Amazon nears Decart acquisition worth up to $7.5 billion | Magica
Amazon nears Decart deal worth up to $7.5 billion to strengthen its AI chips
Editorial Team
••📖7 min read
Amazon is in advanced talks to buy Decart, whose software helps AI models run efficiently across chips. The reported deal could strengthen AWS’s Trainium business, with conditional payments and five-year founder commitments making execution central to its value.
Amazon is nearing a Decart acquisition valued at about $7.5 billion, but negotiations could still collapse.
Decart could add chip-efficiency expertise to AWS, which already offers tools to ease migration from Nvidia hardware.
Reported terms tie part of the price to targets and keep the founders at Amazon for five years.
Amazon is in advanced talks to buy Decart, a startup founded in Israel in 2023 that develops AI efficiency software and real-time video models, in a deal that could reach $7.5 billion. The acquisition would give its cloud business more control over technology that helps models run efficiently on its own chips—and over the engineers developing it.
A report citing people familiar with the matter said an agreement could come together as soon as early next week, while warning that talks could still fall apart. Amazon did not respond to a request for comment; Decart declined to comment.
The strategic prize is adoption of Amazon Web Services’ computing infrastructure. AWS, Amazon’s cloud division, has developed Trainium AI accelerators to compete with Nvidia’s widely used graphics processing units. Better software can make those chips more attractive to developers without requiring Amazon to win on hardware alone.
Decart co-founders Moshe Shalev, left, and Dean Leitersdorf in a 2024 photograph published by Sequoia Capital. Source: Sequoia Capital.
A price tied to people and performance
The headline figure is not simply cash payable at signing. A November 18 report on the negotiations said Amazon was expected to pay more than $7 billion through cash, stock and deferred payments, with target-dependent consideration potentially bringing the total to nearly $7.5 billion.
That report said most legal issues delaying the transaction had been resolved, after changes to its structure. An earlier valuation under discussion was $5 billion to $6 billion. Its expectation of a possible signing that weekend preceded the newer report’s early-next-week timetable; neither establishes a completed deal.
Decart was founded in 2023 by Dean Leitersdorf, its chief executive; Orian Leitersdorf, its chief scientist; and Moshe Shalev, its chief product officer. The founders are veterans of Israel’s military intelligence Unit 8200. Dean Leitersdorf’s relevant background also includes a doctorate from Israel’s Technion and research in distributed computing, according to seed investor Sequoia Capital. Its described Dean Leitersdorf and Shalev starting with training and inference optimization, building an enterprise business before applying that infrastructure to AI-generated experiences. Efficiency software preceded the more visible video products.
The University of California is expanding ChatGPT and Codex access across 10 campuses, with extra charges for advanced models and an unresolved dispute over faculty authority.
Under the reported terms, all three founders have agreed to stay for five years and not establish a competing business. A significant portion of the consideration would be paid only in the fifth year, with employment arrangements contributing to the negotiating delay. Those provisions make retaining the team part of the transaction’s economics.
Decart raised $300 million in May at a $4 billion valuation, with investors including Sequoia, Benchmark, Radical Ventures and Nvidia, according to the report on the latest talks. The November 18 account puts total fundraising at more than $400 million. The potential $7.5 billion package is about 1.9 times May’s valuation, but the comparison includes conditional and deferred acquisition payments; it is not a measure of an investor’s return.
The same November 18 report put revenue, by some estimates, in the low tens of millions of dollars, driven mainly by semiconductor-company projects and varying from quarter to quarter. It did not provide a defined revenue period or audited figures, preventing a reliable acquisition-to-revenue multiple. An October analysis also relayed a Globes report that Decart had been profitable almost from inception. The analysis also cited Bloomberg in identifying eBay as both a customer and an investor. These accounts supply evidence of an existing commercial business, although neither provides financial statements to assess the proposed price.
Reported terms include a valuation of about $7.5 billion and a five-year founder commitment. Consideration includes conditional and deferred payments; talks could still collapse. Source: Financial Times.
Software is the bridge to Trainium
Decart also develops Lucy, which can edit live video, and Oasis, which generates interactive virtual worlds. It was an early user of Trainium3, which Amazon launched in late 2025.
