AWS re:Invent 2026 (30 November to 4 December): keynote schedule and what to watch

re:Invent opens 30 November with five keynotes. The schedule, what AWS has already confirmed, and what to settle before launches land.

Read time
18 min
Word count
2.7K
Sections
10
FAQs
8
Share
Two men review a notebook and checklist at a Las Vegas hotel lobby table, captioned AWS re:Invent watchlist
On this page · 10 sections
  1. re:Invent 2026 at a glance
  2. The five keynotes: schedule in PT, IST and GMT
  3. What AWS has confirmed before re:Invent
  4. What to watch at re:Invent 2026: expectations, labelled
  5. Pre-event checklist for AWS buyers
  6. Worked example: AgentCore web search for an agent served from Mumbai
  7. India, EU and US notes
  8. After re:Invent
  9. How eCorpIT can help
  10. References

Summary. AWS re:Invent 2026 runs from Monday 30 November to Friday 4 December in Las Vegas. AWS's keynotes page, last updated on 9 November, lists five keynotes: the Developer keynote on Monday 30 November at 7:30 pm PT; CEO Matt Garman on Tuesday 1 December at 8:30 am PT (10:00 pm IST, 16:30 GMT); Data & AI with Swami Sivasubramanian on Wednesday 2 December at 8:30 am; the Partner keynote at 3:00 pm that day; and Technical Innovations with Matt Wood and Chief Trainium Architect Ron Diamant on Thursday 3 December at 8:30 am. A full pass costs $2,499, with 10% off for 10 or more passes, and the keynote livestream is free. AWS launches services on stage, so nothing below counts as announced until it is. Already on record: Andy Jassy told Amazon's 29 October results call that Amazon intends to offer Trainium outside AWS late in 2027; the AgentCore Web Search API became generally available on 10 November at $7 per 1,000 queries, but not in Mumbai or Hyderabad; and AWS cost tools now amortise EC2 Capacity Blocks. Watch the Trainium roadmap, AgentCore parity in Indian Regions and any commitment pricing. Our verdict: hold large three-year commitments until after 4 December unless one expires first, and work through the checklist below.

About this article. For CTOs, cloud architects and FinOps leads in India, the UK and the US who buy or run AWS services. On 11 November 2026 we checked the schedule, prices and every pre-event fact against AWS's re:Invent pages, AWS What's New posts, AWS documentation, Amazon's results materials and the EU's official legal text. Anything AWS has not announced is labelled as an expectation. This is general information, not financial or legal advice; AWS can change times and prices, and no vendor paid for inclusion. Researched with AI assistance and reviewed by our editorial team.

re:Invent 2026 at a glance

Sessions run across five Las Vegas venues, with keynotes and the Expo at The Venetian, and Friday sessions end at 12:30 pm. AWS's home page says it expects more than 60,000 attendees and lists over 2,200 sessions, 70% of them interactive.

Item Detail
Dates Monday 30 November to Friday 4 December 2026; badge pickup from Sunday 29 November
Venues The Venetian, Caesars Forum, Caesars Palace, MGM Grand, Wynn and Encore
Expected attendance 60,000+ (AWS figure)
Sessions and speakers 2,200+ sessions, 70% interactive; 3,000+ speakers
Full pass $2,499; 10% off for 10 or more passes in one transaction
Early-bird pricing Ended 25 August 2026
Livestream Free registration; keynotes and over 50 sessions
Expo 420+ sponsors and booths at The Venetian; Monday evening reception, then Tuesday to Thursday, closing 4:00 pm Thursday
Security focus day Thursday 3 December, Wynn
re:Play Thursday 3 December, 6:30 to 10:30 pm, Allegiant Stadium

Pass prices and the speaker count come from AWS's pricing page, and venues, the Expo and re:Play from the agenda. For a team of ten, the bulk discount takes 10 × $2,499 = $24,990 down to $22,491.

The five keynotes: schedule in PT, IST and GMT

AWS's agenda page says its times are Pacific Daylight Time. Las Vegas left daylight saving time on Sunday 1 November 2026, so we read the listed times as local Pacific Standard Time (UTC-8) and converted on that basis: IST is 13.5 hours ahead and GMT is 8 hours ahead. AWS says times may change. Speakers and times below are from the keynotes page.

