I Dug into 5 Crypto AI Projects That Raised Millions in Funding, and Found…

Author｜Golem (@ウェブ3_ゴーレム)

The AI frenzy has thrown VC firms primarily focused on Web3 into disarray. When Paradigm announced on February 28th its plan to raise a new fund of up to $1.5 billion targeting AI and robotics, I analyzed it as a signal that 暗号 capital was shifting its gaze from Web3 to AI. With a lack of promising projects in the 暗号 space, the booming AI sector has become the new battleground for crypto capital. (Related read: When Openclaw Founder Advises Young People to Stay Away from Crypto)

However, what I overlooked is that top-tier AI projects are out of reach for crypto VCs. Promising projects often choose investors based on their “pedigree” and resources. While a well-known crypto VC like Paradigm might squeeze into the first-tier AI investment circle, other small crypto VCs with no reputation in traditional finance can only watch from the sidelines as the big players feast.

So, is it just a waiting game? Of course not. Where there’s a will, there’s a way. Since they can’t invest in top AI projects, the next best thing is to retreat to their crypto comfort zone and find projects with some AI connection to invest in, buying themselves a ticket on the ship of the times.

But as an online joke goes, a Web3 company can transform into an AI-related startup simply by changing all its “loading…” messages to “thinking…”. Driven by crypto VC anxiety, some Crypto+AI projects can secure millions in funding with just a whitepaper and a product lacking Product-市場 Fit (PMF).

To make this “industry abstraction” more concrete, I’ve selected five recently funded “paper AI projects” as examples.

Derivio: A DEX That Isn’t a Meme 道具 is Not a Good AI Trading Platform

A fresh example is the AI-native trading platform Derivio, which announced a $6 million funding round on March 18th, backed by crypto VCs including YZiLabs. However, the official announcement included a fine print note: the $6 million represents the total funds raised to date, not the amount raised in this specific announcement.

It’s rare to see an ultra-early-stage project announce cumulative funding without specifying the new round amount. Is it because the VC funding was too small to mention? Or a PR move to coincide with its AI pivot? Research shows Derivio was part of Binance Labs’ (now YZiLabs) Season 6 Incubation Program in 2023, when it was still a decentralized derivatives exchange on zkSync.

In 2024, Derivio launched the Ethereum L2 Derivio Network, also claiming full compatibility with the Solana Virtual Machine (SVM), aiming to straddle both ecosystems. Unfortunately, or perhaps fortunately, it failed.

Today, if you visit Derivio’s website, you won’t see trading charts or the typical DEX swap interface. Instead, you’re greeted with a chain-scanning monitor reminiscent of Pump.fun.

Yes, you read that right. This former decentralized derivatives exchange, now an AI-native trading platform, primarily operates by helping “P-troopers” scan chains. A DEX that isn’t a Meme tool is not a good AI trading platform. Additionally, you can even buy tokenized US stocks on this platform.

So, as an AI-native trading platform, Derivio must have some AI-related features, right? But clicking on “Agent” in the top-left corner leads to a page that says “COMING SOON”…

Although Derivio currently has nothing to show, its described vision is imaginative. In an article titled “The Last Generation of Human Traders” published on X, it states that most trading terminals are designed for humans. Derivio is building the first full-stack trading terminal designed specifically for AI, and has independently developed a high-performance data streaming engine from scratch, compressing the latency from “on-chain event alert” to “front-end AI processing” to near the physical limit.

After reading it, I also thought this idea was fantastic, not something just anyone could do. So I ran it through a detector and found the article wasn’t written by a human at all—it was AI-generated. Derivio is indeed ahead of the curve, replacing its employees with AI before replacing human traders.

Superpower: The Vanished AI Agent Product

Superpower is an AI agent yield protocol. On March 6th, it announced the completion of a Pre-seed funding round with participation from Taisu Ventures, Paper Ventures, CatcherVC, and 280 Capital, though the specific amount was not disclosed. Superpower aims to build a platform for AI agents to autonomously generate revenue, gain funding access, and achieve capital appreciation.

To realize such an ambitious vision, significant technical accumulation and continuous development are necessary. However, Superpower hasn’t even finished its website. March 6th was not only the day they announced funding but also the day their official X account posted its first tweet.

The first post from Superpower’s official X account

While Derivio lacks an AI Agent, at least it has something to interact with. But if you unfortunately open Superpower’s official website, your vision will be severely assaulted. The website has no clickable buttons, only a string of “YOUR AGENT IS BROKE AF” playing on a slideshow. I don’t even understand what that means, and because the slides change too fast, I couldn’t tell if the last word was “AI” or “AF”.

In the links on Superpower’s official X account, I also found a prediction market project called Prolly, which described how agents could make money on it. I wanted to try it, but the product requires an invitation code, so I had to give up. Of course, the introductory post on Superpower’s homepage was, without exception, also written by AI.

finrob: The Unusable Cryptocurrency-Specific Model

Finrob is an AI-driven crypto market research platform. On February 25th, it completed a $3.9 million seed round with participation from Maven11, Placeholder, Archetype, Fabric Ventures, Dispersion Capital, and Node Capital. What does this project do? Simply put, it’s a conversational large language model, with an interaction method no different from ChatGPT or Gemini. In fact, Finrob is built on top of these very models.

When asked about the difference between using Finrob and directly conversing with ChatGPT or Gemini, Finrob’s answer is that it’s specifically built for the cryptocurrency field, featuring real-time data integration, on-chain analysis, and specialized tools. Specifically, it integrates with CoinGecko for real-time prices and market data, Glassnode for over 200 types of on-chain analysis, Tavily for web search and news, Perplexity for in-depth research, and other sources including DefiLlama, Etherscan, and LunarCrush for social sentiment.

