Amazon

The hyperscaler behind AWS Bedrock, the Nova model family, and Nova Act browser agents.

Seattle, WA, USA · Founded 1994

Amazon AI: Nova 2 Models & Bedrock Platform (2026 Guide)

Amazon builds the Nova 2 model family and AWS Bedrock, the managed multi-model API used by enterprises. Founded 1994, $2.56T market cap, AGI team formed 2023.

Amazon is a publicly traded technology company headquartered in Seattle, founded in 1994, whose AWS division builds the Nova 2 model family, Bedrock managed-API platform, and Nova Act browser-agent service. Its AI infrastructure, formerly the AGI team under Rohit Prasad, now sits inside a unified division led by Peter DeSantis covering models, custom silicon, and quantum computing.

Amazon, founded in 1994 in Seattle by Jeff Bezos, builds the Nova 2 foundation model family and AWS Bedrock, a managed API for accessing Nova and third-party models. Its AI effort is led by Peter DeSantis's Global AI Infrastructure division, formed in 2026, which also runs Amazon's Trainium chip program.

Founded: 1994 · HQ: Seattle, WA, USA · Team: 1500000+ · CEO: Andy Jassy · Funding: Public (NASDAQ: AMZN); AI/data center capex self-funded from operating cash flow, reported near $100B+ in 2025 alone · Valuation: $2.56T market cap (June 2026)

