SpaceX Reportedly Targets Israeli AI Lab Decart: $6B for Faster Inference
Decart’s optimization stack runs AI eight times faster; would fill gap from xAI exodus
SpaceX is reportedly in advanced talks to acquire Israeli AI startup Decart at a valuation of approximately $6 billion to $7 billion — a deal that would hand Musk's company a proprietary inference-optimization stack capable of running AI models at roughly eight times the industry average speed, and establish SpaceX's first engineering foothold in Israel, according to an exclusive report by Calcalist, Israel's leading financial newspaper. The deal could be signed within a week. If completed, it would be SpaceX's second major AI acquisition since its $75 billion IPO in June, following the $60 billion Cursor all-stock deal announced just weeks after the company went public.
Three-year-old Decart originally entered advanced negotiations with Nvidia — which had invested in the company's most recent funding round — before a larger competing bid caused the founders to pivot to a new buyer, according to Calcalist's exclusive reporting. Multiple sources cited by Calcalist believe SpaceX is now the leading candidate, though Amazon and Dutch-based cloud company Nebius are also believed to be in the picture. Decart declined to comment.
What Is Decart, and Why Do Three Tech Giants Want It
Decart is one of a handful of companies worldwide that has made real-time generative AI video commercially viable at scale, according to Radical Ventures' investment case, from the firm that led the company's most recent funding round. Founded in September 2023 by Dean Leitersdorf, 27, and Moshe Shalev, 38 — both veterans of Israel's elite signals intelligence unit, Unit 8200 — the company has raised approximately $450 million across a rapid series of rounds that reflect the intensity of investor interest in its underlying technology.
The company's calling card arrived in October 2024, when it released Oasis: a demo showing that an AI model could generate a fully interactive, Minecraft-style virtual environment at 20 frames per second, with no two players experiencing the same world. Oasis drew one million users in days and briefly surpassed ChatGPT in downloads. Elon Musk, who tried the platform shortly after launch, posted a single word on X: "wow." Musk and Leitersdorf have stayed in regular contact since then, according to Calcalist. A personal bridge also runs through Shaun Maguire — the Sequoia Capital partner who has backed Decart since its seed round and organized Musk's 2023 post-October 7 visit to Israel, during which Musk toured the Gaza border area.
The Oasis demo attracted attention as a remarkable demonstration of generative technology, but what was genuinely new was the economics underneath it. Decart was not just generating virtual worlds — it was generating them at a cost low enough that the technology could be deployed commercially rather than confined to research demonstrations. The company does not want to be a gaming company; its technology was always aimed at something larger, as the founders have described in interviews.
The Technical Case: What DOS Actually Does
The engine behind every Decart product is a piece of infrastructure called DOS — the Decart Optimization Stack — and it is the primary reason three major technology companies are competing to acquire a startup with "several tens of millions of dollars" in annual revenue and fewer than 100 employees.
DOS is a vertically integrated platform that operates at a lower level of the hardware stack than most AI software companies ever touch. It spans hardware-aware model design, custom kernel tooling, proprietary compilers, and inference optimization — effectively rewriting how an AI model talks to a chip, rather than accepting whatever efficiency the chip provides by default. The result, according to Radical Ventures' Series B announcement: DOS delivers over 1,600 tokens per second for agentic AI inference, compared to an industry average of roughly 200 — approximately eight times faster. On Amazon's Trainium chips specifically, Decart's systems achieve over 80% Model FLOPS Utilization — meaning more than 80% of the chip's raw computing capacity is doing actual productive work, compared to an industry norm closer to 40 to 50%.
What makes DOS commercially unusual is that it is hardware-agnostic: it runs across Nvidia GPUs, Amazon's Trainium accelerators, and Google TPUs, integrating into a customer's existing infrastructure rather than requiring a switch to a specific vendor's hardware, per Radical Ventures' investment thesis. This is the specific reason both chip manufacturers and cloud companies have been paying Decart for access to the technology — and why Nvidia's own acquisition negotiations ended not because the deal fell apart over valuation, but because a higher competing offer arrived.
