Periodic table of AI startups founded and funded between February 2025 and February 2026
Cristina Iani from Veridion classified 305 artificial intelligence startups founded or funded between February 2025 and February 2026. Each of the 14 categories is characterized by five parameters: total funding, number of startups, annual growth, momentum trend, and ecosystem level.
Based on this classification, I created an interactive dashboard in Tableau. How to use it:
Based on this classification, I created an interactive dashboard in Tableau. How to use it:
- Top left: ranking by investment volume
- Top right: market trend
- Bottom right: annual growth
- Center: category code and name. Hovering shows example companies and market facts
- Bottom figures and charts: investment volume and number of companies in the category
Key observations from the classification's author
Foundation models still receive the largest share of capital. $80 billion is spread across 43 startups, almost double the combined total of all other categories. But growth is happening in other categories.
AI for developers is the fastest-growing category, with annual growth of +320%. Developers are building tools for developers, and venture investors are "pouring fuel on the fire." AI agents (+260%) and vertical SaaS (+250%) aren't far behind.
For the first time, vertical startups have quietly overtaken horizontal ones in terms of funding volume: healthcare, legal, fintech, drug development. Real-world adoption is happening in domain-specific AI. The era of "AI for everything" is giving way to the era of "AI for a specific task."
Emerging startups are forming the next wave of growth. Security and compliance, autonomy and robotics, voice AI and chatbots: 48 startups, $8.8 billion, and some of the highest growth rates.
AI for developers is the fastest-growing category, with annual growth of +320%. Developers are building tools for developers, and venture investors are "pouring fuel on the fire." AI agents (+260%) and vertical SaaS (+250%) aren't far behind.
For the first time, vertical startups have quietly overtaken horizontal ones in terms of funding volume: healthcare, legal, fintech, drug development. Real-world adoption is happening in domain-specific AI. The era of "AI for everything" is giving way to the era of "AI for a specific task."
Emerging startups are forming the next wave of growth. Security and compliance, autonomy and robotics, voice AI and chatbots: 48 startups, $8.8 billion, and some of the highest growth rates.
My analytical context by category
Foundation models and LLMs
The $40 billion OpenAI funding deal in a single investment round was announced in March 2025 and closed by the end of the year. The lead investor was the Japanese conglomerate SoftBank, which provided about $30 billion of the total amount. This round pushed OpenAI's valuation to $300 billion.
As of February 2026, OpenAI is already in the process of closing its next round, which could exceed $100 billion at a valuation of up to $850 billion. The funds are earmarked for building out compute infrastructure (the Stargate project) and achieving AGI (a point at which a machine can perform any intellectual task at or above human level).
As of February 2026, OpenAI is already in the process of closing its next round, which could exceed $100 billion at a valuation of up to $850 billion. The funds are earmarked for building out compute infrastructure (the Stargate project) and achieving AGI (a point at which a machine can perform any intellectual task at or above human level).
AI infrastructure
The "picks and shovels" idiom has never been more relevant: while application-layer startups fight for survival and user loyalty, compute providers are capturing the lion's share of budgets. 80% of AI venture capital flows into the pockets of infrastructure providers. This has given rise to "compute bartering," where investments are made not in cash but in compute capacity.
The chip shortage has turned semiconductors into "the new oil." Companies like NVIDIA, AMD, and startups building specialized tensor processors are setting the pace for the entire industry. Without access to H200 or Blackwell GPUs, it's impossible to train a competitive model.
As the market matures, the focus has shifted from "training" to "operating." Startups building tools for monitoring, security, and scaling models (Weights & Biases or Anyscale) have become a critical link in the chain. Without a quality serving and maintenance system, a "raw" model never turns into a reliable business product.
The chip shortage has turned semiconductors into "the new oil." Companies like NVIDIA, AMD, and startups building specialized tensor processors are setting the pace for the entire industry. Without access to H200 or Blackwell GPUs, it's impossible to train a competitive model.
As the market matures, the focus has shifted from "training" to "operating." Startups building tools for monitoring, security, and scaling models (Weights & Biases or Anyscale) have become a critical link in the chain. Without a quality serving and maintenance system, a "raw" model never turns into a reliable business product.
