Amazon’s third-quarter earnings release, filed with the SEC on 29 October 2026, says total Bedrock token volume grew “nearly 17x year-over-year”, with Fortune 500 companies spending “nearly 7x more year-over-year” on AWS’s inference service.
On Alphabet’s call a day earlier, Sundar Pichai said Google’s first-party model APIs process nearly 32 billion tokens per minute, “more than four times what we processed a year ago”. The two figures measure different things; by our arithmetic, Amazon’s reported multiple is about four times Google’s.
A short Business Engineer explainer on the difference between an AI model and the harness around it. It teaches the general idea only and says nothing about any company in this piece.
The key insight: By our arithmetic, AgentCore customers deployed about two thirds of the last four quarters’ agents in the third quarter alone, while Amazon’s release puts Bedrock token growth at nearly 17 times in a year.
The release calls Bedrock “AWSโs inference service”. It lists the models added since the last earnings announcement: four from OpenAI (GPT-6 Astra, GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol), four from Anthropic (Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5), Moonshot AI’s Kimi K3, SpaceXAI’s Grok 4.7, Z.ai’s GLM 5.3, and “two purpose-built cybersecurity models from OpenAI”.
By our count, that is 13 models from five developers, six of them from OpenAI. Amazon also launched a public preview of Bedrock Managed Agents, which the release describes as “a managed service powered by OpenAI that makes it faster for enterprises to build and deploy production-ready AI agents”.
Year-on-year growth multiples for AI platform metrics, as each company reported them: Amazon (Bedrock token volume, Fortune 500 Bedrock spend) in its Q3 2026 release; Microsoft (Foundry revenue, multi-model agent customers, open-source modelrevenue) in its FY27 Q1 prepared remarks; Google (first-party API tokens) on its Q3 2026 call. Different metrics and scopes, not like-for-like.
The Agent Layer
AgentCore, AWS’s managed service for running agents, has “Tens of thousands of customers”, according to the release, “and they deployed 2x more agents in Q3 than the prior three quarters combined”.
By our arithmetic, that puts about two thirds of the agents deployed on AgentCore over the last four quarters in the third quarter alone. Our earlier piece on AgentCore covered how Amazon built it.
Amazon released Strands harness, which it says “lowers the cost of agent tasks by 32% through greater token efficiency than other harnesses when using the same frontier models”, and previewed Strands Box, “an open source sandbox that gives developers the ability to define rules for agent behavior and access”.
The release also says Amazon made AWS Continuum, an AI-powered security service, generally available and is bringing it “into Anthropicโs Claude Code, OpenAIโs Codex, and AWSโs Kiro”.
It acquired DuckLabs, the company behind the open source analysis tool DuckDB, and said it made Trainium easier to adopt with TorchNeuron, which “lets developers run existing PyTorch code (which most already use) on Trainium without rewriting a thing”.
Microsoft and Google Gave Their Own Multiples
In Microsoft’s prepared remarks on 28 October, Satya Nadella said: “We now have over 120,000 Foundry customers, with revenue more than tripling year over year”.
He added that the number of customers building agentic solutions with multiple models “increased 5X year over year” and that “revenue from open source models on Foundry increased 10X”, as covered in our piece on Microsoft’s Foundry.
Nadella also said Microsoft is “making the harness behind our own Copilot experiences available to customers in Foundry”. On Alphabet’s call, Google Cloud chief Thomas Kurian said more than 2,600 customers have used at least 100 billion tokens on Gemini Enterprise over the past 12 months, as covered in our piece on Google’s API tokens.
As we read the release, Amazon now reports its AI platform in usage multiples (tokens, Fortune 500 spend, agents deployed) rather than in revenue for Bedrock or AgentCore.
As we read the model list, Bedrock is sold as a neutral shelf: by our count 13 models from five developers were added, six of them from OpenAI, and OpenAI also powers the new Bedrock Managed Agents.
As we read Amazon’s release and Microsoft’s remarks side by side, both now sell an agent harness: Strands at Amazon, and the harness behind Copilot, offered in Foundry, at Microsoft.
Amazon Q3 2026 earnings release, 29 October 2026
“Tens of thousands of customers currently use AgentCore, and they deployed 2x more agents in Q3 than the prior three quarters combined”
Three Implications
FOR ENTERPRISE BUYERSAmazon says its Strands harness lowers the cost of agent tasks by 32% against other harnesses on the same frontier models; the figure is Amazon’s own.
FOR MODEL MAKERS The release names OpenAI, Anthropic, Moonshot AI, SpaceXAI and Z.ai models added to Bedrock since the last earnings announcement.
FOR MICROSOFT AND GOOGLE Nadella said Foundry has over 120,000 customers with revenue more than tripling; Pichai said Google’s API tokens are more than four times a year ago.
The framework’s core line: “The surface, not the model, governs autonomy.”
As we read it, Amazon’s quarter adds three surfaces at once: a harness (Strands), an open source sandbox for agent rules (Strands Box) and a managed agent service powered by OpenAI. The framework was published in July 2026 and does not discuss this quarter.
What Is Not Established
Amazon gives growth multiples, not totals: the release does not state Bedrock’s token count, its revenue, or how many agents AgentCore customers deployed.
Its 17x covers total Bedrock token volume, while Google’s figure covers its first-party model APIs, so the comparison is between different scopes. The 32% saving is Amazon’s own claim and is not independently tested.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Amazon’s release puts Bedrock token growth at nearly 17 times in a year and says AgentCore customers deployed twice as many agents in the third quarter as in the previous three combined, about two thirds of the four-quarter total by our arithmetic.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Every Amazon figure comes from its third-quarter 2026 earnings release, filed with the SEC on 29 October 2026 as Exhibit 99.1 to a Form 8-K. Microsoft figures come from Satya Nadella’s prepared remarks of 28 October 2026; Google figures from Alphabet’s Q3 2026 earnings call, as transcribed by us. Model counts and ratios are our arithmetic. Amazon’s call had not been read for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
