most influential companies in artificial intelligence. Amazon's role in AI spans the entire stack: it designs custom AI accelerators (AWS Trainium and AWS Inferentia), operates the Amazon Bedrock foundation-model marketplace, builds its own large language models under the Amazon Nova program, ships consumer AI products such as Alexa and the Rufus shopping assistant, runs the world's largest mobile-robotics fleet, and is the largest single corporate backer of Anthropic, the maker of Claude, with cumulative investment commitments reaching up to $33 billion by April 2026.[4][5][32]
Founded by Jeff Bezos in 1994 as an online bookstore, Amazon grew into the world's largest e-commerce retailer and one of the most influential technology firms in artificial intelligence, cloud computing, digital streaming, advertising, logistics, and robotics. As of early 2026, Amazon is one of the five most valuable public companies in the world by market capitalization and is the largest cloud provider on the planet through its Amazon Web Services (AWS) subsidiary.[1]
Amazon's role in artificial intelligence spans nearly every layer of the modern AI stack. The company designs custom AI accelerators (AWS Trainium and AWS Inferentia), operates the Amazon Bedrock foundation-model marketplace, builds proprietary large language models under the Amazon Nova and Project Olympus programs, ships consumer AI products such as Alexa, Alexa+, the Rufus shopping assistant and the Amazon Q enterprise assistant, runs the largest mobile-robotics fleet in the world through Amazon Robotics, and operates an autonomous-vehicle subsidiary, Zoox, that began commercial robotaxi service in Las Vegas in September 2025.[2][3] Amazon is also the largest single corporate backer of Anthropic, with cumulative investment commitments that reached up to $33 billion by April 2026.[4][5]
The company is led by Andy Jassy, who succeeded founder Jeff Bezos as president and chief executive officer on July 5, 2021, exactly 27 years after Amazon was founded.[6] Jassy previously built and led AWS for 18 years, and his elevation to the top job has been widely interpreted as a sign that the company's center of gravity has shifted from retail to cloud, software, and AI.
Amazon participates in AI at four distinct layers, an unusually vertically integrated position. At the infrastructure layer it designs its own training and inference chips (Trainium and Inferentia) through the Annapurna Labs team and operates the largest data-center fleet in the industry. At the model layer it builds the Amazon Nova family of foundation models and bankrolls Anthropic's frontier Claude models. At the platform layer it sells access to dozens of models through Amazon Bedrock and developer tooling through Amazon Q and Amazon SageMaker. At the application layer it ships consumer and enterprise AI products, from Alexa+ and Rufus to warehouse robots governed by an in-house AI foundation model called DeepFleet.[33] Andy Jassy has summarized the strategy by arguing that owning the full stack yields durable cost advantages: "We are monetizing capacity as fast as we can install it," he told investors when defending Amazon's roughly $200 billion 2026 capital-expenditure plan, which he said spans AI, chips, robotics, and low-earth-orbit satellites.[34]
Amazon was founded on July 5, 1994 by Jeff Bezos in his garage in Bellevue, Washington. Bezos, then a 30-year-old former senior vice president at the New York hedge fund D. E. Shaw, drove cross-country with his wife MacKenzie to start the company after reading projections that web usage was growing 2,300% per year.[7] The company was initially called "Cadabra" but Bezos quickly renamed it after a lawyer misheard the name as "cadaver." He flipped through a dictionary and chose "Amazon" because it began with "A" (a help in alphabetical listings of the time) and because the Amazon River was the largest river in the world, matching his ambition to build the largest bookstore in the world.[7]
Amazon.com opened to the public on July 16, 1995. The first book sold was Douglas Hofstadter's Fluid Concepts and Creative Analogies: Computer Models of the Fundamental Mechanisms of Thought, an unintentionally fitting title given the company's later AI focus.[7] Within two months Amazon was selling to all 50 U.S. states and over 45 countries, with sales reaching $20,000 per week.
