AI21 Labs
AI21 Labs was founded in 2017 by Amnon Shashua, Yoav Shoham, and Ori Goshen, academics from Israel with extensive research backgrounds—Shoham spent 28 years as a computer science professor at Stanford.
Profile
Build large language models and AI orchestration systems for enterprise knowledge agents.
AI21 Labs was founded in 2017 by Amnon Shashua, Yoav Shoham, and Ori Goshen, academics from Israel with extensive research backgrounds—Shoham spent 28 years as a computer science professor at Stanford. The company raised $208 million in Series C funding at a $1.4 billion valuation in November 2023, bringing total capital raised to $336 million. By 2024, the company was generating approximately $50 million in annualized revenue, though this represented a significant underperformance relative to competitors.
AI21 built a portfolio of proprietary and open-source models: Jurassic (proprietary, available via AWS and Google Cloud), Jamba (open-source, released in 2024–2025 with 256K token context windows), and Maestro (an orchestration platform for building AI agents). The company saw rapid deterioration in 2025. It discontinued Wordtune, its consumer writing tool, in April 2025, and subsequently halted development of the Jamba model family to focus solely on Maestro.
In late 2025, AI21 cut 60% of its workforce, from 180 to roughly 70 employees, after failing to close a reported $300 million Series D fundraising round. As of January 2026, Nvidia is in advanced acquisition talks valued at $2–3 billion, a deal that sources characterize as talent-driven given the company's pivot away from model development. Concurrent reporting suggests Nebius, a cloud infrastructure provider, has secured multi-million-dollar contracts to implement Maestro systems.
Who buys this
- Financial services firms using contract analysis and document understanding
- Enterprise customer support operations requiring grounded, cited responses
- Retail and e-commerce businesses automating product descriptions and customer inquiries
- Cloud infrastructure and SaaS platforms integrating AI agents via API
- Edtech and knowledge management verticals requiring long-context retrieval
Publicly disclosed clients
- Digital banking platform (using Jamba for customer support)
- Educa (edtech provider using Jamba for personalized chatbots)
- Nebius (cloud infrastructure, contracted for Maestro implementation)
Strengths and what to watch
Strengths
- Jamba architecture delivers 2–2.5× faster inference on long-context tasks compared to Transformer-only competitors, with 256K token windows.
- Scientific founding team with established credibility in AI research; backed by major infrastructure providers (Google, Nvidia, Samsung) through 2023.
- Maestro platform claims up to 50% accuracy improvements on complex reasoning tasks and 95%+ correctness rates when integrated with reasoning-optimized models.
Watch for
- Severe workforce reduction (60% in late 2025) and discontinuation of Jamba model development—company now exclusively focused on Maestro orchestration after strategic pivots and layoffs.
- Failed Series D fundraising ($300M reported but never closed) and current acquisition talks with Nvidia; acquisition characterization is talent-focused, suggesting products alone lack market viability.
- Revenue stagnation below $50M annually against competitors (Mistral ~$400M, Cohere ~$200M) despite $200M+ funding; consumer product Wordtune discontinued April 2025 after years of operation.
Recent moves
Key Information
- Industry
- Closed Source Models
- Founded
- 2017
- Headquarters
- Tel Aviv
Frequently Asked Questions
What is AI21 Labs?
AI21 Labs, founded in 2017 by Israeli academics, builds large language models and AI orchestration systems for enterprises. The company raised $336 million total, including $208 million Series C funding. By 2024, it generated roughly $50 million in revenue but faced significant underperformance versus competitors like Mistral and Cohere.
Who founded AI21 Labs?
AI21 Labs was founded in 2017 by Amnon Shashua, Yoav Shoham, and Ori Goshen, academics from Israel with extensive research backgrounds. Shoham notably spent 28 years as a computer science professor at Stanford. The company was backed by major infrastructure providers including Google, Nvidia, and Samsung through 2023.
What is Jamba?
Jamba is AI21's open-source large language model released in 2024–2025, featuring a 256K token context window and hybrid Transformer-Mamba architecture. It delivers 2–2.5× faster inference on long-context tasks than Transformer-only competitors, and is used by financial services, retail, and edtech companies for document understanding and customer support applications.