Keynote Speakers listed by AWS PT (Las Vegas) IST (India) GMT (UK)
Developer Clare Liguori, Eric Johnson, Marc Brooker, Neha Rungta Mon 30 Nov, 19:30 to 20:45 Tue 1 Dec, 09:00 to 10:15 Tue 1 Dec, 03:30 to 04:45
CEO Matt Garman, with Barbara Karuth-Zelle (Allianz) and Tony Xu (DoorDash) Tue 1 Dec, 08:30 to 10:30 Tue 1 Dec 22:00 to Wed 2 Dec 00:00 Tue 1 Dec, 16:30 to 18:30
Data & AI Swami Sivasubramanian, Andy Warfield, Hannes Mühleisen, with Delta Air Lines and Cox Automotive Wed 2 Dec, 08:30 to 10:00 Wed 2 Dec, 22:00 to 23:30 Wed 2 Dec, 16:30 to 18:00
Partner Ruba Borno, Matt Garman, Matt Yanchyshyn, with Presidio and Marriott International Wed 2 Dec, 15:00 to 16:00 Thu 3 Dec, 04:30 to 05:30 Wed 2 Dec 23:00 to Thu 3 Dec 00:00
Technical Innovations Matt Wood, Amy Herzog, Colm MacCárthaigh, Ron Diamant, Oliver Cameron (Odyssey) Thu 3 Dec, 08:30 to 10:00 Thu 3 Dec, 22:00 to 23:30 Thu 3 Dec, 16:30 to 18:00

AWS's short descriptions show where each keynote is aimed. In our own words, with our expectations labelled:

  • Developer keynote. Building apps quickly, agents that hold up in real use, and confidence in code that AI writes. With Marc Brooker (VP and Distinguished Engineer) and Neha Rungta (Director, Applied Science) listed, we expect agent reliability and code verification to feature, not only new tools.
  • CEO keynote. AWS says Garman will present new launches spanning AI infrastructure and agents. The keynotes page recalls that the 2025 launches, from chips to agents, were made on stage.
  • Data & AI keynote. Taking agents from demo to production: dependable data access, demonstrated value and clear limits. AWS lists Sivasubramanian as VP, Agentic AI and Emerging Technologies.
  • Partner keynote. Where partners should invest, and simpler co-selling and go-to-market. Matt Yanchyshyn is VP, Marketplace and Partner Services, so we expect Marketplace terms to come up.
  • Technical Innovations keynote. AWS infrastructure from silicon to storage, built for unpredictable agent workloads. With Ron Diamant (VP and Chief Trainium Architect) and CISO Amy Herzog listed, this is the keynote for chips and security engineering. The agenda page calls it the Technology Innovations keynote, at the same time.

For Indian teams the timing is awkward: the CEO, Data & AI and Technical Innovations keynotes all start at 10:00 pm IST, the Partner keynote at 4:30 am IST, and the Developer keynote at 9:00 am IST on Tuesday 1 December.

What AWS has confirmed before re:Invent

Each fact below is dated and published by Amazon or AWS. Read every on-stage launch against them.

Trainium: strong demand and an outside-AWS plan for late 2027

Amazon reported third-quarter results on 29 October and held its results call that afternoon. On 30 October it published Andy Jassy's call remarks on custom chips. He said the chips business is growing at triple-digit percentages year over year, that revenue commitments for Graviton and Trainium have each grown more than 20 times year over year, and that customers using Amazon's custom silicon saved more than $2 billion in the quarter. He added that short-term capacity on prior-generation Trainium instances now sells at a premium to first-half levels, that inference is well over half of AWS's AI revenue, and that Amazon intends to make Trainium available outside AWS late in 2027. Amazon has published no form, price or terms for that offer.