The difficulty Amazon is addressing extends beyond processor performance. That analysis described Nvidia’s CUDA software ecosystem as a major source of its strength: developers have spent years building around it, and moving workloads to other chips normally requires additional engineering. The analysis, citing the Wall Street Journal, said Decart sells software that speeds models on Nvidia, Google and Amazon processors and eases switching between them, licensing it to cloud providers and AI labs. Decart is based in San Francisco with a development center in Tel Aviv.
Amazon already has an alternative software foundation. Its Neuron documentation, for the release dated November 11, describes a development kit for Trainium and Inferentia chips, supporting model training and inference through frameworks including PyTorch and JAX. More directly, AWS says TorchNeuron, its native PyTorch backend for Trainium2 and Trainium3, lets most PyTorch workflows run with few or no changes. That tool is in beta. The same documentation offers beta software for serving diffusion and video-generation models on those chips. Decart would join an existing effort to simplify migration and run video workloads; its incremental value needs to be judged against those capabilities.
There is also a concrete, if company-supplied, performance baseline. In its December 2, 2025 launch announcement, Amazon said Decart was achieving four times faster inference for real-time generative video at half the cost of GPUs. Inference means running a model to produce outputs, rather than training it.
That comparison concerns Decart’s video workload. The announcement does not identify the GPU model or lay out a reproducible test method for that claim, so it cannot establish equivalent savings across other models. Amazon was already promoting results from its work with Decart well before the acquisition talks reached this stage.
The infrastructure constraint also extends beyond software portability. Amazon’s launch announcement describes communication between chips and limits on parallel computation as obstacles to scaling AI workloads. A model that runs on Trainium still needs to run efficiently across the hardware and networking available to the customer. Decart’s video result alone does not quantify savings at that broader scale.
AWS’s Trainium3 chip, pictured by Amazon. Decart was an early user of the accelerator. Source: Amazon.
An established chip team, after other buyers fell away
The executive reportedly leading Amazon’s negotiations is Nafea Bshara, who heads engineering for its Annapurna Labs chip operation. He co-founded Annapurna in 2011 after serving as chief technology officer at chipmaker Marvell’s Israeli operation, according to a report identifying his role in the deal.
Annapurna was itself an Israeli startup acquired by Amazon and subsequently developed its cloud processors and AI chips. Its work includes Trainium, used for training and inference, and Inferentia, designed for inference, competing with processors from Nvidia and AMD. Bshara’s involvement places the prospective purchase alongside Amazon’s established silicon effort, where software efficiency could influence cloud customers’ costs and hardware choices.
Amazon was not Decart’s only prospective owner. The November 18 account said AI model developer Anthropic had pursued a $6 billion acquisition, exchanging draft agreements and effectively completing due diligence before withdrawing in early September. Decart then sought to revive earlier discussions with Nvidia, already an investor, but the chipmaker declined to reopen negotiations.
Anthropic is also an Amazon-backed AI developer and a Trainium customer, using the chips to train and run its Claude models, as the October analysis explains. That account, citing Bloomberg, said Anthropic had pursued Decart to get more out of its existing chips. The former bidder and the current one therefore belong to overlapping commercial relationships.
Nvidia has several roles in this story: Decart investor, former suitor and supplier of processors its software supports. Amazon’s own launch announcement also says it is designing Trainium4 to support Nvidia’s NVLink Fusion interconnect technology, with infrastructure intended to accommodate both GPU and Trainium servers. The announcement presents that integration as a development plan. Competition between processors can coexist with cooperation on the systems connecting them; a Decart acquisition would change who controls its software priorities.
The next test is an agreement, then broader results
The immediate question is whether the parties sign and disclose the final price, conditions and employment commitments. The report identifying Bshara suggested a possible unveiling at AWS’s re:Invent event at the end of November, but that remains an expectation rather than an announced plan.
