Amazon is reportedly in negotiations to buy Decart AI for approximately US$7 billion, in a potential deal that could bring together the online retail and cloud computing giant with a developer of specialised AI infrastructure and real-time video models. The Wall Street Journal reported the discussions on Sunday, 11 October 2026, citing people familiar with the matter. Reuters subsequently reported the story but said it had been unable to verify the discussions independently.
The distinction is essential: there has been no publicly confirmed signed or completed US$7 billion acquisition in the sources reviewed for this report. Negotiations can be delayed, repriced or abandoned, and another buyer could emerge. Neither Amazon nor Decart had provided a confirmation in the Reuters reporting. Reuters, 11 October 2026; Wall Street Journal’s original report.
The essentials: what has actually been reported?
| Question | Current verified position |
|---|---|
| Potential buyer | Amazon.com, according to the Wall Street Journal and subsequent reporting. |
| Reported target | Decart AI, a San Francisco-based AI infrastructure and generative video startup founded by Israeli engineers in 2023. |
| Approximate reported price | US$7 billion. The proposed form of consideration, final price and transaction structure have not been publicly confirmed. |
| Status on 12 October 2026 | Reported talks, not a signed, approved or completed transaction. Reuters said it could not independently verify the negotiations. |
| Recent financing benchmark | Approximately US$4 billion valuation at a May 2026 US$300 million funding round, according to news reports citing private-market data. |
| Technology at stake | Decart Optimization Stack (DOS), Lucy real-time video editing and Oasis interactive world models. |
| Existing relationship | Decart is an AWS Trainium customer and showcases its optimisation work across AWS Trainium, Nvidia and Google TPU hardware. |
Sources: Reuters syndicated detail; Barron’s coverage; Decart’s own product descriptions; AWS’s customer case study.
Who is Decart AI — and why could it command a multibillion-dollar price?
Decart was founded in 2023 by Israeli engineers Dean and Orian Leitersdorf and Moshe Shalev, with operations centred in San Francisco. The company initially attracted attention by demonstrating real-time interactive video generation and has since positioned itself as an AI infrastructure business as well as a model developer.
That dual identity is important. Decart does not simply offer another chatbot competing for subscriptions. It develops software that makes sophisticated AI workloads run on specialised computing hardware and then builds consumer and enterprise applications on top of the same software stack. In May 2026 the company announced a US$300 million funding round led by Radical Ventures, and Nvidia joined its investors. News reports placed its valuation at around US$4 billion at that point. Decart has stated that the round brought its cumulative funding to more than US$450 million. Decart’s May 18 funding announcement; Ctech report on funding and valuation.
If the reported purchase price of about US$7 billion were realised, it would be approximately 75% above the roughly US$4 billion May valuation. That is only an illustrative comparison: a transaction’s actual equity consideration, cash, assumed liabilities, earn-outs, retention packages and valuation definition can differ materially from a startup funding-round valuation. The published reports do not establish a final premium or investment return for existing shareholders.
The strategically valuable asset: Decart Optimization Stack (DOS)
Decart’s Optimization Stack (DOS) is designed to extract more performance from AI accelerator hardware. Its public documentation describes optimisation across training, inference and the interaction between models and particular chips. In practical terms, the goal is to make AI software execute with lower latency, more throughput or improved cost efficiency by adapting workloads to the capabilities of the machines available.
This has become strategically significant because AI development is expensive and the hardware market is increasingly varied. Developers use Nvidia GPUs, Amazon’s Trainium accelerators, Google’s TPUs and other processors. Switching chips is not always a simple matter of moving the same code from one server to another: compilers, memory behaviour, model operators, communication frameworks and deployment tools can differ.
Decart says DOS supports multiple hardware systems including Nvidia, AWS Trainium and Google TPU, among others. That flexibility is potentially valuable for organisations trying to manage AI workloads across suppliers. However, its performance and portability claims are descriptions of the company’s technology and should not be interpreted as independently tested guarantees for every model or customer. Decart Optimization Stack details.
Why this could matter to Amazon Web Services
AWS competes to provide the infrastructure on which businesses train and run AI models. Amazon has invested in its own Trainium accelerator programme as an alternative to relying solely on third-party chips. In December 2025 it announced general availability of EC2 Trn3 UltraServers powered by Trainium3, targeting demanding model training and inference workloads. AWS Trainium3 general availability announcement.