About Amazon

Amazon was founded in July 1994 by Jeff Bezos in Bellevue, Washington, initially as an online bookstore before expanding into e-commerce, cloud infrastructure, devices, and now foundation models. The company is incorporated as Amazon.com, Inc., a Delaware C-Corp, and is headquartered at 410 Terry Avenue North, Seattle, Washington. Andy Jassy became CEO in July 2021, succeeding Bezos, after previously building and running Amazon Web Services from its 2006 launch. Amazon's foundation model effort traces to the 2023 formation of the Artificial General Intelligence (AGI) organization under Rohit Prasad, who had led Alexa science; Prasad departed at the end of 2025 after 12 years, and in early 2026 Andy Jassy folded the Nova model team, custom silicon (Graviton, Trainium), and quantum computing research into a single Global AI Infrastructure division led by Peter DeSantis, formerly SVP of AWS Utility Computing. Amazon's primary AI product line runs through Amazon Bedrock, the managed service for accessing foundation models (Amazon's own Nova family plus third-party models from Anthropic, Meta, Mistral AI, and others) via a single API. The Nova family, first announced at AWS re:Invent in December 2024, was expanded into Nova 2 at re:Invent in December 2025: Nova 2 Lite (fast, cost-efficient reasoning with a 1M-token context window and built-in code interpreter and web grounding), Nova 2 Pro (the flagship for complex multi-step reasoning and long-range agentic planning, in preview for Nova Forge customers as of mid-2026), Nova 2 Sonic (a unified speech-to-speech model for real-time conversational AI), and Nova 2 Omni (multimodal text, image, video, and speech ingestion with text and image generation). Alongside the models, AWS launched Nova Forge, a service letting enterprises pre-train custom Nova variants on proprietary data via SageMaker AI, and Nova Act, an agent-building service for browser-based UI automation that AWS reports hits 90% reliability on early customer workflows. In the six months to mid-2026, AWS expanded Amazon Bedrock AgentCore with a Policy capability that blocks unauthorized agent actions in milliseconds, tightened Nova Act's native integration with AgentCore for deployment and CloudWatch for observability, and brought the AWS European Sovereign Cloud to general availability in January 2026, an EU-only region physically and logically separate from all other AWS regions, aimed at data-sovereignty-sensitive customers. Matt Garman, AWS CEO since June 2024 and a 19-year Amazon veteran who started as an intern, has framed 2026 as the year task-completing agents (not just content generation) start returning measurable enterprise value, citing Nova 2 alongside custom Trainium silicon as the AWS differentiator on price-performance. Amazon does not disclose external funding rounds: it is a publicly traded company (NASDAQ: AMZN) that self-funds AI infrastructure from operating cash flow. Reported 2025 capital expenditure guidance for AI and data center buildout exceeded $100B, with some 2026 estimates citing a combined roughly $200B AI infrastructure commitment across chips, data centers, and model training. AWS itself generated $37.59B in Q1 2026 revenue, up 28% year over year, and now accounts for over one-fifth of Amazon's total revenue; Amazon's total 2025 revenue was approximately $716.4B with $77.67B net income, and its market capitalization stood near $2.56 trillion as of June 2026. Amazon monetizes AI primarily through Bedrock's per-token API pricing for hosted models (including Nova), Nova Forge's custom-training service fees, SageMaker compute consumption, and Nova Act's agent-automation pricing, layered on top of AWS's existing EC2/Trainium compute margins. Enterprise customers pay standard AWS consumption-based billing rather than a separate AI subscription, which lets Amazon bundle inference against existing cloud commitments and volume discounts. Andy Jassy (CEO, Amazon.com) sits above Matt Garman (CEO, AWS) and Peter DeSantis (SVP, Global AI Infrastructure, overseeing Nova models, Graviton/Trainium silicon, and quantum computing). Swami Sivasubramanian, previously VP of AI/ML at AWS and an early architect of Bedrock and SageMaker, remains a senior AI leader inside the org. Amazon does not disclose a Nova-specific headcount, but AWS overall is one of Amazon's largest hiring segments, with the company citing thousands of new Gen Z graduate and intern hires in 2026 specifically to staff AI and cloud teams. Amazon's stated AI research focus centers on price-performance efficiency (Nova models are explicitly marketed as beating comparable OpenAI and Google models on cost per token at similar quality), agentic reliability (Nova Act's browser-automation benchmarks), and enterprise trust controls (Bedrock Guardrails, AgentCore Policy, and the European Sovereign Cloud). Compared to research-first labs like Anthropic or Google DeepMind, Amazon publishes far fewer standalone safety papers; its safety posture is expressed operationally, through Bedrock's content filtering, agent policy enforcement, and compliance certification breadth, rather than through a public responsible-scaling framework equivalent to Anthropic's RSP. Amazon's competitive position rests on distribution: Bedrock is the default multi-model marketplace for existing AWS customers, giving Nova default reach that standalone model labs lack, but Nova models trail OpenAI's GPT-5-series and Anthropic's Claude on independent reasoning and coding benchmarks, and Amazon has no consumer-facing chat product comparable to ChatGPT or Claude.ai. AWS's primary competitors in cloud AI infrastructure are Microsoft Azure (bundling OpenAI models) and Google Cloud (bundling Gemini); Amazon's answer to both is model-agnostic hosting through Bedrock rather than betting exclusively on in-house models. As a US public company operating globally, Amazon is subject to the EU AI Act as both a general-purpose AI model provider (via Nova on Bedrock) and a deployer, and has been building EU-specific infrastructure (the European Sovereign Cloud) partly in anticipation of GDPR and AI Act data-residency obligations. AWS holds HIPAA eligibility with signed Business Associate Addendums, and its European Sovereign Cloud region carries SOC 2, C5 Type 1, and seven ISO certifications (27001, 27017, 27018, 27701, 22301, 20000-1, 9001) as of its January 2026 general availability, alongside AWS's global roster of over 140 compliance programs including FedRAMP and NIST 800-171. Looking into the second half of 2026, Amazon's near-term AI priorities center on scaling Nova 2 Pro out of preview, expanding Trainium3 chip deployment to reduce reliance on Nvidia GPUs for both training and inference, and pushing Nova Act and AgentCore toward production-grade reliability for enterprise agent fleets. There is no indication of a Nova-specific spinoff or IPO; the model business remains fully consolidated inside Amazon.com, Inc.

Mission

To be Earth's most customer-centric company, offering the broadest selection of foundation models and AI infrastructure through AWS Bedrock.

