Anthropic, the San Francisco-based artificial intelligence company valued at $965 billion in its most recent private funding round, is in advanced discussions to acquire Israeli AI startup Decart for approximately $6 billion, a deal that would represent its largest known acquisition and arrive weeks before a planned Nasdaq listing that investment banks are positioning as a debut above $1 trillion.
Key Takeaways
- Anthropic is in talks to acquire Decart AI for approximately $6 billion, a roughly 50% premium over the startup’s nearly $4 billion valuation from a May 2026 Series B round led by Radical Ventures.
- Decart’s core product, DOS (Decart Optimization Stack), delivers 1,600 tokens per second for agentic inference across Nvidia GPUs, Google TPUs, and Amazon Trainium chips, roughly eight times the industry benchmark of 200 tokens per second.
- Anthropic filed a confidential draft S-1 with the SEC on June 1, 2026, and began scheduling investor meetings in mid-July; Goldman Sachs, Morgan Stanley, and JPMorgan Chase are leading the underwriting.
- Anthropic’s annualized revenue run rate has accelerated from approximately $1 billion at the start of 2025 to an estimated $69 billion by July 2026, driven by enterprise adoption of Claude across more than 300,000 business customers.
- OpenAI has delayed its own IPO to 2027, leaving Anthropic positioned to capture first-mover advantage among frontier AI companies entering the public markets.
The Acquisition Targets Inference Economics, Not Model Architecture
For investors evaluating the strategic logic of the deal, the distinction matters: Anthropic is not acquiring a competitor or a new model family. Decart builds infrastructure that makes existing models run faster and cheaper on the same hardware. The DOS optimization stack functions as a hardware-to-model layer that reduces the cost of inference, the computationally expensive process of generating each response a user receives from Claude, by extracting more throughput from chips already deployed in Anthropic’s data centers.
The performance claims are specific. Decart reports that DOS delivers 1,600 tokens per second for agentic inference workloads, compared with an industry average the company benchmarks at approximately 200 tokens per second. For video processing, DOS handles full-HD inference at up to 100 frames per second. The stack runs across Nvidia GPUs, Google TPUs, and Amazon Trainium chips, a cross-platform capability that aligns with Anthropic’s multi-cloud deployment strategy.
On top of DOS sit two model families. Lucy is a real-time video transformation model deployed in e-commerce virtual try-on and live advertising applications, processing video feeds with sub-30-millisecond response times. Lucy 2.5, the current production version, runs on Amazon Trainium3 silicon. Oasis is a world model that generates interactive three-dimensional simulation environments for robotics training, autonomous vehicle development, and manufacturing logistics. The original Oasis demo, a fully playable AI-generated game world released in late 2024, reached one million users within 72 hours.
Decart was founded in 2023 by Israeli brothers Dean and Orian Leitersdorf and Moshe Shalev. The founding team’s background spans deep systems engineering and GPU-level optimization. The company emerged from stealth with a seed round led by Sequoia Capital after demonstrating what it described as unprecedented efficiency in real-time AI training and inference. Its May 2026 Series B was led by Radical Ventures with participation from Nvidia, Sequoia, Benchmark, Zeev Ventures, Atreides Management, Valor Equity Partners, and Adobe Ventures.
The Margin Implications Are What Matter for the S-1
Anthropic filed a confidential draft S-1 registration statement with the SEC on June 1, 2026, and began scheduling investor meetings in mid-July. Goldman Sachs, Morgan Stanley, and JPMorgan Chase are leading the underwriting process. The listing is targeted for Nasdaq, with October 2026 as the current window. Investment bankers cited in reporting now consider a debut valuation above $1 trillion the base case, up from the $965 billion set in the May 2026 Series H-1 round.
The revenue trajectory heading into the roadshow is steep. Anthropic’s annualized revenue run rate has scaled from approximately $1 billion at the beginning of 2025 to roughly $5 billion by August 2025, $30 billion by April 2026, $47 billion by May 2026, and an estimated $69 billion by July 2026. The growth has been driven overwhelmingly by enterprise adoption of Claude. The company reports more than 300,000 business customers, over 25 billion API calls per month, and more than 1,000 accounts spending above $1 million in annual recurring revenue.
Within that context, the Decart acquisition is a margin play as much as a technology acquisition. If DOS can reduce the cost per query for Claude by extracting more inference throughput from existing hardware, the improvement flows directly to gross margin, the metric public market investors will scrutinize most aggressively in the S-1. AI companies face structural pressure on margins because every user interaction carries a real compute cost. A technology that lowers that cost per interaction, without requiring proportional increases in data center capacity, strengthens the unit economics story that Anthropic will present to institutional investors during the roadshow.
The $6 billion price tag, while substantial, is modest relative to Anthropic’s own valuation. At $965 billion, the Decart acquisition would represent approximately 0.6% of enterprise value, a ratio that is well within the range of accretive technology acquisitions by pre-IPO companies seeking to lock in infrastructure advantages before listing.
Nvidia’s Position Creates a Layer of Investor Complexity
Nvidia participated in Decart’s May 2026 Series B and has committed over $40 billion in AI equity investments in 2026 alone. Nvidia’s typical investment pattern involves taking a stake in a portfolio company, which then signs a long-term GPU purchase commitment, creating a revenue loop that benefits Nvidia’s top line. Decart is an atypical fit for that pattern because DOS is explicitly chip-agnostic, designed to optimize performance across Nvidia, Amazon, and Google hardware rather than deepening dependency on any single chip vendor.
