The race to build the most capable AI model is no longer just about raw compute—it's increasingly about squeezing every last drop of performance from the silicon you already own. Anthropic appears to be betting its future on that premise. The Claude developer is in advanced negotiations to acquire Israeli AI infrastructure startup Decart for approximately $6 billion, according to multiple reports from Bloomberg and other outlets citing anonymous sources familiar with the matter. If finalized, the deal would represent Anthropic's largest known acquisition and land just weeks before the company's highly anticipated initial public offering. Neither Anthropic nor Decart has officially commented on the talks, and sources caution that negotiations remain fluid and could still collapse. But the strategic signal is unmistakable: as the AI industry's leading labs pour hundreds of billions into data centers, the next competitive frontier may be how efficiently those investments translate into usable output.
A Short but Heavily Valued History
Decart is barely three years old, but its growth trajectory reads like a Silicon Valley fairy tale with a Tel Aviv twist. Founded in 2023 by Israeli engineers Dean Leitersdorf, Orian Leitersdorf, and Moshe Shalev, the company was profitable almost from inception, per company statements cited in industry reports. Its core offering is the Decart Optimization Stack (DOS 2.0), a training and inference platform designed to run AI workloads across NVIDIA GPUs, Google TPUs, and Amazon Trainium—essentially, a hardware-agnostic software layer for performance. The company's claims are eyebrow-raising. Decart says DOS 2.0 delivers over 1,600 tokens per second for agentic inference, compared to an industry average of roughly 200, and handles full-HD video and world model inference at up to 100 frames per second. It also claims a more than 100x improvement in cost efficiency and real-time AI performance 8x faster than comparable systems on equivalent hardware. Notably, these are company-provided figures and have not been independently verified. Beyond the optimization stack, Decart has developed two "world models": Lucy, a platform for real-time experiences in gaming, e-commerce, advertising, and streaming with sub-30ms response times, and Oasis, positioned as a world model for Physical AI—real-time simulation for robotics and autonomous driving systems. The price tag of $6 billion is a substantial premium over Decart's May 2026 valuation of nearly $4 billion, following a $300 million funding round led by Radical Ventures with participation from NVIDIA, Atreides Management, Valor Equity Partners, Adobe Ventures, Toyota Ventures, and eBay Ventures. Existing investors Sequoia Capital, Benchmark, and Zeev Ventures also participated. The company's cap table reads like a who's who of AI and media: OpenAI co-founder Andrej Karpathy, former Disney CEO Michael Eisner, and the Nintendo founding family all hold stakes, according to public reports.
Efficiency as a Defense Strategy
The timing and rationale of this deal are deeply intertwined with Anthropic's immediate future. The company has committed to spending hundreds of billions of dollars on data centers packed with expensive—and, critically, supply-constrained—accelerator chips. Decart's software is essentially a leverage tool: it promises to extract more usable capacity from the same physical infrastructure, effectively lowering the marginal cost of every inference call Claude processes. This is not merely a cost optimization play. In a landscape where customers increasingly choose models based on price per token and latency, the ability to deliver more compute-per-dollar is a product differentiator. Anthropic's multi-hardware strategy—it has recently onboarded AMD MI series GPUs alongside Google TPU and Amazon Trainium—suggests the company wants to avoid being locked into a single vendor's roadmap. Decart's hardware-agnostic stack would align neatly with that approach. Some industry analysts have framed the acquisition as an "IPO declaration." Investors on prediction market Kalshi place an 85% probability on Anthropic announcing its IPO in 2026. The company confidentially submitted a draft registration statement to the SEC on June 1, 2026, and reports suggest it aims to complete a listing by September or October, with a valuation target of $965 billion to $1 trillion. That ambition comes after a meteoric 2026: a February valuation of $380 billion, followed by a May Series H round of $65 billion at a post-money valuation of $965 billion. For context, that surpasses OpenAI's last reported valuation of $852 billion. But the pre-IPO narrative has cracks. Reportedly, investor meetings have surfaced three persistent concerns: competition from low-cost Chinese AI models, the company's public tensions with the Trump administration, and the "SpaceX cautionary tale" of the rocket company's IPO stock decline from $225 to $108 after a record-setting launch. In that context, an acquisition that promises better unit economics could be a persuasive counterargument.
