Amazon, 아마존은 AI 개발자들이 Trainium 칩 대신 Nvidia를 선택하는 이유를 해결하기 위해 Decart와 약 70억 달러 규모의 거래를 임박한 것으로 알려졌다.

Amazon is in active negotiations to acquire Decart, an Israeli AI startup, at a valuation of approximately $7 billion. If finalized, the deal would give Amazon the inference optimization software it needs to make its custom AI chips viable against Nvidia's dominant hardware ecosystem. SiliconAngle confirmed on October 11 that Amazon was in acquisition talks with Decart at that figure. The Wall Street Journal first reported the talks on Sunday, citing people familiar with the discussions, and said an agreement could be finalized within days.

The deal would be Amazon's largest pure AI software acquisition and one of the most expensive exits in Israeli technology history. It would also conclude a months-long bidding war in which Nvidia, SpaceX, and Anthropic each circled the three-year-old company before Amazon moved to the front. Amazon already had a commercial relationship with Decart through a $300 million strategic partnership formed in May 2026.

**What Decart Has That Amazon Needs**

Decart's primary product is not a language model or a consumer application. It is the Decart Optimization Stack, internally called DOS, a vertically integrated software platform that makes AI models run dramatically faster on existing hardware. Announced in May 2026, the Decart Optimization Stack combines hardware-aware model design, proprietary compilers, custom kernel tooling, and inference optimization into a single platform that runs across Nvidia GPUs, Google TPUs, and Amazon's Trainium chips.

DOS is strategically valuable because of what it automates. When an AI developer wants to run workloads on a non-Nvidia chip, including Trainium, it typically takes weeks or months of specialized engineering work to port models to that hardware efficiently. Nvidia's CUDA platform, built over 15 years, has accumulated the libraries, frameworks, and developer tooling that make Nvidia hardware the path of least resistance for most AI teams. AWS's competing Neuron SDK lags behind CUDA's maturity in ways that matter to enterprise customers. According to Decart, DOS compresses that porting process from months to weeks, removing the main barrier to Trainium enterprise adoption.

Amazon has confirmed one aspect of DOS's Trainium performance. The company stated that running Decart's Lucy2 model on Trainium3 chips achieved 80% Model FLOPS Utilization, a metric that measures how efficiently software exploits a chip's theoretical maximum performance. Chip engineers consider that figure excellent for any non-Nvidia hardware. Decart claims DOS delivers more than 1,600 tokens per second for agentic AI inference, compared with an industry average of approximately 200. The company also claims 8x speed and 100x cost efficiency gains over baseline systems on equivalent hardware. These figures are company-stated and have not been independently verified by third-party benchmarkers.

**Why Inference Is the Battleground Now**

The strategic value of DOS depends on where AI economics have shifted. Training an AI model, the resource-intensive process of teaching a neural network on vast datasets, is an occasional investment. Inference, the continuous process of serving that model to users, is where costs accumulate. Industry research indicates inference can account for 80% to 90% of an AI system's lifetime operating cost. The AI inference market is projected to grow from approximately $106 billion in 2025 to $255 billion by 2030.

Per-query inference costs have already fallen dramatically, from roughly $20 per million tokens in 2022 to approximately $0.07 per million tokens by 2025, driven largely by optimization advances rather than new hardware. Still, as AI products scale from thousands of users to millions, inference bills remain one of the largest line items in an AI product company's operating budget. Software that cuts those bills by a factor of eight, or by a more conservative three or four in production deployments, is not a peripheral tool. It is infrastructure for AI at scale and a source of competitive advantage for any company running AI at production volume.

Amazon Web Services faces this dynamic as both a cloud provider and an AI product developer. It runs other companies' inference workloads on its infrastructure and builds AI products of its own. Its position in cloud AI increasingly depends on offering customers better inference economics than rivals. Those economics depend on whether customers can use Trainium chips efficiently rather than defaulting to more expensive Nvidia hardware.

**Trainium's Problem and Decart's Proposed Solution**

Amazon has invested heavily in Trainium as a custom AI silicon play. Jeff Bezos has described chips as the next pillar alongside AWS, Prime, and Marketplace. Trainium3 reportedly delivers a 4x performance improvement and 40% better energy efficiency compared with prior generations. The hardware case for Trainium is strong. The adoption case has been harder to make.

The obstacle is the CUDA ecosystem. Nvidia has spent 15 years building the software environment, including compilers, libraries, debugging tools, and training frameworks, that makes running AI on its GPUs straightforward for any competent ML engineer. AWS's Neuron SDK, the Trainium equivalent, is improving but is widely acknowledged to still be catching up. AWS engineers reportedly recognize internally that CUDA familiarity keeps developers on Nvidia hardware even when Trainium's price-performance metrics are favorable.

DOS addresses this gap directly. Rather than requiring engineers to learn the Neuron SDK in depth, DOS automates the hardware-specific optimization work, including kernel selection, memory layout decisions, and compute scheduling, that previously required months of manual engineering. For an enterprise customer considering moving inference workloads from Nvidia to Trainium, DOS turns a major engineering project into a manageable transition.

That is the core logic of a $7 billion acquisition: Amazon would be buying the software key to unlocking hardware investment it has already made. If DOS became a first-party AWS tool, every Trainium customer would gain access to an optimization stack that currently requires engaging with Decart separately. If it works as described, the result could be a meaningful pull toward Trainium adoption across AWS's enterprise base.

The acquisition also carries a broader platform logic. DOS does not work only on Trainium; it runs on Nvidia GPUs and Google TPUs as well. An Amazon-owned DOS that continued serving the full hardware ecosystem would give AWS a software layer between AI developers and all major chip platforms. That is a more powerful position than any single-hardware play.

**The Bidders Decart Outlasted**

The $7 billion figure is 75% above Decart's most recent valuation of approximately $4 billion, set in a $300 million funding round five months ago. That round drew a notable group of strategic investors: Nvidia, which had reportedly also explored acquiring Decart outright; eBay Ventures; Adobe Ventures; and Toyota Ventures. Returning investors Sequoia Capital and Benchmark also participated, along with individual backers including Andrej Karpathy, the former Tesla AI director and OpenAI co-founder who has become one of the more visible technology investors in applied AI infrastructure.

The path to Amazon was circuitous. Nvidia reportedly explored acquiring Decart earlier in 2026 before a larger bid emerged. CTech in Israel first reported in August that Decart's founders were weighing a $6 billion to $7 billion sale, with SpaceX also described as circling. On August 13, Bloomberg reported that Anthropic, the Claude developer, had entered advanced acquisition talks at a reported $6 billion valuation. That deal would have embedded Decart's engineering team inside Anthropic's inference and performance organization.

On September 8, Bloomberg reported that Anthropic had completed due diligence and walked away without explanation. Neither company commented. Analysts at the time focused on timing. Anthropic was preparing for an initial public offering, with marketing expected to begin in mid-October, and absorbing a stock-based acquisition worth $6 billion to $7 billion just before going public would have added significant balance sheet complexity. Whatever Anthropic's reasoning, its withdrawal left Decart available, and Amazon, which already had a commercial relationship with the startup, moved to close the gap.

**Three Years, $450 Million, $7 Billion**

Decart was founded in 2023 by three engineers from Israel's Unit 8200 intelligence unit: brothers Dean and Orian Leitersdorf and Moshe Shalev. Dean Leitersdorf, who comple [Source text ends here.]