Amazon went public on May 15, 1997, at $18 per share, valuing the company at $429 million.[8] The IPO funded an aggressive expansion beyond books into music, video, electronics, toys, and tools. Bezos's now-famous 1997 letter to shareholders set out the principles that still guide Amazon: long-term thinking, obsessive customer focus, and a willingness to invest heavily in the future at the expense of short-term profits. The company survived the dot-com crash of 2000-2001 and posted its first quarterly profit in the fourth quarter of 2001.
Amazon's most consequential strategic decision was the creation of Amazon Web Services. Internally, the company had been struggling with the cost of building and maintaining its own server infrastructure, and a small team led by Andy Jassy began productizing the internal services that ran Amazon.com. AWS launched publicly in March 2006 with Simple Storage Service (S3) and Elastic Compute Cloud (EC2). What started as a way to rent out spare server capacity became, within a decade, the dominant cloud platform on Earth and the most profitable part of Amazon by a wide margin.
The cloud business gave Amazon several structural advantages that later proved decisive in the AI race. It built out the largest fleet of data centers in the industry, accumulated deep operational expertise in running GPU and accelerator clusters at scale, and gave the company direct relationships with millions of developers who would later build AI applications. By 2014 AWS was a multibillion-dollar business and the runaway leader in infrastructure as a service.
Amazon's consumer AI footprint began with the launch of the Kindle e-reader in November 2007 and accelerated dramatically with the introduction of the Echo smart speaker and its Alexa voice assistant in November 2014. Echo and Alexa were Amazon's bet that voice would become a primary computing interface, and for several years the device category dominated U.S. smart-speaker sales. Alexa shipped with cloud-based automatic speech recognition and natural language understanding, both built on AWS, and gradually grew a third-party "skills" ecosystem of more than 100,000 voice apps.
In 2016 AWS introduced Amazon Rekognition, a managed computer vision service for image and video analysis. Rekognition was used widely in retail, media, and security, but its sale to law-enforcement agencies generated significant controversy. The American Civil Liberties Union showed in 2018 that Rekognition incorrectly matched 28 members of the U.S. Congress to mugshots, with errors disproportionately concentrated among lawmakers of color.[9] After the murder of George Floyd in 2020, Amazon announced a one-year moratorium on police use of Rekognition for face recognition, which it later extended indefinitely.[10]
Amazon SageMaker launched at AWS re:Invent in November 2017 as a managed platform for building, training, and deploying machine learning models. SageMaker became one of the fastest-growing services in AWS history and underpinned Amazon's broader strategy of letting customers run any ML workload on AWS, regardless of framework.[11]
When Andy Jassy succeeded Jeff Bezos as CEO on July 5, 2021, Amazon was widely viewed as a laggard in the emerging generative-AI race that OpenAI, Microsoft, and Google were beginning to dominate. The release of ChatGPT in November 2022 made the gap explicit. Jassy's response over the next four years was to flood every layer of the AI stack with capital and to build a new partnership with Anthropic that, by 2026, had grown to rival Microsoft's investment in OpenAI in scope and ambition.
Key milestones included the launch of Amazon Bedrock in April 2023, the first $4 billion investment in Anthropic in September 2023, the unveiling of the Amazon Q enterprise assistant at AWS re:Invent in November 2023, the launch of the Amazon Nova family of foundation models at re:Invent in December 2024, the rollout of Alexa+ in February 2025, the launch of Trainium2 capacity in 2024 and Trainium3 in late 2025, the debut of the Amazon Nova 2 model family at re:Invent in December 2025, and the announcement in April 2026 of an additional Anthropic investment of up to $25 billion in exchange for a more than $100 billion, 10-year compute commitment from Anthropic to AWS.[4][12][32] Over the same period, Amazon's data-center capital expenditures climbed from roughly $48 billion in 2023 to $128 billion in 2025, with management guiding to approximately $200 billion in 2026, the vast majority directed at AI infrastructure.[13]
Amazon is organized into three reportable segments: North America, International, and Amazon Web Services. AWS is by far the most profitable, accounting for the majority of operating income despite contributing only around 17 to 20 percent of total revenue. The retail business operates Amazon.com, Whole Foods Market (acquired in 2017 for $13.7 billion), Amazon Fresh grocery, and a global logistics network. The company also owns Twitch, IMDb, Audible, MGM Studios, and Ring, among other subsidiaries.