The current product is Trainium3. AWS's Trn3 page says a Trn3 UltraServer scales to 144 Trainium3 chips, with up to 362 MXFP8 PFLOPs, 20.7 TB of HBM3e and 706 TB/s of aggregate memory bandwidth; each chip carries 144 GB of HBM3e. AWS claims up to 4.4 times the performance and 4 times the performance per watt of Trn2 UltraServers; those are vendor figures. For on-premises buyers, AWS AI Factories already put Trn2 and Trn3 inside customer data centres. AWS says deployment takes about 3 to 6 months once the site is handed over, and pricing follows a joint assessment. Our guide to Trainium outside AWS in late 2027 covers that decision in depth.

Bedrock AgentCore: a Web Search API, with regional gaps

On 10 November AWS made the Web Search API on Amazon Bedrock AgentCore generally available. The InvokeWebSearch call returns ranked results from an Amazon-operated web index, with filters for domain and publication time, IAM authorisation, PrivateLink access and CloudTrail data events. It costs $7 per 1,000 queries, the same rate as Web Search through AgentCore Gateway, with no gateway charge, and runs in six Regions: N. Virginia, Ohio, Oregon, Sydney, Tokyo and Ireland. On 5 November AWS also made Web Answer generally available, a tool that returns a synthesised, cited answer in one call. It runs only in N. Virginia, and the AgentCore pricing page lists it at $10 per 1,000 tool calls.

The AgentCore Regions table matters for India. Mumbai supports the AgentCore harness, Runtime microVMs and Runtime Instances, Memory, Gateway, Identity, built-in tools, Observability, Policy, Evaluations and optimisation. It does not support AgentCore payments, AWS Agent Registry, the Web Search tool, the Web Search API or Web Answer. Hyderabad also lacks the harness, Runtime Instances and Memory. Our August comparison of agent web search APIs sets the AgentCore rate against other providers; it predates the standalone API and the wider Region list.

Cost visibility: Capacity Blocks amortisation

Since 6 November, AWS Data Exports for CUR 2.0 and Cost Explorer spread the upfront fee for EC2 Capacity Blocks for ML across each reservation. Before, the fee appeared as one charge on the purchase date. You can now see amortised cost per instance and the cost of unused reserved capacity, at no extra charge, in all Regions except GovCloud (US) and China. If you buy GPU capacity in blocks, this is the data to have before any re:Invent capacity announcement.

Agent isolation: Nitro Sandboxes

On 9 November AWS introduced Nitro Sandboxes: lightweight VMs managed by the Nitro Hypervisor that give each EKS pod, ECS task or lightweight VM its own hardware-isolated boundary, with running LLM-generated code as a named use. AWS says enabling them in EKS or ECS is a one-line change, that cryptographic attestation can gate the release of secrets, and that they cost nothing extra on most Nitro-based instances in all commercial Regions.

Model retirements already scheduled on Bedrock

Bedrock's Legacy lifecycle page lists 26 open-weight models that entered Legacy on 30 September 2026 and reach end of life on 30 March 2027. It also lists Claude Opus 4.1 ending on 8 January 2027 and Claude Sonnet 4.5 on 8 April 2027 in commercial Regions (27 April 2027 in GovCloud). Our guides to the 26 open-weight retirements and to Claude on Bedrock cover the migrations. AWS has not said re:Invent will change these dates.

Regions: 34, two in India

AWS's Regions list shows 34 Regions for a standard AWS account; GovCloud and China use separate account types. India has two: Asia Pacific (Mumbai), ap-south-1, with three Availability Zones and enabled by default, and Asia Pacific (Hyderabad), ap-south-2, with three Availability Zones and opt-in.

The table restates the dated facts cited in this section.