Products

Compliance

SOC 1, SOC 2, SOC 3, ISO 27001, ISO 27017, ISO 27018, ISO 27701, HIPAA-eligible, GDPR, FedRAMP Moderate, HITRUST CSF

Links

Website · GitHub · Twitter · LinkedIn · Blog · Docs

Frequently Asked Questions

What is Amazon and what do they build in AI?

Amazon is a publicly traded technology company founded in July 1994 by Jeff Bezos, headquartered in Seattle, Washington. Its cloud division, Amazon Web Services, builds Amazon Bedrock, a managed API that gives enterprise customers access to Amazon's own Nova model family alongside third-party models from Anthropic, Meta, and Mistral AI. The flagship Nova 2 lineup, launched at AWS re:Invent in December 2025, includes Nova 2 Lite for fast everyday reasoning with a 1M-token context window, Nova 2 Pro for complex multi-step agentic workloads, Nova 2 Sonic for real-time speech-to-speech conversation, and Nova 2 Omni for unified multimodal text, image, video, and speech processing. Amazon also offers Nova Forge, letting enterprises pre-train custom Nova variants on proprietary data, and Nova Act, a service for building browser-automation agents that AWS reports hits 90% reliability on early customer workflows. Users access these through the Bedrock console, API, or SageMaker AI. Amazon's AI position is built on distribution: Bedrock reaches AWS's existing enterprise customer base by default, rather than competing purely on model capability.

Who founded Amazon and who leads its AI efforts in 2026?

Jeff Bezos founded Amazon in Bellevue, Washington in July 1994 as an online bookstore, later serving as CEO until 2021 and remaining Executive Chairman. Andy Jassy became CEO in July 2021, having previously built and run Amazon Web Services from its 2006 launch. Amazon's dedicated AI model effort began in 2023 as the Artificial General Intelligence organization under Rohit Prasad, formerly Alexa's head scientist, who built twelve Nova-branded foundation models before departing at the end of 2025 after 12 years at the company. In early 2026, Jassy restructured AI leadership, folding the Nova model team, Graviton and Trainium custom silicon, and quantum computing research into a single Global AI Infrastructure division led by Peter DeSantis, previously SVP of AWS Utility Computing. Matt Garman, a 19-year Amazon veteran who started as an intern, has served as AWS CEO since June 2024 and oversees Bedrock and enterprise AI go-to-market. There has been no Nova-specific spinoff; the AI business remains fully inside Amazon.com, Inc.

How much funding has Amazon raised for AI?

Amazon does not raise external venture funding; it has traded publicly on NASDAQ under ticker AMZN since 1997 and self-funds AI infrastructure from operating cash flow. Reported AI and data center capital expenditure guidance for 2025 exceeded $100B, with some 2026 industry estimates citing a combined roughly $200B commitment across custom silicon, data centers, and model training. Amazon's total 2025 revenue was approximately $716.4B with $77.67B net income, giving it substantial internal cash flow to fund AI buildout without outside investors. AWS itself, the segment housing Bedrock and Nova, generated $37.59B in Q1 2026 revenue alone, up 28% year over year, and accounts for over one-fifth of Amazon's total revenue. Amazon's market capitalization stood near $2.56 trillion as of June 2026. Because Amazon is a diversified public company rather than an AI-only startup, there is no separate 'AI valuation' distinct from the overall market cap, and no lead investor or funding round applies.

What AI products does Amazon make?

Amazon's core AI product is Amazon Bedrock, a managed multi-model API letting developers call Nova models or third-party models (Anthropic, Meta, Mistral AI, and others) through one interface with unified billing on standard AWS consumption pricing. The Nova 2 model family, launched December 2025, spans four variants: Nova 2 Lite (cost-efficient reasoning, 1M-token context, built-in code interpreter and web grounding), Nova 2 Pro (the flagship for advanced multi-step reasoning and agentic planning, in preview for Nova Forge customers as of mid-2026), Nova 2 Sonic (unified speech-to-speech for real-time voice AI), and Nova 2 Omni (multimodal ingestion across text, image, video, and speech with text and image generation output). Nova Forge lets enterprises build custom Nova variants by pre-training on proprietary data via SageMaker AI. Nova Act, built on a custom Nova 2 Lite model, automates browser-based UI workflows and integrates natively with Bedrock AgentCore for deployment, CloudWatch for monitoring, and IAM for authentication. There is no Amazon consumer chat product comparable to ChatGPT or Claude.ai; all products target developers and enterprises.