| Executive | Role | Background |
|---|---|---|
| Andy Jassy | President and CEO | Former CEO of AWS (2003-2021); joined Amazon 1997 |
| Jeff Bezos | Executive Chairman | Founder; CEO 1994-2021 |
| Brian Olsavsky | Chief Financial Officer | Joined Amazon 2002 |
| Matt Garman | CEO, Amazon Web Services | Succeeded Adam Selipsky as AWS CEO in June 2024 |
| Doug Herrington | CEO, Worldwide Amazon Stores | Leads the retail business |
| Adam Selipsky | Former CEO of AWS (2021-2024) | Returned to AWS from Tableau |
| Dave Limp | Former SVP, Devices and Services | Led Echo, Alexa, Ring; departed in 2023 to lead Blue Origin |
| Panos Panay | SVP, Devices and Services | Joined from Microsoft in 2023, now leads Alexa hardware |
| Rohit Prasad | SVP and Head Scientist, AGI | Leads Amazon's foundation-model and AGI research |
Andy Jassy joined Amazon in 1997 right after earning his MBA from Harvard Business School and spent 18 months as Bezos's "shadow," essentially serving as a technical chief of staff before being asked to start what became AWS.[14] Under his leadership, AWS grew from 57 employees and an internal experiment to the largest cloud platform in the world. He has framed his agenda as CEO around "refounding" Amazon for AI, cutting bureaucratic layers, and reaccelerating innovation across both retail and cloud.
Amazon Web Services is the financial engine that funds Amazon's AI ambitions. AWS reported $35.6 billion of revenue in the fourth quarter of 2025, growing 24% year over year and the fastest growth rate in 13 quarters, with operating income of $12.5 billion at a 35% operating margin.[15] On an annualized basis, the cloud business reached a $142 billion run rate by the end of 2025, with an order backlog of $244 billion (up 40% year over year) reflecting multi-year commitments from large customers.
AWS holds roughly 28 to 32% of the global cloud-infrastructure market, ahead of Microsoft Azure (around 20 to 22%) and Google Cloud (around 12 to 13%), although competitors have grown faster on a percentage basis as enterprise AI workloads have shifted some demand toward the providers tied most closely to OpenAI and Google's foundation models.[16] Amazon's response has been to position AWS as the most neutral, multi-model platform in cloud AI, with first-class support for Anthropic, Meta Llama, Mistral, Cohere, DeepSeek, Stability AI, Qwen and even, since 2026, OpenAI models on Bedrock, alongside Amazon's own Nova and Olympus families.