Pre-event fact Date Published by Who should act
Trainium to be offered outside AWS late in 2027 30 Oct 2026 About Amazon, call remarks On-premises and sovereign AI buyers
Prior-generation Trainium short-term capacity sells at a premium 30 Oct 2026 About Amazon, call remarks Teams buying Trn1 or Trn2 capacity
Web Answer on AgentCore GA, N. Virginia only 5 Nov 2026 AWS What's New Agent builders
Capacity Blocks amortised in CUR 2.0 and Cost Explorer 6 Nov 2026 AWS What's New FinOps teams with GPU blocks
Nitro Sandboxes for EKS, ECS and EC2, no extra charge 9 Nov 2026 AWS What's New Teams running agent-generated code
AgentCore Web Search API GA, $7 per 1,000 queries, six Regions 10 Nov 2026 AWS What's New Agent builders, especially in India
26 open-weight models reach end of life on Bedrock 30 Mar 2027 Bedrock documentation Bedrock users on Llama, DeepSeek, Mistral and others

What to watch at re:Invent 2026: expectations, labelled

AWS has not announced any of the following. Each row is our expectation, based on the keynote descriptions, the speaker list or AWS's pre-event statements cited above.

Watch item Why we expect it What would change your plan
Trainium roadmap and Trn3 availability Ron Diamant on the Technical Innovations stage; the late-2027 statement Trn3 in Mumbai or Hyderabad, or published terms for outside-AWS Trainium
Trainium capacity pricing Prior-generation capacity sells at a premium New Savings Plans or Capacity Block terms for Trainium
AgentCore and Bedrock agent services Data & AI keynote framed on agents in production Web Search, payments or Agent Registry reaching Mumbai
Model choice on Bedrock 26 open-weight models end on 30 March 2027 New models in Mumbai that can replace them
AI code verification Developer keynote promises confidence in AI-written code Tools you can put in CI before merging agent code
Commitment and capacity pricing AWS's event page names token spend as a problem attendees bring Changes to Savings Plans, reservations or block pricing
India Region and Local Zone news Two Indian Regions, with Hyderabad behind Mumbai on AgentCore Feature parity for Hyderabad, or new India capacity
Security engineering CISO Amy Herzog on the Thursday keynote; Security focus day at Wynn Agent identity or isolation features that replace your own

Trainium and AI infrastructure

Expect Trainium content at the Technical Innovations keynote. What would change a buying decision is narrower: whether Trn3 reaches Regions your data must stay in, whether AWS publishes terms for outside-AWS Trainium, and whether prior-generation capacity pricing eases. Do not assume a launch date for any future chip; plan on what AWS publishes.

Bedrock and AgentCore

AgentCore shipped Web Answer and the Web Search API in the first ten days of November. For Indian teams the test is parity: the Web Search API, payments and Agent Registry are not in Mumbai today. If AWS extends them, agents that must keep data in India can drop cross-Region workarounds.

Pricing

AWS's re:Invent home page lists token spend among the problems attendees bring. AWS does not pre-announce price changes, so treat any expectation of cuts as speculation. The practical step is to know your commitment expiry dates and your Capacity Block utilisation before 1 December, so you can act on whatever is announced.

Pre-event checklist for AWS buyers

Settle these before 30 November. Each one turns a keynote launch into a decision you can make in days.

Capacity and Trainium

  1. List current GPU and Trainium usage by instance family, Region and commitment type.
  1. Turn on the amortised view in Cost Explorer and check Capacity Block utilisation for the last three months.
  1. Record your Neuron SDK version and which models already run on Trainium.
  1. Shortlist two workloads to benchmark on Trn3 if AWS expands availability.

Bedrock and AgentCore

  1. Inventory every Bedrock model in use against the 8 January, 30 March and 8 April 2027 end-of-life dates.
  1. Map each AgentCore feature you use, or plan to use, against the Regions table for Mumbai and Hyderabad.
  1. Estimate monthly web search volume and price it at $7 per 1,000 queries.
  1. Decide which agents may call services outside India and which may not.

Pricing and commitments

  1. List Savings Plans and Reserved Instances expiring before 31 March 2027.
  1. Avoid signing new three-year commitments before 4 December unless an expiry forces it.
  1. Prepare questions for your account team on Trainium capacity pricing and Capacity Block terms.

Regions and data residency

  1. Confirm which workloads must stay in ap-south-1 or ap-south-2 under your contracts and DPDP design.
  1. If Hyderabad is your disaster recovery target, check that the services you need run there and that the Region is enabled in your account.

