Decart Achieves Real Time AI Video Generation With AWS Trainium3

Amazon Web Services (AWS) has scored another major win for its custom AWS Trainium accelerators after striking a deal with AI video startup Decart. The partnership will see Decart optimise its flagship Lucy model on AWS Trainium3 to support real-time video generation — highlighting the growing popularity of AI accelerators as a credible alternative to Nvidia's graphics processing units (GPUs).

Decart is essentially going all-in on AWS. As part of the deal, the company will also make its models available through the Amazon Bedrock platform, allowing developers to integrate Decart's real-time video generation capabilities into almost any cloud application — without worrying about underlying infrastructure.

💡 Key Takeaway: The distribution through Bedrock enhances AWS's plug-and-play capabilities, demonstrating Amazon's confidence in the surging demand for real-time AI video — while giving Decart broader reach within the global developer community.

AWS Trainium provides Lucy with the extra processing power needed to generate high-fidelity video without sacrificing quality or latency. Custom AI accelerators like Trainium are increasingly seen as a viable alternative to Nvidia's GPUs for AI workloads. While Nvidia still dominates the AI market — with its GPUs processing the vast majority of AI workloads — it faces a growing threat from custom silicon processors.

⚡ Why All the Fuss Over AI Accelerators?

AWS Trainium isn't the only option available to developers. Google's Tensor Processing Unit (TPU) product line and Meta's Training and Inference Accelerator (MTIA) chips are other notable examples of custom silicon. Each shares a similar architectural advantage over Nvidia's GPUs — their ASIC architecture (Application-Specific Integrated Circuit).

As the name suggests, ASIC hardware is engineered specifically to handle one kind of application — and to do so far more efficiently than general-purpose processors. Here's a useful way to think about the difference:

This yields massive performance and energy efficiency benefits compared to GPUs — and may explain their rapidly growing popularity across the AI industry.

📌 Notable Example: Anthropic has partnered with AWS on Project Rainier — an enormous cluster built from hundreds of thousands of Trainium2 processors. The project is set to deliver hundreds of exaflops of computing power to run Anthropic's most advanced AI models, including Claude Opus-4.5.

The AI coding startup Poolside is also using AWS Trainium2 to train its models, with plans to extend its use to inference workloads in the future. Meanwhile, Anthropic is hedging its bets — also exploring training future Claude models on a cluster of up to one million Google TPUs. Meta Platforms is reportedly collaborating with Broadcom to develop a custom AI processor for its Llama models, and OpenAI has similar plans in the pipeline.

💻 The Trainium Advantage

Decart chose AWS Trainium2 specifically for its performance, which enabled the company to achieve the ultra-low latency required by real-time video models. Lucy has a time-to-first-frame of just 40ms — meaning it begins generating video almost instantly after receiving a prompt. By streamlining video processing on Trainium, Lucy can also match the quality of much slower, more established video models like OpenAI's Sora 2 and Google's Veo-3, generating output at up to 30 fps.

Decart believes Lucy will only improve from here. As part of its agreement with AWS, the company has secured early access to the newly announced Trainium3 processor — capable of outputs of up to 100 fps with even lower latency.

💬 "Trainium3's next-generation architecture delivers higher throughput, lower latency, and greater memory efficiency – allowing us to achieve up to 4x faster frame generation at half the cost of GPUs."

— Dean Leitersdorf, Co-founder and CEO, Decart

🎯 What This Means for Nvidia and the Broader AI Chip Market

Nvidia might not be overly worried just yet. The AI chip giant is reportedly designing its own ASIC chips to rival cloud competitors. Moreover, ASICs are unlikely to replace GPUs entirely — each chip type has its own distinct strengths. The flexibility of GPUs means they remain the only real option for general-purpose models like GPT-5 and Gemini 3, and they continue to dominate in AI training.

However, many AI applications have stable, predictable processing requirements — making them particularly well-suited to running on ASICs. The rise of custom AI processors is expected to have a profound impact on the industry as a whole.

📷 Photo courtesy AWS re:Invent