| Service | Launched | Function |
|---|---|---|
| Amazon SageMaker | November 2017 | End-to-end managed ML platform for training and deploying models |
| Amazon Rekognition | 2016 | Managed image and video analysis |
| Amazon Lex | 2017 | Conversational chatbot framework, the engine behind Alexa |
| Amazon Polly | 2016 | Text-to-speech using deep learning voices |
| Amazon Transcribe | 2017 | Speech recognition for audio and video |
| Amazon Translate | 2017 | Neural machine translation across 75+ languages |
| Amazon CodeWhisperer | 2022 | AI code assistant; succeeded by Amazon Q Developer |
| Amazon Bedrock | April 2023 | Managed access to foundation models from multiple providers |
| Bedrock Marketplace | December 2024 | Catalog of 100+ specialized and enterprise foundation models |
| Bedrock Agents | 2023 | Tool-using AI agents over Bedrock models |
| Bedrock Guardrails | 2024 | Safety and content-filtering layer for foundation-model output |
| Amazon Q Developer | November 2023 | Generative AI assistant for software development on AWS |
| Amazon Q Business | 2024 | Enterprise assistant connected to internal data sources |
| Amazon Nova | December 2024 | Amazon's own family of foundation models |
| Amazon Nova 2 | December 2025 | Second-generation Nova models with extended reasoning |
Amazon Bedrock is the centerpiece of AWS's foundation-model strategy. Launched in general availability on April 13, 2023, Bedrock provides a unified API for accessing dozens of large language models, image models, and embedding models without the customer having to provision GPU clusters. Bedrock now hosts model families from Anthropic, Meta, Mistral, AI21 Labs, Cohere, DeepSeek, Luma AI, Qwen, Stability AI, TwelveLabs, Writer, and Amazon itself.[17]
Bedrock Marketplace, launched in December 2024, expanded the catalog to more than 100 specialized and emerging foundation models, including IBM Granite models, Nvidia Nemotron models, Upstage's Solar Pro for Korean, and Evolutionary Scale's ESM3 for protein research. Bedrock also includes Bedrock Agents (an agentic framework with tool use), Knowledge Bases (managed retrieval-augmented generation), and Bedrock Guardrails (content moderation and safety controls).[17]
Amazon Q is Amazon's family of generative-AI assistants for technical and business users. Amazon Q Developer (formerly CodeWhisperer) is positioned as a competitor to GitHub Copilot and Cursor; it integrates into the major IDEs, AWS console, and command line, and it can write, test, and refactor code, troubleshoot operational issues, scan for security vulnerabilities, and modernize legacy applications. Amazon Q Business is an enterprise search and chat assistant that respects the user's existing identity and permissions and connects to more than 50 enterprise systems including Salesforce, ServiceNow, Microsoft 365, Slack, Atlassian, and S3.[18]
Amazon Nova, unveiled at AWS re:Invent on December 3, 2024, is Amazon's first publicly named family of foundation models. Nova spans text, multimodal, image generation, and video generation, with Nova Micro, Lite, Pro, and Premier text models targeting different cost and latency tiers, plus Nova Canvas for image generation and Nova Reel for video.[35] Nova models are trained on AWS Trainium silicon and are available on Bedrock alongside third-party models. Internally, Amazon refers to its larger flagship model program as Project Olympus, a multi-trillion-parameter multimodal model intended to power complex consumer experiences such as Alexa+ and Rufus.[19]
At re:Invent on December 2, 2025, Amazon introduced Amazon Nova 2, the second generation of its foundation models, described as delivering reasoning capabilities with industry-leading price performance. The lineup includes Nova 2 Lite, Nova 2 Pro, a preview of the multimodal Nova 2 Omni, and Nova 2 Sonic for speech-to-speech interaction. The Nova 2 models support extended thinking with step-by-step reasoning and task decomposition across three selectable "thinking intensity" levels (low, medium, and high).[36] Alongside Nova 2, Amazon launched Nova Forge, a customization service that lets enterprises build tailored models, priced from $100,000 per year.[36]
One of Amazon's most strategic AI bets has been on its own custom silicon. The Annapurna Labs team, which Amazon acquired in 2015, designs three families of AWS chips: Graviton (general-purpose Arm CPUs), AWS Inferentia (inference accelerators), and AWS Trainium (training accelerators). The combined Graviton plus Trainium business reached an annualized run rate of more than $10 billion by the end of 2025 and was growing at triple-digit percentages year over year.[15]