Security

  1. Identify where agents run untrusted or generated code and how it is isolated today.
  1. Evaluate Nitro Sandboxes on one EKS or ECS workload before re:Invent.
  1. Confirm each agent has its own IAM identity and CloudTrail coverage.

Logistics

  1. Register for the free livestream and block the IST or GMT slots above.
  1. Assign one person per keynote to log launches against this checklist.

Worked example: AgentCore web search for an agent served from Mumbai

Take a hypothetical support agent for an Indian SaaS company, running in Mumbai. It makes 5,000 web searches a day to check product and regulatory updates.

  • Monthly queries: 5,000 × 30 = 150,000.
  • Cost: 150,000 ÷ 1,000 × $7 = $1,050 a month, or $12,600 a year.
  • In rupees, at Rs 93.93 per US dollar (ECB reference rates, 10 November 2026): $1,050 × 93.93 = about ₹98,627 a month.
  • At 50,000 searches a day the same arithmetic gives $10,500 a month, or $126,000 a year.

The price is only half the decision. The Web Search API is not available in Mumbai or Hyderabad. By distance from Mumbai, the closest Regions that offer it are Tokyo (about 6,700 km) and Ireland (about 7,600 km), so the agent would call a Region outside India. Search queries often contain user text, so strip personal data from queries before they leave India, or route through a provider and Region your data protection review has approved. If AWS adds Mumbai at re:Invent, rerun this check: the cost stays the same but the residency picture changes.

For GPU-heavy teams, the parallel exercise uses the new Capacity Block amortisation. If the amortised view shows, say, 30% of a block unused, that idle share is the first number to bring to any conversation about new capacity offers.

India, EU and US notes

India. Three keynotes start at 10:00 pm IST, so for most teams the free livestream, or a recording, is the practical option. Mumbai and Hyderabad are the only Indian Regions, and their AgentCore coverage differs, so map every agent call that leaves India. Watch for launches that list either Region on day one.

EU and UK. Three keynotes start at 16:30 GMT, a workable afternoon slot. Among European Regions, the AgentCore Web Search API runs only in Ireland. Teams planning a provider or Region move should note that from 12 January 2027, Article 29(1) of the EU Data Act bars providers from charging customers for switching, as our EU Data Act switching explainer sets out.

US. For in-person attendees, Thursday is the Security focus day at Wynn, and the Expo closes at 4:00 pm that day.

After re:Invent

We plan a recap after 4 December that maps each launch to the watchlist above, with availability status, Regions at launch and published prices, and we will refresh our Trainium, Bedrock retirement and AgentCore guides where announcements change them. For the Microsoft side of the season, our Microsoft Ignite 2026 watchlist covers Ignite in San Francisco from 17 to 20 November.

How eCorpIT can help

eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our cloud computing services team can map Bedrock and AgentCore features to Indian Regions, size Trainium and GPU capacity against real utilisation, and plan model migrations ahead of Bedrock retirement dates. Talk to us at /contact-us/.

Last updated: 11 November 2026.

References

  1. AWS re:Invent 2026, home page, updated 10 November 2026 — Amazon Web Services
  1. AWS re:Invent 2026 keynotes: speakers and schedule, updated 9 November 2026 — Amazon Web Services
  1. AWS re:Invent 2026 pricing — Amazon Web Services
  1. Q3 Earnings: Andy Jassy on why demand for Amazon's custom chips keeps growing, 30 October 2026 — About Amazon
  1. Amazon Bedrock AgentCore now offers a Web Search API, 10 November 2026 — AWS What's New
  1. AWS launches support for amortization of EC2 Capacity Blocks for ML, 6 November 2026 — AWS What's New
  1. AWS re:Invent 2026 agenda — Amazon Web Services
  1. Amazon.com announces third quarter results, 29 October 2026 — About Amazon
  1. Amazon EC2 Trn3 UltraServers — Amazon Web Services
  1. AWS AI Factories FAQs — Amazon Web Services
  1. Announcing Web Answer on Amazon Bedrock AgentCore, 5 November 2026 — AWS What's New
  1. Amazon Bedrock AgentCore pricing — Amazon Web Services
  1. Supported AWS Regions, Amazon Bedrock AgentCore Developer Guide — AWS Documentation
  1. Introducing AWS Nitro Sandboxes on Amazon EKS, ECS and EC2 for hardware isolated pods, tasks and VMs, 9 November 2026 — AWS What's New
  1. Model lifecycle (Legacy), Amazon Bedrock User Guide — AWS Documentation
  1. AWS Regions, AWS Regions and Availability Zones User Guide — AWS Documentation
  1. Regulation (EU) 2023/2854 (Data Act), Article 29 — EUR-Lex
  1. Microsoft Ignite, 17 to 20 November 2026, San Francisco — Microsoft