Beyond signing, the unresolved question is how widely Decart’s gains can extend beyond its demonstrated workloads. Comparable tests across those workloads, including the GPU baseline, migration effort and total operating cost, would help establish the value beyond the existing Neuron tools. Customers licensing Decart’s software also need clarity about continued support for rival chips. Those are the tests of whether a signed acquisition could translate into wider Trainium adoption.
Ray-Ban’s US store cuts the Gen 2 Wayfarer by 25% to $284.25. Buyers still face separate prescription-lens prices, a cheaper Gen 1 option and newer glasses outside Meta’s frame sale.
Ubisoft has made the team behind its Teammates experiment a dedicated subsidiary targeting AI characters with memory and outside game creators. Earlier playtests show both progress and limitations, while commercial plans remain undisclosed amid group cost cuts and refinancing.
Neuralink has shown a trial participant controlling GTA VI through intended movements. Earlier gaming use and company-reported trial results put the demonstration in context, while its reliability across users remains unmeasured in the reports.
SoftBank founder Masayoshi Son reportedly explored financing a $1 trillion acquisition of memory-chip maker SK hynix, but the company says the deal is no longer under active consideration. The target brings surging earnings, existing shareholders and customer commitments that complicate any simple equation between ownership and access to AI-memory supply.
Nvidia reported $115.8 billion in quarterly revenue, but its shares fell 14% Friday amid a wider technology selloff, a leadership transition and a reported accounting investigation. Strong cash generation offers a counterweight to concerns about the durability of AI spending.
The Nasdaq 100 fell 5.6% on November 20 and closed 19.8% below its record after an intraday bear-market breach. Nvidia’s revenue surge and Anthropic’s huge IPO did not halt the rout; infrastructure bills and resilience elsewhere in the market complicate the picture.
OpenAI reports sharply lower reasoning-extraction rates in internal GPT-6.2 tests and backs sanctions against illicit foreign distillation. Its own research and rival Anthropic’s experience show why protecting hidden reasoning leaves capability transfer and customer costs unresolved.
Alberta proposes 600-metre setbacks for large data centres and expanded utility oversight, but exemptions and rules still to be written will determine how much protection communities receive.
President Donald Trump backed Commerce Secretary Howard Lutnick as lawmakers scrutinize alleged delays in restricting a suspected route for AI chips to China. A new congressional inquiry also questions Nvidia’s influence on an existing investigation.
Amazon’s month-old Echo Time is $50 off for Black Friday. Its touchscreen, home hub and optional sleep-aware alarm cost $90 more than the reported Echo Spot price, while Alexa+ access and evidence of better waking remain key buying questions.
Merck will pair its experimental data with Isomorphic Labs’ AI to seek small-molecule cancer drug candidates. The deal follows Isomorphic’s $5 billion financing, but its payments, target and development timetable remain undisclosed.
Thinking Machines Lab released a roughly 20GB open-weights model that reuses its middle layers for reasoning. It brings the Inkling family to local hardware, with competitive coding results but workload-dependent costs and preliminary evidence on oversight.
Apple’s first HomePod mini update since 2020 adds Mac audio and four-speaker surround, but Siri AI remains unavailable, core home-hub features carry over, and reviews disagree on playback reliability.
Treasury Secretary Scott Bessent says an unnamed Anthropic founder will announce a large gift to children's investment accounts within weeks, but the amount, recipients and connection to existing philanthropic pledges remain undisclosed.
Google says air cooling will limit water use at its Visakhapatnam AI hub, but environmental proceedings and residents’ accounts raise broader questions about land, electricity, public incentives and the evidence behind promised benefits.
Nvidia slipped behind Alphabet to fourth in U.S. market value during November 20 trading. The reversal follows a steep selloff despite strong reported revenue growth, with a permanent CEO still to be chosen.
Google will close the Nest app on January 25, 2027 and retire Assistant on home speakers and displays in August. Device transfers are automatic, but continued voice access requires accepting Gemini; basic assistance remains free.
Amazon says Cairn agents exploited a MONAI Label zero-day to take over a university server. Other attacks failed, and maintainers list no patched package release.
Federal Reserve Governor Lisa D. Cook sees early AI pressure on coding, translation and young workers, but says a solid overall labor market and persistent inflation still call for tighter monetary policy.