Built on top of DOS are two product lines. Lucy is Decart's real-time video world model: it transforms live video feeds — from a camera, a screen, or a game — responding to user inputs in under 30 milliseconds and running at 1080p at 30 frames per second. Applications include virtual clothing try-on, dynamic advertising, live streaming, and gaming. Oasis is Decart's world model for physical AI: it generates physically accurate, real-time simulations for robotics and autonomous vehicle training, targeting scenarios that traditional physics-engine tools like CARLA and Isaac Sim cannot produce at scale because those tools require hand-built environments, according to Decart's product documentation. Together, the two product lines — and the DOS stack powering both — make the case that Decart has built not a product but an infrastructure layer: the piece that makes real-time generative AI economically viable at deployment scale.
Chris Penrose, Global Vice President for Telecom at Nvidia, described the significance in a March 2026 collaboration announcement: running Decart's Lucy model on Nvidia's distributed edge computing infrastructure enabled interactive AI applications at global scale with near-instant response times.
Musk's AI Empire and the Gap DOS Would Fill
To understand why SpaceX would acquire a company with modest current revenues at a $6 to $7 billion price tag, it helps to understand the state of SpaceX's AI division after a turbulent 18 months.
SpaceX acquired Musk's AI startup xAI in February 2026 at a valuation of $250 billion for the AI unit, merging under one roof the Grok large language model and the Colossus supercomputer cluster in Memphis. Within weeks of the acquisition, the cultural friction between xAI's research-oriented founding team and SpaceX's hardware-execution culture became visible. All 11 original xAI co-founders departed between February and March 2026 — an attrition rate described by analysts as virtually unprecedented at this scale for a technology company post-acquisition. Musk publicly acknowledged the outcome: "xAI was not built right first time around," he wrote, "so is being rebuilt from the foundations up."
Cursor, acquired in June 2026, represents one layer of that rebuild — developer tools and coding assistance. Decart, if the deal closes, would represent a different and more foundational layer: the optimization infrastructure that determines how efficiently SpaceX's entire AI compute estate runs.
SpaceX's financial disclosures make the logic concrete. In the second quarter of 2026, the company reported $14.1 billion in contracted AI cloud sales — agreements with Alphabet and Anthropic under which SpaceX's compute infrastructure serves as an AI cloud provider. Musk told investors that a $100 billion annualized revenue run rate by December is "not a question mark," according to CNBC's reporting on SpaceX's earnings call. Reaching that target requires that SpaceX's hardware — including the Colossus supercomputer and any future compute expansion — runs as efficiently as possible. DOS is specifically the software that closes that gap.
Calcalist notes that Decart's technology "could help SpaceX generate more computing output from the same physical resources" — an insight that translates directly into margin: more AI revenue per GPU purchased.
What the Integration Risk Looks Like
The xAI co-founder exodus is not only a piece of context — it is the most important precedent for evaluating what a Decart acquisition by SpaceX would actually mean for the company and its founders.
Leitersdorf has spoken in consistent and specific terms about ambitions that exceed any conventional acquisition outcome. In a Calcalist interview about his ambitions, he described his goal as building something comparable to Google, Apple, or Facebook in AI — a company that changes what the world looks like before and after. Shalev described their shared goal at founding: to solve "a huge challenge because we lived in a time that was bigger than us," according to Calcalist's reporting. These are the stated ambitions of founders who have said it would take five years to build something world-changing — at a company that is barely three years old.
A sale to SpaceX, or to Amazon, or to Nebius, resolves the financial question immediately but leaves open whether Decart's founders will thrive inside a larger organization where the engineering culture, execution tempo, and strategic priorities have been set by other hands. The xAI precedent — where eleven accomplished AI researchers departed within weeks of acquisition by SpaceX — is not a prediction of the same outcome for Decart's founders. But it is the most recent data point on what happens when a research-oriented AI lab encounters SpaceX's operating model, and a new acquirer would need to structure the deal differently to avoid repeating it.
Calcalist notes the significance directly: a deal with Musk, unlike a deal with a chipmaker or cloud company that already has an Israeli presence, "could potentially give Decart a different role in the global technology landscape" and bring a major new technology company into Israel. If SpaceX is the buyer, it would establish an R&D center in Tel Aviv — its first engineering footprint in Israel — using Decart as the foundation.
Two Founders From Opposite Worlds
Behind the technical platform are two founders with strikingly different origins who, by Shalev's account, complement each other precisely because of the gap between them.