Inferentia launched in late 2019 as a low-cost inference accelerator. Trainium followed in late 2022 and was followed by Trainium2 in late 2024 (used at scale by Anthropic for training and serving Claude) and Trainium3 in late 2025. Trainium3 is the company's first 3-nanometer AI chip, built on TSMC's 3 nm process and delivering 2.52 petaflops of FP8 compute, 144 GB of HBM3e memory, and roughly 4.4x the throughput of Trainium2 at up to 4x better energy efficiency in EC2 Trn3 UltraServer form factors that scale to 144 chips (about 362 FP8 petaflops).[20] Trainium4 has already been previewed and is expected to ship in late 2026 or early 2027 with 3x FP8 performance over Trainium3 and support for Nvidia's NVLink Fusion, allowing Trainium and Nvidia GPUs to share a unified memory fabric in mixed clusters.[20]
Amazon's silicon strategy is closely tied to the Anthropic partnership: Anthropic has committed to running Claude on Trainium for the next decade, currently uses more than one million Trainium2 chips to train and serve Claude, and Anthropic engineers have helped optimize the Trainium software stack.[32] Apple, OpenAI, and other major customers are also reported to be using or evaluating Trainium for production inference and training workloads.[21]
Anthropic, the AI-safety lab founded in 2021 by former OpenAI researchers including Dario and Daniela Amodei, has become Amazon's most important AI partner. The relationship now combines a multi-billion-dollar equity investment, a primary cloud and training relationship on AWS, deep silicon co-design on Trainium, and the integration of Claude into Amazon's own consumer and enterprise products such as Alexa+ and Bedrock. By April 2026 the partnership had grown to a scale comparable to Microsoft's relationship with OpenAI and was widely seen on Wall Street as the basis for a meaningful re-rating of Amazon's AI position.[5] In announcing the April 2026 expansion, Amazon CEO Andy Jassy said: "Anthropic's commitment to run its large language models on AWS Trainium for the next decade reflects the progress we've made together," while Anthropic CEO Dario Amodei said the collaboration "will allow us to continue advancing AI research while delivering Claude to our customers."[32]
| Date | Event | Amount |
|---|---|---|
| September 2023 | Initial Amazon investment in Anthropic | Up to $4 billion ($1.25B initial) |
| September 2023 | AWS named Anthropic's primary cloud provider | n/a |
| November 2024 | Amazon doubles total investment to $8 billion | +$4 billion |
| November 2024 | AWS named Anthropic's primary training partner | n/a |
| 2024-2025 | Trainium2 capacity dedicated to Claude training | n/a |
| February 2025 | Claude integrated into Alexa+ via Bedrock | n/a |
| April 2026 | Amazon commits up to $25 billion more (over time) | $5B initial, +$20B milestones |
| April 2026 | Anthropic commits $100 billion+ of AWS spend over 10 years | $100 billion+ |
| April 2026 | Anthropic secures up to 5 GW of AWS capacity | 5 gigawatts |
The April 2026 expansion was structured around what The Information and the Financial Times described as a Microsoft-OpenAI style "compute-for-equity" loop: Amazon invests in Anthropic, Anthropic spends most of the money on AWS Trainium capacity, and Amazon books the resulting cloud revenue. Per Anthropic's announcement, Amazon invested $5 billion immediately (at Anthropic's then-current $380 billion valuation) with up to $20 billion more tied to commercial milestones, while Anthropic committed more than $100 billion to AWS technologies over the decade and secured up to 5 gigawatts of AWS capacity, with nearly 1 gigawatt of combined Trainium2 and Trainium3 capacity slated to come online by the end of 2026.[32] The arrangement is reported to mirror the structure of an even larger $50 billion AWS-OpenAI compute deal that Amazon announced just two months earlier.[4][5] Anthropic's momentum underpinned the deal: its annualized revenue run rate climbed from about $9 billion at the end of 2025 to more than $30 billion, and the number of enterprise customers spending at least $1 million per year crossed 1,000.[37]
Alexa launched alongside the original Echo speaker in November 2014 and remains Amazon's most widely used AI consumer product, with hundreds of millions of devices deployed across the Echo, Echo Show, Echo Dot, Fire TV, and Fire Tablet lines. The original Alexa stack used hand-built natural language understanding and a directed-dialogue "skills" model. By 2023 the architecture was visibly aging compared to ChatGPT-style assistants.