The NVIDIA Subplot and "Hijacking" Narrative
Every good deal story needs friction, and this one has plenty. Israeli business outlet Calcalist reported that Decart had been close to a sale agreement with NVIDIA before a "larger" buyer entered the process. Amazon, Nebius, and SpaceX were also floated as potential acquirers, though Elon Musk publicly stated SpaceX will not be buying Decart. The image of Anthropic sweeping in to outbid NVIDIA—a company that just invested in Decart at a $4 billion valuation months ago—has captured the industry's imagination. If the $6 billion price is realized, NVIDIA stands to make a handsome return on that investment in roughly three months. But the chip giant also loses strategic control over a startup whose software could be used to make competitors' hardware more attractive. (NVIDIA's December 2025 acquisition of Groq for roughly $200 billion was a hardware-side play on inference efficiency; Decart represents the complementary software side.)
Community Reaction: Buying Speed vs. Building It
The developer community, as is often the case, is skeptical. Comments on Hacker News range from "Anthropic is jumping a shark for sure. I think they're trying everything to grab headlines, instead of making their model cheaper, better to use" to a more derisive "Before they buy Dehorse?" The criticism reflects a broader unease about massive M&A in an industry where open-source models are improving at breakneck speed. Some industry observers question whether efficiency gains are better built in-house, rather than bolted on through billion-dollar acquisitions. That said, Anthropic is not alone in this strategy. OpenAI has reportedly acquired patent portfolios from neuromorphic chip designer Rain AI, taken a stake in Cerebras, and acquired AI consulting firm Tomoro. Nebius bought AI optimization startup Eigen AI for $643 million. The "efficiency layer" is rapidly becoming the industry's hot acquisition category.
[SPONSORED]
NEXT-GEN NPU CHIPSETS
Empower your local devices with desktop-class inference capabilities.
The Physical AI Vector
There's a second strategic dimension often buried beneath the M&A coverage: Decart's world models. Oasis, with its focus on real-time simulation for robotics and autonomous driving, positions Anthropic to compete in what NVIDIA calls "Physical AI"—a market estimated by some analysts to be worth as much as $6 trillion by 2040. At a minimum, it gives Anthropic a beachhead in verticals—manufacturing, autonomous vehicles, humanoid robotics—where AI labs like Google's DeepMind and Elon Musk's xAI have already established inroads. According to public reporting, Oasis 3, released in June 2026, can generate photorealistic driving environments in real time via API, with initial customers among autonomous vehicle companies. Pricing is advertised at $0.02 per second. This is a highly speculative—but potentially game-changing—investment area, far removed from the enterprise chat assistants that anchor Anthropic's current revenue.
What Happens Next
If the deal closes, Decart's team—which spans Tel Aviv and San Francisco—will join Anthropic's inference and performance organization, bringing its optimization expertise into the company's core engineering operations. Decart's investors, including Sequoia, Benchmark, and NVIDIA, would see a tidy return on a three-year-old company. Anthropic would gain its first Israeli footprint, joining Google, Microsoft, and Amazon as major AI-powered R&D presences in the country. Talks are still ongoing and could collapse. M&A of this size and complexity often does, especially with an IPO clock ticking. But the fact that Anthropic is even willing to consider a $6 billion acquisition in the final sprint to a public listing—a period when companies typically emphasize cash preservation and simple metrics—tells you how seriously it takes the efficiency problem. When OpenAI and Anthropic are each committed to spending hundreds of billions on datacenters, the difference between winner and also-ran may well come down to who can make those dollars work hardest. Decart could be a significant piece of that math—or just another expensive lesson in Silicon Valley hubris. Right now, both outcomes look plausible.