Press reports place Amazon in the final stages of buying Israeli AI startup Decart for more than $7 billion, with earnouts that could reach about $7.5 billion. Calcalist reported on November 18, 2026 that legal hurdles cleared and signing could happen within days. Globes reported on November 19 that talks led by Amazon chip executive Nafea Bshara were nearing completion. Neither Amazon nor Decart has issued a definitive acquisition press release on the explainx.ai check date of November 21, 2026. This post separates reporter-sourced deal terms from verified public statements.
What is claimed today
| Claim | Source | Verified by Amazon or Decart? |
|---|---|---|
| Deal close to signing, could finalize within days of Nov 18 | Calcalist exclusive | No public closing post |
| Price more than $7B, up to ~$7.5B with performance payments | Calcalist | No 8-K or joint release |
| Cash, stock, deferred consideration; earlier $5B–$6B valuation band | Calcalist | No filing |
| Founders stay five years; large payment in year five; non-compete | Calcalist, Globes | No contract published |
| Talks led by Nafea Bshara (Annapurna Labs / Amazon chips) | Globes | No Amazon comment cited |
| Possible unveil at AWS re:Invent Nov 30 | Secondary November 21 summaries | No Amazon agenda confirmation |
Treat the table as journalism with unnamed or industry sources, not a term sheet. When Amazon buys a public company it files with the SEC; Decart is private, so confirmation may arrive only through a blog post, re:Invent keynote, or regulatory merger filing in relevant jurisdictions.
The deal timeline: Anthropic, Nvidia, Amazon
Calcalist reconstructs a multi-buyer auction that explainx.ai readers may remember from summer headlines.
-
August 10, 2026 — Calcalist first reported Decart was exploring a sale. Bloomberg coverage relayed by explainx.ai put Anthropic in talks near $6 billion for video and inference technology. Anthropic never confirmed a binding deal on the record.
-
Early September 2026 — Calcalist says Anthropic withdrew after drafts and diligence, scuttling a path some outlets framed as Anthropic’s Israel entry ahead of its IPO.
-
September–October 2026 — Decart reportedly approached Nvidia, an existing investor, but Calcalist says Nvidia did not reopen acquisition talks.
-
October–November 2026 — Amazon emerged as the active buyer. Calcalist’s November 18 piece says most legal issues resolved in the prior week after structural changes delayed the pact.
That sequence matters for builders because Decart’s stack is chip-portable inference and media, not a single closed model API. Whoever owns it decides whether Lucy, Oasis, and routing software stay multi-vendor or become AWS-exclusive accelerators.
What Decart builds (from primary company context and reporting)
Multi-chip compute racks illustrating Decart routing software across accelerator families
Calcalist and earlier Bloomberg-style summaries describe three layers relevant to practitioners:
Multi-chip routing. Decart software lets teams move workloads between accelerator families. That aligns with Amazon’s Trainium and Inferentia push documented across explainx.ai’s Trainium inference guides and chip architecture primer. If Amazon closes the deal, expect messaging that Decart squeezes more tokens per watt on Annapurna silicon — similar to how Luminal’s compiler story targets CUDA lock-in from the software side.
World models. Lucy and Oasis headline real-time video and interactive 3D simulation. They sit in the same embodied AI conversation as robotics and gaming stacks, though Calcalist notes revenue today is low tens of millions and semiconductor project-driven, not mass consumer subscription scale.
Capital and talent. Calcalist lists $400M+ raised, $4B May 2026 valuation, and backers including Sequoia, Benchmark, and Radical Ventures. Founders Dean Leitersdorf (CEO), Orian Leitersdorf (chief scientist), and Moshe Shalev (CPO) are Unit 8200 alumni in press accounts; Calcalist says former 8200 commander Yossi Sariel joined in February 2026.
None of those product names require the acquisition to be true — they describe the asset Amazon would be buying if reports are accurate.
Why Amazon would pay $7 billion-plus (strategic read)
Semiconductor wafer grid evoking Annapurna Labs and Trainium silicon strategy behind a reported Decart deal
Amazon’s largest recent tech-adjacent deals in Calcalist’s comparison set include MGM (~$6.5B, 2022) and the Globalstar (~$11B) satellite agreement announced in 2026. A $7B–$7.5B Decart price would sit among the largest AI infrastructure bets, not a content catalog.
The strategic logic reporters repeat:
-
Trainium 3 needs killer software, not only silicon. Calcalist says Amazon gave Decart early Trainium 3 access in 2025 after founder meetings.
-
Nvidia substitution is an AWS revenue story. Customers who can port inference layers without rewriting entire stacks reduce GPU rental dependence, echoing Anthropic multi-chip orchestration claims but under Amazon’s ownership.
-
Real-time video competes with generative media startups and could feed Bedrock, Nova, and gaming on AWS. That is speculative until Amazon demos integration.
Again, this section interprets reported motives. It is not confirmation the board signed.