Alexa+, announced on February 26, 2025, is a complete rebuild of Alexa around generative AI. It uses a routing layer that selects between Amazon Nova models and Anthropic Claude models hosted on Amazon Bedrock, with Claude reportedly handling the majority of complex reasoning, agent, and tool-use tasks demonstrated at the launch.[22] Alexa+ ships at $19.99 per month and is included with Amazon Prime. It can hold extended natural conversations, make restaurant reservations, book Ubers, place grocery orders, manage smart-home routines, and operate as an agent across third-party services. Amazon began rolling Alexa+ out to U.S. Prime members in 2025, with more than one million people having access by June 2025, and the goal of reaching most compatible Echo households by mid-to-late 2025.[38]
Rufus is Amazon's generative-AI shopping assistant, embedded directly into the Amazon mobile app and the amazon.com homepage. Rufus is built on a custom large language model that Amazon trained on its own catalog, customer reviews, Q&A content, and the broader web. It uses retrieval-augmented generation to ground its answers in product data and a streaming architecture for low-latency responses. Rufus runs at massive scale on more than 80,000 Inferentia and Trainium chips, the largest production deployment of Amazon's own AI silicon, serving an average of 3 million tokens per minute while holding latency to the first response under one second for Prime Day shoppers, and served more than 250 million customers in 2025.[23] Amazon reported that Rufus monthly users rose 149% and interactions 210% year over year, that shoppers who use Rufus are about 60% more likely to complete a purchase in that session, and that the assistant was on pace to drive over $10 billion in incremental annualized sales.[37]
The underlying model is part of the Project Olympus program; press reports have described Olympus as a 2-trillion-parameter multimodal model intended to compete with GPT-4 class systems on Amazon's most demanding internal workloads.[19]
Amazon's Ring (acquired in 2018) and Blink home-security brands have become significant outlets for Amazon's computer vision and on-device AI work. The 2025 product line introduced "Retinal 4K Vision," a suite of AI-tuned image-processing features on the new Wired Doorbell Pro, Spotlight Cam Pro, Floodlight Cam Pro, and Outdoor Cam Pro, alongside Smart Video Descriptions (an LLM-powered captioning system that describes events in human-readable language), Familiar Faces (face recognition for known visitors enrolled by the homeowner), and a Search Party feature that crowdsources lost-pet recognition across nearby Ring cameras.[24] The face-recognition rollout was controversial; civil-liberties groups have raised concerns about Ring's relationships with law-enforcement agencies and the broader normalization of always-on neighborhood surveillance.
Amazon One is a palm-recognition biometric identification service launched in 2020. It uses near-infrared imaging of the palm and the underlying vein structure to generate a vector "palm signature" used for payments, venue access, and workplace authentication. The retail version was deployed in Whole Foods, Amazon Go, and select third-party stadiums and offices. In 2026 Amazon announced it would discontinue the retail version on June 3, 2026 due to limited customer adoption while continuing to offer Amazon One Enterprise for workplace authentication.[25]
Amazon operates the largest mobile-robotics fleet on the planet. The company crossed more than one million industrial robots across its global fulfillment network in 2025, and its long-term goal is to automate roughly 75% of its fulfillment operations, with projected savings of about $12.6 billion cumulatively from 2025 through 2027 (roughly 30 cents per item delivered).[26][33] The robotics organization, headquartered in North Reading, Massachusetts, traces its lineage to the 2012 acquisition of Kiva Systems.