Where is Amazon headquartered and how big is the team?

Amazon is headquartered at 410 Terry Avenue North, Seattle, Washington, USA, with a major secondary headquarters (HQ2) in Arlington, Virginia. The company reported 1,576,000 full-time and part-time employees globally as of December 31, 2025, making it one of the largest private employers in the world, though this spans all of Amazon's businesses (retail, logistics, AWS, devices), not AI specifically. Amazon does not disclose a Nova-specific headcount. AWS, the segment running Bedrock and Nova, is one of Amazon's largest ongoing hiring areas; AWS CEO Matt Garman said in mid-2026 that the company is hiring thousands of Gen Z graduates and interns specifically for AI and cloud roles. Major AWS AI engineering hubs include Seattle, East Palo Alto (former Amazon Science AGI offices), and Boston, alongside AWS's global network of regional offices supporting Bedrock customers.

What is Amazon's mission or research focus in AI?

Amazon frames its AI mission around being Earth's most customer-centric company, applied to AI as offering the broadest selection of foundation models and infrastructure through Bedrock rather than betting solely on in-house models. Concretely, Amazon's research priorities center on three areas: price-performance efficiency (Nova models are explicitly marketed as beating comparable OpenAI and Google models on cost per token at similar quality), agentic reliability (Nova Act's browser-automation benchmarks and Bedrock AgentCore's production agent infrastructure), and enterprise trust controls (Bedrock Guardrails and AgentCore Policy, which blocks unauthorized agent actions against policy in milliseconds). Unlike research-first labs such as Anthropic or Google DeepMind, Amazon publishes comparatively few standalone AI safety papers and has not published a public responsible-scaling framework equivalent to Anthropic's RSP; its safety posture is expressed operationally through Bedrock's content filtering and compliance certifications rather than published alignment research.

Is Amazon compliant with SOC 2, GDPR, HIPAA?

Yes. AWS infrastructure, which underlies Bedrock and Nova, holds SOC 1, SOC 2, and SOC 3 attestations, ISO 27001, 27017, 27018, and 27701 certifications, and HITRUST CSF certification, with over 140 total compliance programs including FedRAMP Moderate and NIST 800-171, documented at aws.amazon.com/compliance/. AWS signs Business Associate Addendums (BAAs) and designates over 100 services, including Bedrock, as HIPAA-eligible for healthcare workloads. For GDPR, Amazon offers Standard Contractual Clauses and, as of January 2026, the AWS European Sovereign Cloud, a general-availability EU-only region physically and logically separate from all other AWS regions, carrying SOC 2 and C5 Type 1 attestations plus seven ISO certifications specific to that region. Customers access compliance reports on demand via AWS Artifact in the management console. By default, Bedrock does not use customer inputs or outputs to train Amazon's underlying foundation models.

Who are Amazon's main AI competitors?

Amazon's closest AI infrastructure competitors are Microsoft Azure, which bundles OpenAI's GPT models as its flagship offering, and Google Cloud, which bundles Gemini through Vertex AI; both compete directly with Bedrock for enterprise cloud-AI spend. On raw model capability, Amazon's Nova 2 family trails OpenAI's GPT-5-series and Anthropic's Claude models on independent reasoning and coding benchmarks, positioning Nova instead as the price-performance option within Bedrock's model-agnostic marketplace, where customers can also run those very competitor models. Amazon has no consumer-facing chat product to compete with ChatGPT or Claude.ai, a category it has largely ceded. An emerging competitive pressure not present 12 months ago is the rise of specialized agent-infrastructure vendors competing with Bedrock AgentCore for production agent deployment and governance, as enterprises increasingly evaluate agent reliability and policy controls as a separate purchasing decision from model choice.