Frequently asked

Quick answers.

01 When is AWS re:Invent 2026?
AWS re:Invent 2026 runs from Monday 30 November to Friday 4 December 2026 in Las Vegas, across The Venetian, Caesars Forum, Caesars Palace, MGM Grand, and Wynn and Encore. Badge pickup opens on Sunday 29 November, sessions start on Monday, and Friday sessions end at 12:30 pm local time.
02 When are the re:Invent 2026 keynotes?
AWS lists five. The Developer keynote is on Monday 30 November at 7:30 pm PT, CEO Matt Garman on Tuesday 1 December at 8:30 am, Data & AI on Wednesday 2 December at 8:30 am, Partner that day at 3:00 pm, and Technical Innovations on Thursday 3 December at 8:30 am. AWS says times may change.
03 What time is Matt Garman's keynote in India?
The CEO keynote runs from 8:30 to 10:30 am Las Vegas time on Tuesday 1 December, which is 10:00 pm IST on Tuesday to midnight. In the UK it runs from 16:30 to 18:30 GMT. AWS's agenda labels its times as daylight time, but Las Vegas is on standard time in December, which we used.
04 How much does a re:Invent 2026 pass cost?
AWS lists the full conference pass at $2,499. Buying 10 or more passes in one transaction earns a 10% discount, subject to AWS terms, so ten passes cost $22,491. Early-bird pricing ended on 25 August. Registering for the livestream, which carries the keynotes and over 50 sessions, is free.
05 Will AWS announce Trainium news at re:Invent 2026?
Likely, but not confirmed. Chief Trainium Architect Ron Diamant speaks at the Technical Innovations keynote on 3 December, and Andy Jassy told Amazon's 29 October results call that Amazon intends to offer Trainium outside AWS late in 2027. AWS has published no terms for that offer, so treat anything further as unannounced.
06 Is Bedrock AgentCore available in the AWS Mumbai Region?
Yes, for most features. AWS's Regions table lists Mumbai for the AgentCore harness, Runtime, Memory, Gateway, Identity, built-in tools, Observability, Policy and Evaluations. Mumbai does not yet support AgentCore payments, AWS Agent Registry, the Web Search tool or API, or Web Answer. Hyderabad supports fewer features than Mumbai.
07 How much does the AgentCore Web Search API cost?
AWS prices it at $7 per 1,000 queries, the same rate as Web Search through AgentCore Gateway, with no gateway charge. An agent making 5,000 searches a day, or 150,000 a month, would cost $1,050 a month. The API runs in six AWS Regions in the US, Europe and Asia Pacific, none of them in India.
08 Should we sign AWS commitments before re:Invent?
If a Savings Plan or reservation expires before December, renew only what you must, for the shortest term that fits. Otherwise wait until after 4 December, because AWS announces services, and sometimes pricing, on stage. Use the new Capacity Block amortisation data to size any GPU or Trainium commitment against real utilisation.

About the author

Manu Shukla

Founder & Director

Founder of eCorpIT. Hands-on engineer leading senior-only delivery for AI apps, custom software, and cloud systems for global clients.

Subscribe

One engineering note a week. No fluff, no spam.

Senior-architect playbooks on AI agents, mobile apps, cloud, security, data, and marketing — delivered every Wednesday.

Past the reading

Read enough. Let's build something.

A senior architect responds in 24 working hours with scope, indicative cost, and a timeline. NDA before any technical conversation.

Loading search…

Browse all guides