For Anthropic, Decart’s cross-platform capability is part of the strategic appeal. Anthropic runs inference across multiple cloud providers, and a technology that improves performance on all of them, rather than locking the company into a single silicon vendor, reduces concentration risk and preserves negotiating leverage with hyperscalers. For Nvidia investors, the question is whether a Decart acquisition by Anthropic would alter the commercial relationship between the two companies. Decart’s optimization of non-Nvidia hardware could theoretically reduce Anthropic’s dependency on Nvidia GPUs over time, though the near-term demand for Nvidia silicon across the AI industry remains supply-constrained regardless.
The $65 billion Series H-1 round that set Anthropic’s $965 billion valuation was co-led by Altimeter Capital, Dragoneer Investment Group, Greenoaks Capital, and Sequoia Capital, with additional participation from Baillie Gifford, Blackstone, Brookfield, D.E. Shaw Ventures, Fidelity, Samsung, SK Hynix, and Micron. Secondary market activity in mid-2026 has at times implied valuations ranging from $1.05 trillion to $1.2 trillion, though secondary prices carry liquidity risk and are not guaranteed to reflect the final IPO pricing.
Anthropic Would Be the Fifth AI Company This Cycle to Consolidate Before Listing
The Decart deal would be Anthropic’s fifth acquisition of 2026, a pace that marks a shift from the company’s historically organic growth strategy. The acceleration suggests that Anthropic’s leadership, led by CEO Dario Amodei, views the pre-IPO window as the optimal period to lock in infrastructure advantages while the company still has the flexibility of private capital and the willingness of investors to tolerate dilutive acquisitions.
The pattern is not unique to Anthropic. Multiple AI companies preparing for public markets in 2026 have pursued pre-IPO acquisitions designed to strengthen their infrastructure layers. The logic is straightforward: public market investors will evaluate AI companies not only on model performance but on the sustainability of their compute economics. A company that can demonstrate declining cost per query alongside rising revenue per customer presents a more compelling growth-and-margin story than one that scales revenue but watches margins compress under infrastructure costs.
OpenAI’s decision to delay its own IPO to 2027 has opened a window for Anthropic to establish the first public market benchmark for frontier AI company valuation. SpaceX’s June 12, 2026, debut at $135 per share and a $1.77 trillion valuation demonstrated that the public market appetite for large-scale technology IPOs remains intact, despite elevated interest rates and ongoing geopolitical uncertainty. The Anthropic IPO, if executed in October, would test whether that appetite extends to a pure-play AI company whose revenue has scaled faster than any enterprise software company in history but whose path to sustained profitability remains a subject of active investor debate.
The Deal Has Not Closed and Faces Multiple Execution Risks
Neither Anthropic nor Decart has issued an official statement confirming the acquisition, and representatives for both companies have declined to comment. The deal could collapse over valuation disagreements, regulatory scrutiny, or shifts in Anthropic’s IPO timeline that alter the strategic calculus for pre-listing acquisitions.
For Decart’s existing customers and partners, including Amazon Web Services, which features Decart as a reference deployment on Trainium, the acquisition would raise questions about platform access. If DOS becomes exclusive to Anthropic’s internal infrastructure, AWS and other cloud providers would lose a third-party optimization tool that currently helps them attract AI workloads. If Anthropic maintains DOS as an externally available product, it would create a business model tension between optimizing its own inference costs and selling that same optimization capability to competitors.
The next scheduled catalyst for Anthropic’s public market narrative is the transition of the S-1 from confidential to public filing, which is expected to occur before the roadshow begins. That document will provide the first audited view of revenue, costs, margins, customer concentration, and capital expenditure projections, the data points that will determine whether the $1 trillion-plus valuation implied by the roadshow holds under institutional scrutiny.
Disclaimer: This article is for informational purposes only and does not constitute financial, investment, or trading advice. Readers should consult a licensed financial professional before making investment decisions.
FAQs
Why Is Anthropic Acquiring Decart Before Its IPO?
The acquisition is designed to strengthen Anthropic’s inference infrastructure ahead of its planned Nasdaq listing. Decart’s DOS optimization stack can reduce the cost of generating each Claude response by extracting more throughput from existing hardware. Lower cost per query improves gross margins, the metric public market investors will evaluate most closely in the S-1. The pre-IPO window also gives Anthropic the flexibility to absorb a $6 billion acquisition using private capital without the disclosure and shareholder approval requirements that would apply post-listing.
When Is Anthropic Expected to Go Public?
Anthropic filed a confidential draft S-1 with the SEC on June 1, 2026, and began scheduling investor meetings in mid-July. The current target is a Nasdaq listing in October 2026. Goldman Sachs, Morgan Stanley, and JPMorgan Chase are leading the underwriting. The company’s most recent private valuation was $965 billion following its May 2026 Series H-1 round, and investment banks expect a debut above $1 trillion. The timeline remains subject to SEC review, market conditions, and final pricing decisions.
What Does Decart AI Actually Build?
Decart operates three product lines. DOS (Decart Optimization Stack) is a cross-platform inference and training engine that runs on Nvidia GPUs, Google TPUs, and Amazon Trainium chips, delivering 1,600 tokens per second for agentic inference. Lucy is a real-time video transformation model used in e-commerce virtual try-on and live advertising. Oasis is a world model that generates interactive 3D simulation environments for robotics, autonomous vehicles, and manufacturing. All three products are built around real-time performance with sub-30-millisecond response latency.