What changes for builders if the deal closes
If Calcalist’s terms hold after closing:
| Area | Likely builder impact |
|---|---|
| Trainium inference | Decart routing may land inside SageMaker, Bedrock, or DLC images — watch re:Invent sessions and GitHub org changes. |
| Lucy / Oasis APIs | Early access could move from direct Decart contracts to AWS GA with IAM billing. |
| Multi-cloud portability | Founders’ five-year lockups suggest Amazon will prioritize AWS, but prior customers on other clouds may get transition periods — unspecified in press. |
| Pricing | No public price list change yet. Compare against Baseten and Together serverless when AWS publishes bundles. |
If the deal does not close, Decart remains an independent vendor selling chip optimization to labs and clouds, and the Anthropic chapter stays a case study in failed mega-acquisitions before IPO.
What is verified right now
Verified on November 21, 2026:
- Calcalist published the November 18 exclusive with named journalists and detailed consideration structure.
- Globes published November 19 with Nafea Bshara attribution to sources.
- explainx.ai’s August Anthropic-Decart post documents earlier $6B talk claims; no contradiction — buyer changed.
Not verified:
- A signed purchase agreement, regulatory approval, or employee communication from Amazon or Decart.
- Exact earnout metrics, breakup fees, or antitrust review timelines.
- Product roadmap integrations.
Search demand around “amazon decart acquisition” and “is decart ai publicly traded” reflects confusion between this startup and unrelated names — the company remains private until a buyer files or announces.
How to watch for confirmation
Check these surfaces in order:
- Amazon newsroom and AWS blog for acquisition or re:Invent partner posts.
- SEC EDGAR if Amazon issues an 8-K describing material acquisitions (amount thresholds apply).
- Israeli registrar filings if Calcalist’s tax and structure claims require local disclosure.
- Decart’s own site and GitHub for org transfers or license changes.
explainx.ai will treat a joint press release or Amazon 8-K as the line between rumor and closed.
Regulatory and competitive context
A $7B+ Israeli founder-led startup sale to a U.S. hyperscaler typically triggers multi-jurisdiction review, even when the target is private. Calcalist notes legal restructuring delayed the Amazon pact — consistent with tax treatment, founder earnouts, and employment lockups that Israeli founders negotiate on large exits. No filing explainx.ai retrieved on November 21 lists FTC or EU merger review numbers for this specific deal; reporters have not published HSR filing dates.
Competitively, Amazon already competes with Nvidia on silicon while partnering on cloud instances. Owning Decart’s chip-portability layer would mirror AMD buying Taalas for etched inference and Anthropic courting Decart for video — both paths say the same thing: software that moves workloads between accelerators is strategic property in 2026. If the deal fails, Nvidia’s refusal to re-enter talks (per Calcalist) leaves Decart independent but still aligned with Trainium early-access history.
For AWS customers, the near-term product question is whether Decart routing appears inside SageMaker HyperPod or Bedrock custom model import flows. Today’s verified surface is Trainium DLC tooling and global inference gateway — not Decart-branded panels.
Lucy, Oasis, and why video labs cared
GPU chiplets on a board, standing in for real-time video and inference stacks Decart optimizes
Lucy (real-time interactive video) and Oasis (3D world simulation) explain why Anthropic and Amazon both reportedly bid high despite modest revenue. Video stacks burn predictable tokens per frame; optimizing them across Trainium vs GPU changes margins on inference-heavy products. Anthropic’s August interest tied to efficient video + inference, not consumer social apps.
If Amazon closes, expect first integrations on re:Invent stages or Annapurna blog posts, not a separate Decart homepage forever. Founders’ five-year retention (Calcalist) implies Amazon wants roadmap continuity, not a quick talent acqui-hire shutdown — but retention clauses are reporter-sourced, not a contract readers can read.
Questions still open after November 21 reporting
| Open question | Why practitioners care |
|---|---|
| Signed or still negotiating? | Procurement cannot budget Decart bundles on AWS until confirmation |
| Earnout metrics | Tied to Trainium adoption or Lucy revenue — unknown |
| Multi-cloud customers | Will existing Decart contracts grandfather non-AWS routes? |
| Anthropic relationship | If Amazon owns Decart, Claude training partners lose a neutral optimizer vendor |
| Nvidia investor stake | Calcalist says Nvidia invested; divestiture terms on change of control unreported |
Searchers comparing “amazon decart acquisition” to “anthropic decart” should read both timelines: Anthropic talks died September 2026; Amazon talks active November 2026. Same asset, different buyer thesis — IPO adjacency versus cloud silicon substitution.
How to prepare before confirmation
- Map your inference spend by chip family (Nvidia vs Trainium vs other) using CUR or internal dashboards.
- List workloads that depend on third-party routing SDKs; flag any Decart pilots in vendor questionnaires.
- Watch re:Invent keynotes for Nafea Bshara or Annapurna Labs segments naming acquisition targets.
- Do not reprice long-term GPU commits on rumor alone — Spacex Rubin financing headlines show how fast uncertain deals move markets without changing your next sprint’s stack.