What is Maestro?
Maestro is AI21's AI orchestration platform designed for building intelligent agents. It claims up to 50% accuracy improvements on complex reasoning tasks and 95%+ correctness rates when paired with reasoning-optimized models. After discontinuing Jamba model development in 2025, AI21 pivoted entirely to Maestro as its core product.
What happened to AI21 Labs in 2025?
In 2025, AI21 Labs faced severe challenges. It discontinued Wordtune in April, then cut 60% of workforce from 180 to 70 employees after failing to close its reported $300 million Series D round. Subsequently, it halted Jamba development to focus exclusively on Maestro orchestration.
Is Nvidia acquiring AI21 Labs?
As of January 2026, Nvidia is in advanced acquisition talks to buy AI21 Labs for an estimated $2–3 billion. Sources characterize the deal as talent-driven, suggesting the acquisition targets AI21's research team rather than its products. The negotiation follows AI21's failed Series D fundraising effort.
How AI21 Labs compares
Direct head-to-head against 2 competitors. Picked by 7wData.
AI21 Labs
- Positioning
- Build large language models and AI orchestration systems for enterprise knowledge agents.
- Customer segments
- Financial services firms using contract analysis and document understanding
- Strengths
- Jamba architecture delivers 2–2.5× faster inference on long-context tasks compared to Transformer-only competitors, with 256K token windows.
- Watch for
- Severe workforce reduction (60% in late 2025) and discontinuation of Jamba model development—company now exclusively focused on Maestro orchestration after strategic pivots and layoffs.
- Recent moves
- Nvidia in advanced talks to acquire AI21 Labs for $2–3 billion
Cohere
- Positioning
- Cloud-neutral enterprise LLM vendor; sells BYOC and VPC-isolated deployment to Fortune 500 and regulated industry buyers.
- Customer segments
- Fortune 500 IT and security buyers in finance, healthcare, manufacturing; European government sovereign AI procurement teams.
- Strengths
- Model Vault (September 2025) enables inference inside customer VPCs or on-prem; data never leaves the customer environment.
- Watch for
- 2026 IPO publicly confirmed by CEO Aidan Gomez; listing pressure may force pricing changes on current enterprise contracts.
- Recent moves
- $500M funding round closed August 2025 at $6.8B valuation; Nvidia, AMD Ventures, and Salesforce Ventures among backers.
Mistral AI
- Positioning
- European open-weight and proprietary LLM vendor; data-sovereign deployment and orchestration targeting enterprise and government buyers.
- Customer segments
- European enterprises and governments requiring data-sovereign deployment; global enterprise via NTT Data and Dell systems integrators.
- Strengths
- Split control-plane/data-plane architecture runs inference inside customer Kubernetes environments; 60% of $400M-plus ARR comes from European deployments.
- Watch for
- $830M debt financing closed March 2026 for Nvidia chip purchases; leverage load before $1B revenue target creates financial exposure.
- Recent moves
- Workflows orchestration layer launched public preview April 27, 2026; durable execution engine already running millions of daily enterprise tasks.
Sources
- www.ai21.com — Company mission, founders, and investor list
- www.ai21.com — Series C funding amount ($155M in August 2023), valuation ($1.4B), and total capital raised
- www.calcalistech.com — Workforce reduction details (60% cut from ~180 to ~70 employees), Wordtune discontinuation April 2025, revenue figures (~$50M), Jamba development halt, Maestro focus shift
- www.calcalistech.com — Strategic pivot to Maestro platform only, discontinuation of standalone model sales, Nebius partnership and acquihire arrangements
- siliconangle.com — Nvidia acquisition talks, $2–3B valuation range, talent-driven acquisition characterization
- www.prnewswire.com — Jamba 1.5 release date (August 22, 2024), model specifications, performance claims, and cloud platform partnerships
- siliconangle.com — Maestro launch date, capabilities (task decomposition, inference scaling, quality verification), early access status