AWS already identifies Decart as a Trainium customer. In a testimonial published by AWS, Decart co-founder Dean Leitersdorf said early testing yielded up to four times higher frame throughput, twice the cost efficiency and a reduction in latency from 40 milliseconds to 10 milliseconds relative to certain GPU comparisons. These are vendor-provided results from specific testing, not universal performance benchmarks independently validated by NEXTNEWS. AWS Trainium customer case study.
For Amazon, a closer relationship with software that adapts AI workloads to its accelerators could help strengthen the value proposition of Trainium and make high-performance video or agentic applications easier for AWS customers to adopt. This is a strategic interpretation, not a confirmed explanation from Amazon’s management of why it is negotiating a purchase.
There is also a business-model paradox: a tool advertised as working across rival accelerators may become less attractive to some customers if it is owned by one of the cloud providers competing for their workloads. Maintaining genuine hardware neutrality could therefore be commercially important if a deal proceeds. No post-acquisition strategy has been publicly confirmed.
Beyond chips: what are Lucy and Oasis?
Lucy: generative video as it happens
Lucy is Decart’s family of real-time video transformation models. Instead of waiting for an edited video to render offline, the system aims to alter objects, backgrounds or visual styles while a stream is running. Potential applications include live advertising, virtual try-on, e-commerce, streaming events, games and creative production.
Decart announced Lucy 2.5 in July 2026, describing more realistic editing, stronger frame-to-frame consistency and generation at 30 frames per second. The company’s commercial site also advertises particular latency and throughput characteristics. Those are product claims that can vary with hardware, workload and deployment conditions; they should not be mistaken for independent benchmarks or an indication that every output is factually authentic. Decart Lucy 2.5 launch announcement; Lucy product page.
Oasis: interactive world models for simulation
Oasis is Decart’s interactive world-model research and product line. A world model generates a sequence of evolving scenes in response to actions or inputs rather than merely producing a fixed image. Decart markets Oasis 3 for interactive simulation related to robotics, autonomous vehicles and other physical-AI applications. It announced the system in June 2026. Oasis product details; Decart’s research overview.
AI-generated simulations could help developers create varied training or test scenarios without having to reproduce every possible road, warehouse or physical environment in reality. But generated worlds can misrepresent physical conditions and rare hazards. A realistic-looking scene does not prove its underlying physics are accurate, and synthetic testing must complement rather than replace rigorous real-world safety evaluation for high-stakes systems.
Why an AI video company could be more useful than another chatbot
The acquisition reports arrive as technology giants seek to move artificial intelligence beyond text prompts. Real-time video, robotics, multimodal applications and low-latency automated agents place intense demands on computing infrastructure. Improvements to the underlying software can matter across many products rather than only one consumer-facing application.
Consider the potential economics of running millions of requests. If an AI system can process the same workload with better utilisation or less compute time, a provider may be able to serve more users on the same hardware. This does not automatically translate into lower consumer prices or higher profits: demand may grow, more powerful models may consume the savings, and integration costs can offset headline improvements. The value of a platform acquisition depends on how widely the optimisation is adopted and whether its benefits endure.
Amazon has multiple possible business uses for low-latency, interactive AI, from cloud services and digital advertising to retail product demonstrations and potentially entertainment workflows. Those are plausible applications, not announced post-acquisition product plans.
A complicated competitive picture: Nvidia, Anthropic and other chip makers
Decart’s May funding round included Nvidia, which may appear surprising for a company that promotes software flexibility across alternative AI processors. In fact, its stated approach is to optimise compute on multiple hardware platforms rather than to replace every GPU with another chip. Nvidia investment is evidence of a financial relationship; it does not establish that Nvidia endorses an Amazon takeover or that all its competitors will receive identical commercial support.
Another twist concerns Anthropic. Bloomberg reported that the Claude developer had explored buying Decart for about US$6 billion and had carried out due diligence before deciding in September not to proceed. This is previously reported background, not evidence that Anthropic currently has a competing offer on the table. The financial and strategic reasons for withdrawing have not been conclusively established in public. Bloomberg on Amazon talks and previous Anthropic discussions.
Amazon and Anthropic already work together extensively on AI infrastructure, making the reported history worth watching. Nevertheless, it is too early to conclude that any agreement with Decart would fundamentally change cloud-market share or relations among the companies.
Why a US$7 billion deal might be contentious
If discussions produce an agreement, questions will likely extend beyond the purchase price:
- Competition and access: Would independent developers continue to have equal access to software that optimises for rival chips and cloud platforms?
- Integration risk: Can Decart continue to innovate at startup speed within Amazon’s much larger organisation, while retaining key engineering staff?
- Valuation and return: Can the technology generate enough durable revenue, strategic savings or competitive advantage to justify the proposed consideration?