At the one-million-robot milestone in July 2025, Amazon introduced DeepFleet, a generative AI foundation model that coordinates the movement of robots across its fulfillment network, much like a traffic-control system. Amazon says DeepFleet, which was built with internal logistics data and AWS tools such as Amazon SageMaker, improves robotic fleet travel time by about 10%, enabling faster deliveries at lower cost. Over the past decade, the number of packages Amazon ships per employee each year has climbed from roughly 175 to nearly 3,870, a measure of the productivity gains from automation.[33]
In March 2012 Amazon acquired Kiva Systems for $775 million, at the time its second-largest acquisition.[27] Kiva, founded in 2003 by Mick Mountz and co-founders Raffaello D'Andrea and Peter Wurman, made flat orange mobile robots that lifted entire shelving pods and carried them to human pickers. By the time Kiva robots were fully deployed across Amazon's network in 2014, the "click-to-ship" cycle time at fulfillment centers had dropped from 60-75 minutes to roughly 15 minutes, and operating costs in retrofitted facilities had fallen by approximately 20%.[27] Amazon rebranded the unit as Amazon Robotics in 2015 and stopped selling the technology to third parties, an event that catalyzed the broader warehouse-robotics industry as competitors rushed to fill the gap.
| Robot | Launched | Function |
|---|---|---|
| Kiva (Hercules, Pegasus, Xanthus) | 2012-2018 | Mobile pod-lifting robots; the original Kiva platform |
| Robin | 2021 | Fixed robotic arm for sorting packages onto mobile drives |
| Cardinal | 2022 | Robotic arm using AI and computer vision to lift packages, read labels, and load GoCarts |
| Sparrow | 2022 | First Amazon arm capable of detecting and handling individual items in inventory |
| Proteus | 2022 | First fully autonomous mobile robot allowed to operate around humans without restricted zones |
| Sequoia | 2023 | Multi-level automated storage and retrieval system; identifies and stores inventory up to 75% faster |
| Digit (Agility Robotics) | 2023 | Bipedal humanoid tested for tote handling at fulfillment centers |
| Vulcan | 2025 | First Amazon robot with a sense of touch; uses force-feedback gripping to stow ~75% of inventory |
| Blue Jay | 2025 | Ceiling-mounted picking system coordinated by the Project Eluna AI |
The newest generation of Amazon robots is increasingly governed by AI "orchestrators" rather than rigid scripted workflows. Project Eluna is Amazon's internal name for an AI decision-support model that coordinates the activity of warehouse robots and human workers in real time, and Vulcan, unveiled in May 2025, is the first Amazon robot with tactile force feedback. Vulcan combines stereo vision, end-of-arm tooling that resembles a ruler stuck onto a hair straightener, and physical-AI algorithms to stow approximately 75% of items in a typical Amazon inventory pod at speeds comparable to human workers. Vulcan was deployed in fulfillment centers in Spokane, Washington and Hamburg, Germany and had processed more than 500,000 orders by the time of its public unveiling.[28]
Amazon's most automated facility, opened in Shreveport, Louisiana in 2024, uses eight different robot models in concert (Sequoia, Robin, Cardinal, Sparrow, Proteus, Hercules, Titan, and an automated packager) and reduced fulfillment processing times by up to 25%; Amazon plans to replicate the Shreveport design across roughly 40 facilities by the end of 2027.[26][33]
Astro, launched in September 2021, is Amazon's home robot. It is roughly the size of a small dog, has a 10-inch touchscreen, navigates autonomously around the home, and can be used for home monitoring, video calls, and as a mobile platform for Alexa. Astro Business launched in 2022 for small-business security applications. Astro for consumers was discontinued in mid-2024 in favor of further development on the Astro for Business platform and a planned next-generation home robot.
