Amazon in Talks to Acquire Decart for $7B
Breaking News — October 11, 2026
Amazon is reportedly in advanced negotiations to acquire AI infrastructure startup Decart for approximately $7 billion, according to The Wall Street Journal. This follows Anthropic’s withdrawal from a prior $6 billion acquisition discussion after due diligence — highlighting Decart’s rapidly escalating strategic value in the AI hardware-software convergence race.
Why Decart? The Power of Hardware-Agnostic AI Optimization
At its core, Decart solves one of AI’s most pressing bottlenecks: chip fragmentation. As cloud providers and AI labs deploy diverse accelerators — NVIDIA GPUs, Google TPUs, and Amazon’s own Trainium — porting and optimizing large language models (LLMs) and multimodal systems across architectures remains time-consuming and costly.
Decart’s flagship product, the Decart Optimization Stack (DOS), is a software-first platform that:
✅ Abstracts hardware complexity: Enables seamless model deployment across NVIDIA, Google, and AWS chipsets — without rewriting kernels or retraining.
✅ Boosts training & inference efficiency: Dynamically optimizes memory layout, kernel fusion, and tensor scheduling per chip architecture.
✅ Delivers measurable gains: In AWS benchmark tests, Decart’s real-time video model achieved:
– 4× higher frame throughput on Trainium vs. comparable GPU setups
– 2× cost efficiency (lower $/inference)
– Latency reduced from 40ms → 10ms
DOS supports NVIDIA GPU, Google TPU, and Amazon Trainium — unifying training and inference optimization across vendor ecosystems.
Beyond Infrastructure: Real-Time AI Models in Action
Decart doesn’t just optimize — it builds production-grade AI systems that push latency and fidelity boundaries:
🎥 Lucy2.5 — Real-Time Video Editing
- Edit live 1080p@30fps video while playing: swap faces, change clothing, replace backgrounds, add/remove objects — all via text or reference image.
- Enables use cases like live-stream virtual try-ons, dynamic ad insertion, and interactive content creation.
Lucy2.5 real-time video editing: character transformation, virtual fitting, and background replacement (Source: Decart)
🚗 Oasis3 — AI-Powered Simulation for Autonomous Driving
- A photorealistic, physics-aware virtual environment where vehicles interact with dynamic scenes across six synchronized camera views.
- Functions as an infinitely scalable “digital test track” — enabling accelerated validation of perception, planning, and control stacks.
Oasis3 generates six concurrent road scenarios for multi-angle autonomous driving training (Source: Decart)
Strategic Implications: Why This Matters for AI Adoption
| Dimension | Impact |
|---|---|
| For Cloud Providers | Amazon gains a key differentiator: customers can adopt Trainium without massive porting effort — accelerating chip adoption and reducing lock-in risk. |
| For Developers | Faster iteration cycles, lower inference costs, and unified tooling across hardware — lowering the barrier to building performant AI applications. |
| For End Users | Smoother real-time experiences (e.g., instant video edits, responsive AI avatars), more affordable AI services, and broader accessibility of high-fidelity generative tools. |
As AI chip competition shifts from raw FLOPs to software-defined efficiency, Decart sits at the critical intersection — turning heterogeneous silicon into unified, scalable AI compute.
Timeline & Context
- May 2026: NVIDIA invests in Decart; valuation ~$4B
- September 2026: Anthropic abandons $6B acquisition after technical due diligence
- October 11, 2026: Amazon enters exclusive talks; reported $7B valuation
- Strategic Alignment: Decart already runs Lucy2 on AWS Trainium3; listed as Amazon’s strategic customer