- Customer trust: Will businesses using competing clouds regard an Amazon-owned optimisation layer as neutral?
- Regulatory review: Any transaction may be subject to applicable merger and competition law; the existence, scope or outcome of a review cannot be presumed before a deal is announced.
- Video authenticity: Could easier live-video manipulation increase risks of impersonation, misleading advertising, deepfakes or copyright disputes?
These are analysis questions for a potential deal. They do not establish wrongdoing by Amazon, Decart, their investors or partners.
Why Australian businesses and developers should pay attention
For businesses in Australia using AWS or developing AI products, the most immediate relevance is not the reported acquisition price itself. It is the possibility of more efficient AI infrastructure and new real-time services becoming available through familiar cloud platforms.
Australian developers may want to compare the total cost of running models across GPU and Trainium environments, including software compatibility, engineering effort, availability, latency, data transfer and vendor lock-in. Headline performance gains should be evaluated against the exact workload and service-level requirements, not only manufacturer case studies.
Marketing, e-commerce and creative teams could eventually benefit from faster visual personalisation and live editing, but must consider permissions, intellectual property, disclosure requirements, privacy and the risk of creating misleading representations. The Australian Competition and Consumer Commission’s general consumer-protection rules remain relevant to representations made in advertising, including AI-produced content. ACCC guidance on misleading claims.
Investors should distinguish a rumoured transaction from a binding agreement and recognise that neither Amazon’s share price nor potential future AI revenues can be reliably forecast from the report. A possible acquisition is not a recommendation to buy or sell Amazon stock or any security.
What happens next? Five signals that would confirm progress
- Official announcement: Confirmation from Amazon or Decart that a definitive agreement has actually been signed.
- Deal terms: Disclosure of price, payment structure, assumed obligations, retention arrangements or other material conditions.
- Regulatory filings: Evidence of any required competition or other statutory approvals.
- Commercial commitments: Clarification of support for Nvidia GPUs, Google TPUs, Trainium and other systems after any acquisition.
- Transaction completion: A confirmed closing date, at which point ownership would actually transfer.
A reported discussion, a signed definitive agreement and a completed acquisition are three different stages. Even if a transaction is formally announced soon, it may not close immediately.
The NEXTNEWS view: the hidden infrastructure could be the bigger story
The roughly US$7 billion figure is attention-grabbing, but it may be the least instructive part of the report. Decart operates where two major trends intersect: more capable real-time AI applications and intense competition to make AI hardware economically viable. That combination could be strategically useful to a cloud provider spending heavily on its own AI processors.
For now, Amazon has not been confirmed as Decart’s new owner. What is established is Decart’s growing portfolio of optimisation and world-model technology, its previous funding, its relationship with AWS Trainium and the Wall Street Journal’s report that discussions are under way. The market will need an official agreement and more detail on access to multi-chip software before judging the significance of any deal.
Sources, methodology and disclaimers
Primary and independent sources: Wall Street Journal, 11 October 2026 — original acquisition report · Reuters — report and independent-verification caveat · Bloomberg — reported previous Anthropic talks · Decart Optimization Stack · Decart announcements and funding · Lucy technology · Oasis world models · AWS Trainium customer case study · AWS Trainium3 announcement · ACCC advertising guidance.
News status and source disclaimer: Prepared and published 12 October 2026 from publicly available reporting dated 11–12 October. The reported US$7 billion price refers to negotiations, not a completed acquisition, and Reuters said it could not independently confirm the original report. The parties had not confirmed a signed deal in the sources reviewed. Details can change rapidly; readers should consult official company announcements for later developments.
Technology and company-claims disclaimer: Vendor descriptions, throughput, latency, pricing and product capabilities are sourced from their respective providers, and may not generalise to all computing workloads. NEXTNEWS has not independently benchmarked Decart’s software, tested the company’s video models, or reviewed transaction documents. Potential business synergies are editorial interpretation, not disclosed acquisition plans.
Financial and legal disclaimer: This is general news and technology commentary, not personal financial, securities, investment, competition, legal or tax advice. No stock or crypto investment is recommended; investment decisions should take account of official disclosures and qualified independent advice. All financial amounts in this report are US dollars unless stated otherwise.
Image disclosure: The featured visual is an AI-generated picture created for NEXTNEWS. It illustrates AI infrastructure and generative video and does not depict Amazon or Decart employees, facilities, products or the reported transaction itself.
Disclaimer
NextNews strives for accurate news, but use it with caution—content changes often, external links may be iffy, and technical glitches happen. See the full disclaimer for details.