In other words, after Finrob connects these free large models to these data sources (Finrob doesn’t support users choosing advanced models like GPT-4), it can claim to be specifically built for the crypto field, and that’s worth $3.9 million. From another perspective, this is indeed much cheaper than spending money to train a large model.

Judging from the use cases shown officially, Finrob’s ultimate purpose seems to be providing investment advice to users. Leaving aside whether AI can truly ガイド real trading decisions, is the intelligence of Finrob, using free models, really higher than that of GPT-4, Claude Opus, etc.? I initially thought Finrob might be better than ChatGPT at fetching real-time token prices, but after testing, I found that with a simple prompt, ChatGPT can also fetch BTC’s real-time price from CoinGecko.

Despite all these criticisms, I actually wanted to try Finrob myself. But it didn’t give me the chance. Whether I tried logging in with an email or a wallet, the page showed an error.

PlutonAI: The DeFAI in the Corner

PlutonAI is a DeFAI platform aiming to enable AI agents to analyze markets, optimize strategies, manage yield opportunities, and perform complex on-chain operations on behalf of users. On February 17th, PlutonAI completed a $2.7 million private funding round led by kitchenvc, with participation from HyperGPT.

Judging by its positioning, PlutonAI is a standard Crypto+AI project. Regardless of whether DeFAI has PMF, is safe, or can actually help users make money, in today’s rapidly evolving AI landscape, having AI perform on-chain operations for humans is already quite straightforward.

Especially after the OpenClaw craze, the crypto community was swept by a wave of “raising lobsters.” Countless bloggers have shared how to install OpenClaw and configure it to participate in on-chain trading or prediction markets. Meanwhile, exchanges like Binance and OKX have also launched similar AI agent trading tools/assistants. (Related read: Amid the OpenClaw Frenzy, CEXs Vie for AI Agent Trading Entry Points)

Therefore, one could even argue that DeFAI is already a pseudo-proposition. I also wanted to try the AI Agent on PlutonAI, but again, I couldn’t log in.

Unicity: The True Master of Painting Pies

Unicity is an infrastructure developer dedicated to building the “Agentic Autonomous Internet,” dreaming of enabling billions of AI agents to engage in trustless discovery, trading, and settlement at machine speed. On February 19th, Unicity completed a $3 million seed round led by Blockchange, with participation from Outlier Ventures and Tawasal.

This is another company with grand ambitions. To understand what they actually plan to do, I specifically read their project whitepaper (finally, not written by AI!). In summary, Unicity believes existing blockchains are unsuitable for AI’s high-frequency automated trading and collaboration due to throughput, latency, privacy, and cost issues. Therefore, they aim to build an underlying network for AI agents. The specific solution involves moving transactions off-chain for peer-to-peer completion, with the blockchain only confirming state changes to prevent “double-spending,” paired with verifiable agents responsible for supervising execution.

Compared to the previously listed projects, Unicity sounds much more serious and could potentially form a genuine public chain ecosystem in the future. But what has Unicity, founded in 2025, actually delivered so far? No testnet, zero ecosystem projects. In a project update on March 10th, they mentioned plans to launch the agent operating system AstridOS, allowing tools like Claude Code and OpenClaw to run on it. But whether anyone will use it is another matter.

Are Crypto VCs Stupid?

Of course, among the recently funded AI projects, there are solid ones. For example, RoboForce, which completed a $52 million funding round on March 17th led by YZi Labs, is an AI robotics company with no relation to the crypto field. The project is building physical AI, a sector highlighted by Jensen Huang as a future direction during his GTC 2026 keynote. Furthermore, RoboForce’s robots were indeed showcased at NVIDIA GTC 2025.

There are also projects like Kled, an AI data 市場 that raised $5.5 million on March 11th, and VeryAI, focused on building an AI Agent identity system, which raised $10 million on March 12th. These not only have product-market fit but also possess real product ecosystems and track records of delivery.

But such projects are rare gems in the crypto industry. So the question arises: are other crypto VCs stupid? Knowing full well that many so-called AI projects lack a validated product logic, why are they still willing to pour money in?

The answer lies in the question itself. What crypto VCs want is simply the liquidity associated with the letters “A” and “I.” Although they can’t get a seat at the top-tier project table to share the chips, the money still needs to be deployed. Therefore, creating some “AI avatars” within the crypto context isn’t a bad business. Project teams only need to package themselves as AI, and VCs can blindly stuff money in. As long as both sides can eventually recoup their investment, it’s good AI.

この記事はインターネットから得たものです。 I Dug into 5 Crypto AI Projects That Raised Millions in Funding, and Found…

Related: Skills How to Usher in the Era of AI “Specialization and Division of Labor”?

In the past, if you wanted AI to help you write a company weekly report, you needed to input a long instruction (Prompt) every time: “Help me summarize this week’s progress, use size 2 font for the title, follow last year’s template for formatting, remember to pull data from this spreadsheet…” Even then, AI would occasionally miscalculate numbers or mess up the formatting. Now, the emergence of Claude Skills has made this inefficient conversation a thing of the past. 1. What are Skills? From “Teaching Every Time” to “Ready to Work” Simply put, Skills are a set of standardized toolkits that instantly turn AI into a “domain expert.” It’s no longer just a sentence in a chat box, but a tangible smart folder. To help you write that weekly report,…