Leitersdorf is a product of academic acceleration and a prominent Israeli financial family — his brother Yoav founded YL Ventures, a cybersecurity-focused venture capital firm. Leitersdorf began his computer science undergraduate degree at 17, completed his master's shortly after, and earned his doctorate from the Technion at 23, the youngest PhD graduate the institution had recorded until his younger brother Orian surpassed him. His dissertation earned the prestigious ACM PODC Dissertation Award and produced a postdoctoral fellowship offer at the National University of Singapore, which he declined to found Decart instead.
Shalev's path was structurally opposite. Raised in an ultra-Orthodox household in Bnei Brak, he pursued a bachelor's degree in accounting through evening classes while working to support his family, having married young and become a father of two children. At 23, he enlisted in the IDF through a program specifically designed to integrate Haredi men into technical military units. He served in Unit 8200 for 13 years, eventually becoming the right-hand aide to then-unit commander Yossi Sariel, before meeting Leitersdorf by chance at the base in the week after Operation Guardian of the Walls in 2021. Their first conversation, they have said, centered on a "classified technological issue."
They registered Decart as a company on September 7, 2023. Exactly one month later, both were called up for extended reserve duty following the October 7 attacks.
The company recently added Sariel himself to its management. Sariel resigned command of Unit 8200 in December 2024, accepting responsibility for what he described as a "profound intelligence failure" — acknowledging that despite information about Hamas's plans, the unit could not break through its analytical assumptions. He has since joined Decart's leadership while also founding a separate AI company, Alma Labs, which is pursuing Israeli government contracts. His presence at Decart adds a direct connection to Israel's national security infrastructure — a factor any acquirer's due diligence team will weigh.
The Other Bidders
Amazon is named as a potential buyer, though Calcalist notes the company has historically avoided very large acquisition prices. Its most notable recent deal — an $11 billion April 2026 Globalstar acquisition deal — targeted a business generating roughly $250 million in annual revenue. Decart's revenue is described as "several tens of millions of dollars," with uneven growth driven largely by custom engagements with chip manufacturers. At the reported acquisition price, any buyer is primarily paying for the technology and the team, not for current cash flows.
Amazon's interest has a specific technical logic: Decart's DOS stack delivers its most dramatic efficiency gains on Amazon Trainium chips specifically — over 80% Model FLOPS Utilization versus typical production rates — and Amazon has been a strategic API customer since Decart's May 2026 funding round. If Amazon acquires Decart, it secures both the inference optimization advantage for its own AI cloud and prevents a competitor from gaining it.
Nebius — the Amsterdam-headquartered, Nasdaq-listed AI cloud company — is the third named candidate. Nebius has been actively building Israeli infrastructure in 2026: it was selected by the Israel Innovation Authority to build and operate the country's national AI supercomputer, signed an $880 million Mega Or data center agreement for facilities in Israel, and has been acquiring Israeli AI startups, including a reported $275 million deal to buy Tavily in February 2026. An acquisition of Decart would accelerate Nebius's position in Israel while adding world-model inference capability to a company already operating specialized AI cloud infrastructure for major hyperscalers.
Funding History: From $21M Seed to a $6B Exit in Three Years
Decart's fundraising history reflects an unusually rapid compression of the venture-capital cycle for a company building deep technical infrastructure. The $21 million seed round — led by Sequoia Capital with investor Oren Zeev — was already notable for its size at emergence from stealth in October 2024. One month later, Benchmark, one of the most selective American venture capital firms and one that had barely operated in Israel until then, led a $25 million extension at a roughly $500 million valuation.
Less than a year after that, a $100 million round pushed the total to $153 million at a $3.1 billion valuation. In May 2026, Radical Ventures' May 2026 funding round valued the company at approximately $4 billion, with Nvidia and eBay Ventures, Adobe Ventures, and Toyota Ventures also participating. Private investors in that round included OpenAI co-founder Andrej Karpathy and former Disney CEO Michael Eisner.
A sale at $6 billion to $7 billion would represent a roughly 50% to 75% premium to that May valuation in approximately three months — and would be among the largest exits in Israeli technology history, for a company that is just three years old and employs fewer than 100 people.