Zoox is Amazon's autonomous-vehicle subsidiary, acquired in June 2020 for approximately $1.3 billion.[29] Unlike Waymo and Tesla, which retrofit conventional cars, Zoox builds a purpose-built bidirectional electric vehicle with no steering wheel, no pedals, and two rows of facing seats. The vehicle has been described as a "toaster on wheels" because of its rectangular shape.[30]
On September 10, 2025 Zoox launched its first commercial robotaxi service on the Las Vegas Strip, operating fully driverless rides between Top Golf, Area15, Resorts World Las Vegas, New York-New York Hotel and Casino, and Luxor Hotel and Casino, making it the first company to offer fully autonomous ride-hailing in a purpose-built robotaxi.[30] Rides launched free of charge and carry up to four passengers; Zoox has said the free pricing will be phased out once it receives the necessary regulatory approval, and the company opened a waitlist for San Francisco. The Las Vegas launch marked Amazon's official entry into the U.S. robotaxi market five years after the Zoox acquisition.
Amazon is one of the largest capital spenders on Earth. The company's purchases of property and equipment grew from approximately $48 billion in 2023 to $77.7 billion in 2024 and to $128.3 billion in 2025, a 65% year-over-year increase driven primarily by spending on AI data centers, custom silicon, networking, and power.[13] In its Q4 2025 earnings, Amazon guided to approximately $200 billion of capital expenditures for fiscal 2026, predominantly for AWS and AI infrastructure, an additional step-up from 2025.[15][13]
Management has framed this spending as a long-term investment to capture demand from large customers like Anthropic and OpenAI for AI training and inference capacity. AWS reported an order backlog of $244 billion as of Q4 2025, up 40% year over year, suggesting that much of the planned 2026 buildout is already pre-sold. Andy Jassy has publicly defended the spend, telling investors that Amazon is "monetizing capacity as fast as we can install it" and that the demand is "predominantly in AWS."[34] He has also argued that the long-term decline in inference cost will drive a 10x or larger expansion in AI usage, and that Amazon's vertical integration (custom silicon, custom networking, owned data centers, and a large internal AI workload) gives it durable cost advantages over competitors.[31]
Amazon competes across multiple fronts in AI:
Cloud and foundation models: AWS competes with Microsoft Azure (closely tied to OpenAI) and Google Cloud (running Gemini). Microsoft and Google have both grown faster than AWS in percentage terms in recent quarters, in part because of their tighter integration with leading consumer AI brands. Amazon's bet is that a multi-model, neutral platform anchored by Anthropic, plus its own custom silicon and Nova models, will be more durable.[16]
AI assistants and consumer AI: Alexa+ competes with ChatGPT, Google Assistant and Gemini, Apple Intelligence, and Microsoft Copilot. Rufus competes with shopping experiences in Google's AI Overviews and ChatGPT Search.
Enterprise AI: Amazon Q competes with Microsoft Copilot for Microsoft 365 and GitHub Copilot, with Google Workspace AI, and with focused tools such as Cursor and Glean.
AI silicon: AWS Trainium and Inferentia compete with Nvidia GPUs, AMD Instinct GPUs, Google TPUs, and Microsoft's Maia chips.
Robotics and autonomy: Amazon Robotics competes with companies such as Symbotic, AutoStore, Locus Robotics, and Boston Dynamics. Zoox competes with Waymo, Tesla, Cruise (acquired by GM and substantially scaled back), Pony.ai, and WeRide.
Amazon's expansion into AI has attracted significant regulatory and civil-liberties scrutiny. The Federal Trade Commission and several state attorneys general filed an antitrust suit against Amazon in 2023 alleging anticompetitive conduct in online retail; that case was active in 2026. In the AI domain, regulators in the United States and Europe are reviewing the structural implications of the Microsoft-OpenAI and Amazon-Anthropic compute-for-equity arrangements. The U.K. Competition and Markets Authority concluded an initial review of the Anthropic-Amazon relationship in 2024 and elected not to refer it for in-depth investigation, but the larger 2026 expansion is expected to attract renewed scrutiny. Civil-liberties advocates have continued to criticize Ring's police partnerships and Amazon Rekognition's broader role in surveillance. Amazon's accelerating warehouse automation has also drawn labor scrutiny: internal documents reported in 2025 indicated the company believes robotics could let it avoid hiring more than 160,000 workers by 2027 that it would otherwise need.[33]