Almost all new startups supported by market leaders are implementing or using AI. This represents a very high barrier to entry and places high demands on the team.

AI integration has become virtually mandatory for startups. As of March 2026, 100% of Pre-Seed startups & 100% of Seed-stage startups either use or have announced the use of AI in their products.

Part 2. Analytics by Spheres

2.1. AI Infrastructure

This category includes startups developing the hardware, software, and platforms necessary to develop, train, and deploy artificial intelligence models and applications.

In February 2026, among top VC funds, the AI Infrastructure industry accounted for 17.4% of investments in terms of volume (USD 986 ml out of USD 5.6 bn) and 18.5% of deals (29 out of 157). At the same time, the sector is one of the most rapidly developing—we classified 18 out of 29 deals (62%) as Pre-Seed / Incubator, which shows that the industry is considered one of the most attractive.

Also in February, the industry saw two new unicorns: Render (syndicate lead: Georgian; among participants: Bessemer Venture Partners, General Catalyst) with a valuation of USD 1.4 billion, and Ayar Labs (syndicate leads: Nvidia, AMD; among participants: Insight Partners, Sequoia Capital) at USD 3.8 billion.

2.1.1. AI Development & Agent Frameworks

This subcategory includes startups developing software and applications aimed at the development, deployment, and monitoring of AI models and agents, including their stability and permission systems. This category includes 15 startups with a total venture funding volume of USD 319 million.

Among the Pre-Seed stage, we found YC-backed Carrot Labs and Sponge to be the most interesting.

• Carrot Labs is developing continuous-learning AI models for enterprises. The team's premise is that companies can create their own foundation models tailored to their specific tasks and workflows. Furthermore, these models should self-develop and optimize by taking into account their performance results and the company's efficiency metrics.
• Sponge is creating financial infrastructure to enable transactions between businesses and AI agents. Using these protocols and infrastructure, AI agents will be able to autonomously make purchases from real businesses and even participate in financial transactions.

As many as three projects are developing AI agents designed to create, support, and develop other AI agents:

• Portkey - Providing AI developer teams with the tools they need for production: AI Gateway, Observability, Guardrails, Governance, and Prompt Management. (USD 15 million. Syndicate lead: Elevation Capital; among participants: Light Speed Venture, link).
• Union - A unified AI development platform that streamlines the complex operations required to develop, deploy, and manage AI, ML, and agents. (USD 38.1 million. Syndicate lead: NEA, link).
• Braintrust - AI agents to help organizations efficiently test, monitor, and deploy reliable AI applications. (USD 80 million. Syndicate lead: ICONIQ; among participants: a16z, link).

Among Later-stage Startups

• Encord - The platform provides a data infrastructure to streamline the training data lifecycle for corporate AI models via automated data labeling, curation, post-training alignment, data management, and orchestration. (USD 60 million. Syndicate lead: Wellington Management; among syndicate participants: YC).
• Render reached unicorn status after raising USD 100 million (valuation: USD 1.4 billion). The platform is developing a cloud hosting platform for deploying apps and AI agents with zero ops costs and seamless scaling.

2.1.2. AI Gadgets

This month, we would like to highlight one startup: Button Computer (YC backed, link). They develop a wearable AI assistant that answers user questions when a button is pressed. It appears to be a highly convenient and effective gadget.

2.1.3. Applied, Interactive & Research AI

This group includes projects aimed at developing or upgrading AI models or approaches to their creation.

• Confluence Labs (YC backed, Pre-Seed, link) is already developing foundation models optimized for learning on limited data. The project notes that [????]. The startup estimates that its ARC-AGI-2 benchmark reaches 97.9% at a cost of ~USD 12, vastly outperforming other available models in quality.
• BeyondMath (USD 18 million investment. Syndicate lead: Cambridge Innovation Capital; among participants: Insight Partners, link) is creating specialized physics-informed neural networks to accelerate complex multiphysics simulations: engineering, critical tasks, aerospace, automotive, etc.
• Simile (USD 100 million. Syndicate lead: Index Ventures, link) is creating a simulation platform that uses generative agents to model human behavior and predict real-world decision-making.

2.1.4. Compute & Systems Optimization

This group includes hardware and algorithms specifically aimed at accelerating the performance of core AI or its applications.

Both projects backed in February (Piris Labs, backed by YC; Ayar Labs, backed by Nvidia and AMD, including Insight Partners and Sequoia Capital) were startups in the field of applied optical data transmission technologies.

• Piris Labs (link) is developing optical technologies to speed up communication data processing and remove bottlenecks.
• Ayar Labs (link), which became a unicorn, is the developer of optical engines to replace copper interconnects and build compute nodes that function as a single, giant GPU, improving AI hardware efficiency.

2.1.5. Data & Knowledge Infrastructure

This group includes infrastructure aimed at creating new datasets for model training or improving the stability and security of model operations through new tech solutions.

• Asimov (backed by