AI agents and assistants
Over the past 12 months, the autonomous agents category has posted the most aggressive growth in software history. This came from a shift from "deliberative AI" (chatbots) to "executive AI." The market tripled in size, partly because companies began replacing classic software subscriptions (SaaS) with outcome-based pricing.
Copilots remained in the role of "second pilot," assisting a human within an interface (Microsoft Copilot or GitHub Copilot). AI agents became autonomous entities that plan tasks, use tools, and interact with other agents on their own, without human involvement (AutoGPT integrations and enterprise agents from ServiceNow).
The "Employee-as-a-Service" (EaaS) economy: agents began integrating en masse into HR and finance systems. Instead of buying a CRM license, companies are hiring "digital sales reps" that work 24/7. As a result, the vertical AI sector has started absorbing budgets that previously went toward outsourced contractors.
The market's 3x growth is also driven by the rise of "agentic platforms" (LangGraph and CrewAI). They let businesses assemble complex chains of actions from simple models, making AI agents accessible not just to Big Tech but to mid-sized businesses as well.
Copilots remained in the role of "second pilot," assisting a human within an interface (Microsoft Copilot or GitHub Copilot). AI agents became autonomous entities that plan tasks, use tools, and interact with other agents on their own, without human involvement (AutoGPT integrations and enterprise agents from ServiceNow).
The "Employee-as-a-Service" (EaaS) economy: agents began integrating en masse into HR and finance systems. Instead of buying a CRM license, companies are hiring "digital sales reps" that work 24/7. As a result, the vertical AI sector has started absorbing budgets that previously went toward outsourced contractors.
The market's 3x growth is also driven by the rise of "agentic platforms" (LangGraph and CrewAI). They let businesses assemble complex chains of actions from simple models, making AI agents accessible not just to Big Tech but to mid-sized businesses as well.
Enterprise software
Control over AI budgets is starting to shift from central IT departments to the heads of functional business units. Marketing, sales, HR, and logistics departments have begun purchasing specialized solutions directly, bypassing the lengthy approval cycles of a company-wide AI strategy. Companies have started embedding AI into critical business processes: from automated financial period close to predictive supply chain management.
According to reports from Gartner and IDC, deploying specialized AI in customer service departments has cut staffing costs by 30% while maintaining satisfaction levels.
Companies are dropping Salesforce or Oracle licenses in favor of more flexible AI-native platforms (Glean or Writer) that offer built-in automation out of the box, rather than just data storage.
The "patchwork" of different AI tools within a single company has given rise to a new segment within enterprise software: "AI orchestration." It helps unify disparate solutions from different departments into a single, secure ecosystem.
According to reports from Gartner and IDC, deploying specialized AI in customer service departments has cut staffing costs by 30% while maintaining satisfaction levels.
Companies are dropping Salesforce or Oracle licenses in favor of more flexible AI-native platforms (Glean or Writer) that offer built-in automation out of the box, rather than just data storage.
The "patchwork" of different AI tools within a single company has given rise to a new segment within enterprise software: "AI orchestration." It helps unify disparate solutions from different departments into a single, secure ecosystem.
AI for developers
The term "vibe coding" has become the official name for a new era of programming. In this approach, a developer no longer writes program code but describes the product's "vibe" - its logic, interface, and user experience - in natural language. The programmer turns into a "conductor" or "editor" of code.
The barrier to entry for building software has fallen to a historic low. Now, any product manager or designer can "write an app," which has led to an explosive increase in the number of internal corporate tools.
Small teams of 1-2 people are building products that once required teams of 20 engineers. This is radically reshaping the economics of tech startups: developer labor costs are no longer the main expense line, giving way to spending on tokens and AI infrastructure.
The barrier to entry for building software has fallen to a historic low. Now, any product manager or designer can "write an app," which has led to an explosive increase in the number of internal corporate tools.
Small teams of 1-2 people are building products that once required teams of 20 engineers. This is radically reshaping the economics of tech startups: developer labor costs are no longer the main expense line, giving way to spending on tokens and AI infrastructure.
Content generation
Generative video is moving out of the "social media toy" category and into professional filmmaking. Testing of Runway's tools by giants like Netflix and Disney confirms that generation quality (character consistency, motion physics, and 8K resolution) has reached a level suitable for final cut.
Disney and Netflix's primary interest is centered on cutting visual effects costs. Runway's AI tools can do in hours work that used to take weeks: background touch-ups, building complex 3D scenes from a text description, and automatic rotoscoping (cutting out objects). In addition, directors now "run through" entire scenes using generative models before shooting even begins, which allows for radical optimization of location-shoot and casting budgets.
Despite the technological breakthrough, adoption at Disney and Netflix is happening under strict copyright controls. Runway and other category leaders (Pika, Sora) have introduced "commercially safe" models trained only on licensed content.
Analysts at Gartner and Variety predict that the next step will be "dynamic editing," where AI tailors environmental details or a film's pacing to an individual viewer's preferences.
Disney and Netflix's primary interest is centered on cutting visual effects costs. Runway's AI tools can do in hours work that used to take weeks: background touch-ups, building complex 3D scenes from a text description, and automatic rotoscoping (cutting out objects). In addition, directors now "run through" entire scenes using generative models before shooting even begins, which allows for radical optimization of location-shoot and casting budgets.
Despite the technological breakthrough, adoption at Disney and Netflix is happening under strict copyright controls. Runway and other category leaders (Pika, Sora) have introduced "commercially safe" models trained only on licensed content.
Analysts at Gartner and Variety predict that the next step will be "dynamic editing," where AI tailors environmental details or a film's pacing to an individual viewer's preferences.
AI in healthcare
The pharmaceutical industry has crossed the "valley of death" of traditional drug development. Where the path from molecule discovery to market once took more than 10 years and cost billions of dollars, the use of generative chemistry and digital twins has cut that cycle down to 2-3 years. The first drugs fully designed by AI (from Insilico Medicine and Isomorphic Labs) are already in the final stages of clinical trials, with record efficacy rates.
There has been a shift from "physician decision support" toward autonomous diagnostic systems. AI models have learned to analyze multimodal data (MRI scans, genetic code, blood tests, and wearable device data) and detect hidden pathologies (early-stage cancer or Alzheimer's disease) before the first symptoms appear.
There has been a shift from "physician decision support" toward autonomous diagnostic systems. AI models have learned to analyze multimodal data (MRI scans, genetic code, blood tests, and wearable device data) and detect hidden pathologies (early-stage cancer or Alzheimer's disease) before the first symptoms appear.
Science
Periodic Labs' record seed round ($300 million) confirms venture investors' belief in building full-fledged AI scientists. These are specialized agents that independently formulate scientific hypotheses, plan experiments, and interpret results around the clock.
Science is turning from a process of human "intuitive guesswork" into a scalable technological process. Periodic Labs' AI scientists are capable of running millions of virtual simulations in materials science, energy, and genetics - and selecting only the most promising candidates for physical testing in labs.
Periodic Labs' technology is integrated with robotic lab benches. AI designs the experiment, a robot runs it, AI analyzes the result and refines the theory. The human becomes an "architect of goals" rather than an executor.
Science is turning from a process of human "intuitive guesswork" into a scalable technological process. Periodic Labs' AI scientists are capable of running millions of virtual simulations in materials science, energy, and genetics - and selecting only the most promising candidates for physical testing in labs.
Periodic Labs' technology is integrated with robotic lab benches. AI designs the experiment, a robot runs it, AI analyzes the result and refines the theory. The human becomes an "architect of goals" rather than an executor.
Financial services
The financial sector's saturation with AI agents is explained by the fact that financial processes are highly formalized and digitized, making them an ideal testing ground for deploying autonomous systems. AI agents independently perform audits, detect money laundering in real time, and manage liquidity.
A large number of startups are competing for the same bank and hedge fund budgets. This density of players is driven by the category's high ROI. Deploying AI in credit scoring and wealth management has allowed market leaders to cut operating costs by 40%. Startups that can't offer a multiple improvement in these metrics are quickly absorbed by giants as the market consolidates. Major fintech platforms (Stripe or Adyen) are buying up narrowly specialized agents to build unified "operating systems for money."
A large number of startups are competing for the same bank and hedge fund budgets. This density of players is driven by the category's high ROI. Deploying AI in credit scoring and wealth management has allowed market leaders to cut operating costs by 40%. Startups that can't offer a multiple improvement in these metrics are quickly absorbed by giants as the market consolidates. Major fintech platforms (Stripe or Adyen) are buying up narrowly specialized agents to build unified "operating systems for money."
AI in legal
Legal AI is no longer just a document search tool. It has become a full operating system that independently drafts legal filings, spots hidden risks in contracts, and proposes defense strategies based on analysis of case law.
Harvey's success demonstrates the viability of the vertical AI model. Training models on closed (proprietary) legal data has created an insurmountable barrier for general-purpose models like base ChatGPT. Companies are willing to pay a premium price for the absence of "hallucinations" in legal terminology.
Legal giants are being forced to shift to outcome-based pricing or fixed annual subscriptions, since AI can do a junior lawyer's work in seconds rather than 20 hours. However, the trust of law firm partners remains the decisive factor, despite the product's technological sophistication.
Harvey's success demonstrates the viability of the vertical AI model. Training models on closed (proprietary) legal data has created an insurmountable barrier for general-purpose models like base ChatGPT. Companies are willing to pay a premium price for the absence of "hallucinations" in legal terminology.
Legal giants are being forced to shift to outcome-based pricing or fixed annual subscriptions, since AI can do a junior lawyer's work in seconds rather than 20 hours. However, the trust of law firm partners remains the decisive factor, despite the product's technological sophistication.
Vertical SaaS
Modern SaaS solutions are being built around neural networks. What's being created isn't just a "database for dentists," but an autonomous system that maintains the patient chart itself, produces a preliminary diagnosis from an image, and automatically orders supplies.
AI aggregators are buying up profitable but technically outdated businesses in narrow niches (cleaning, repair, local logistics) and, within 3-6 months, rolling out vertical SaaS with AI agents inside them. This cuts administrative staff and boosts business margins by 20-30%.
Vertical solutions outperform horizontal ones thanks to proprietary data. A model trained on 10 million industry-specific documents (construction estimates or agricultural maps) makes 10 times fewer errors than general-purpose models.
Instead of the usual "per-user" pricing, vertical AI is shifting to outcome-based pricing. The customer pays not for access to the software, but for a successfully processed application, a closed deal, or a liter of fuel saved.
AI aggregators are buying up profitable but technically outdated businesses in narrow niches (cleaning, repair, local logistics) and, within 3-6 months, rolling out vertical SaaS with AI agents inside them. This cuts administrative staff and boosts business margins by 20-30%.
Vertical solutions outperform horizontal ones thanks to proprietary data. A model trained on 10 million industry-specific documents (construction estimates or agricultural maps) makes 10 times fewer errors than general-purpose models.
Instead of the usual "per-user" pricing, vertical AI is shifting to outcome-based pricing. The customer pays not for access to the software, but for a successfully processed application, a closed deal, or a liter of fuel saved.
Autonomy and robots
Prometheus's launch-stage funding ($6.2 billion) confirms that building "brains" for autonomous systems has become the most expensive game in the world. This is an attempt to create a single foundation model capable of controlling any mechanism.
Prometheus and its competitors (Figure and Tesla Optimus) have moved from programming robots to "training them through observation." Prometheus's robots learn to perform tasks simply by watching millions of hours of video of human labor.
With this level of funding, Prometheus plans not just to sell robots, but to offer an "autonomous workforce" by subscription. This is aimed at addressing the global shortage of frontline labor.
Prometheus and its competitors (Figure and Tesla Optimus) have moved from programming robots to "training them through observation." Prometheus's robots learn to perform tasks simply by watching millions of hours of video of human labor.
With this level of funding, Prometheus plans not just to sell robots, but to offer an "autonomous workforce" by subscription. This is aimed at addressing the global shortage of frontline labor.
Security and compliance
Having advanced protection systems is not a competitive advantage - it's the bare minimum condition for a startup's survival. Large corporations won't consider buying AI solutions unless they meet strict protocols against model hacking and data leaks.
Bad actors have learned to "poison" training datasets and bypass corporate filters. In response, Armis and its competitors have moved to building autonomous defense systems that monitor, in real time, the behavior of neural networks and connected AI devices (sensors, robots, drones).
Compliance startups offer automatic checks of every model iteration for adherence to ethical standards, privacy laws, and freedom from bias.
Bad actors have learned to "poison" training datasets and bypass corporate filters. In response, Armis and its competitors have moved to building autonomous defense systems that monitor, in real time, the behavior of neural networks and connected AI devices (sensors, robots, drones).
Compliance startups offer automatic checks of every model iteration for adherence to ethical standards, privacy laws, and freedom from bias.
Voice AI and chatbots
Wonderful's technological breakthrough lies in its models' ability to recognize cultural context, sarcasm, and the emotional state of the person they're talking to in real time. AI agents don't just output text - they adapt tone and timbre, which has made it possible to deploy them in sensitive areas such as psychological support and VIP concierge service.
Voice agents have become autonomous: they can independently book tickets, carry out banking transactions, and modify orders in CRM systems, using voice as a full-fledged management tool.
The cost of one minute of conversation with an AI agent has become 15 times lower than the cost of a human operator, while customer satisfaction in pilot projects came in 20% higher.
Analysis of the dashboard and analytical context confirms a global trend: investment is shifting from foundation models toward specialized solutions with a measurable business impact. The advantage goes to vertical startups with their own proprietary data, rather than simply access to the most advanced foundation model.
Voice agents have become autonomous: they can independently book tickets, carry out banking transactions, and modify orders in CRM systems, using voice as a full-fledged management tool.
The cost of one minute of conversation with an AI agent has become 15 times lower than the cost of a human operator, while customer satisfaction in pilot projects came in 20% higher.
Analysis of the dashboard and analytical context confirms a global trend: investment is shifting from foundation models toward specialized solutions with a measurable business impact. The advantage goes to vertical startups with their own proprietary data, rather than simply access to the most advanced foundation model.
See also
Who bought U.S. oil & products in 2025 - how that may change in 2026
A detailed analysis of U.S. Energy Information Administration data on U.S. exports of crude oil and petroleum products, broken down by country and region.
Funding and outcomes analysis of 100,000 startups
A study of model data for predicting IPO, acquisition, or shutdown outcomes. The approach used are applicable to identifying the true drivers of startup success.
25-year risk-return analysis of investment portfolios
Risk and return are directly related: the higher an asset's potential profit, the higher the probability of financial loss. Safe instruments deliver minimal returns.
The materials presented in this section are not individual investment recommendations. The financial instruments mentioned in this section may not be suitable for you and may not correspond to your investment profile, financial situation, investment experience, knowledge, investment objectives, or attitude toward risk and return.
This information does not constitute a public offer, proposal, or invitation to invest in funds and/or strategies, or to buy or sell securities, or to enter into transactions with them.
Past investment performance does not determine future returns. This is not an advertisement for securities. Before making an investment decision, the investor must independently assess the economic risks and benefits, as well as the tax, legal, and accounting consequences of the transaction, and their willingness and ability to accept such risks.
© 2016—2026 Invest Adviser
This information does not constitute a public offer, proposal, or invitation to invest in funds and/or strategies, or to buy or sell securities, or to enter into transactions with them.
Past investment performance does not determine future returns. This is not an advertisement for securities. Before making an investment decision, the investor must independently assess the economic risks and benefits, as well as the tax, legal, and accounting consequences of the transaction, and their willingness and ability to accept such risks.
© 2016—2026 Invest Adviser
Manage cookies
Cookie Settings
Cookies necessary for the correct operation of the site are always enabled.
Other cookies are configurable.
Other cookies are configurable.