When I first saw the term "AI-native oracle", my first reaction was: here’s another marketing concept trying to ride the AI wave.
After all, everything is relying on AI now. Wallets want to create AI assistants, DEXs want to use AI for market making, even token launches claim to be AI-driven. This industry has been so ostentatious for a long time that it's really hard to believe who is genuinely putting in the work and who is just faking it.
But the collaboration between @APRO-Oracle and Nubila changed my mind.
First, let me introduce who Nubila is. This is a project specifically focused on "verifiable real-world data", aiming to feed the data that AI models need (not just prices, but also news, documents, images) to on-chain smart contracts in a trustworthy manner. It sounds a bit abstract, but let me give you an example to clarify.
Suppose you are creating a 'decentralized insurance' protocol on-chain, where users buy flight delay insurance. The traditional method is: when a flight is delayed, users submit proof, undergo manual review, and claim compensation. But this is too slow and contentious—if you say the delay was three hours, and the insurance company says it was only two hours and fifty minutes, who decides?
The approach of AI oracles is to automatically scrape data from airline websites, airport announcements, and flight tracking websites, using natural language processing technology to extract conclusions like 'Flight XX was delayed by 3 hours and 15 minutes,' and then put this conclusion on-chain, triggering smart contracts to automatically compensate.
The entire process requires no manual intervention, data sources are verifiable, and execution logic is transparent. This is the meaning of 'AI-native'—not just adding an AI label to traditional oracles, but designing around AI capabilities from the ground up.
The collaboration between APRO and Nubila aims to jointly develop such a system. APRO is responsible for data collection, aggregation, and blockchain integration, while Nubila handles data verification and trustworthy computing. The technological complementarity between the two companies is strong.
But the question arises: is there really market demand for this?
My answer is: not only is there demand, but it is one of the largest infrastructure gaps in the next three years.
Look at the hottest directions on-chain right now: RWA (Real World Assets), prediction markets, AI Agents, BTCfi. These things have one thing in common: they all require 'non-price data'.
What does RWA need? It needs audit reports, property certificates, and regulatory documents. For a tokenized property, you must prove that the property actually exists, ownership is clear, and valuation is reasonable. This information is not on CoinGecko, but on government websites and in audit company PDFs. Traditional oracles cannot access this data; AI oracles can.
What do prediction markets need? They need real-time event results. Who won the sports game, who was elected, whether a company’s earnings report is a profit or loss. This information is scattered across news websites, social media, and official announcements, with inconsistent formats and difficult to verify authenticity. AI oracles can extract key information from vast amounts of text and cross-verify credibility.
What does an AI Agent need? It needs multimodal data input. An automated trading AI Agent cannot just look at price K-lines; it also needs to consider news ('What did the Federal Reserve say today?'), on-chain data ('Are whales hoarding or selling?'), and social media ('What are KOLs calling for?'). The collection, cleaning, and structuring of this data all require AI capabilities.
What does BTCfi need? It needs cross-chain data from Bitcoin Layer 2, payment path data from the Lightning Network, and metadata from Ordinals. The complexity of this data far exceeds that of Ethereum, and traditional oracle data models are simply not applicable.
So you see, APRO's 'AI-native oracle' is not about showcasing technology; it is because the market genuinely needs infrastructure capable of handling complex data.
Moreover, and more critically, this demand cannot be met by Chainlink or Pyth.
Chainlink's architecture was designed 10 years ago, with the core logic being 'nodes fetch prices → aggregators calculate medians → put on-chain'. If you ask it to process PDF documents or parse news semantics, it cannot do so. Of course, Chainlink can change, but that would require a complete system overhaul, which is costly, and its main clients (DeFi blue chips) do not have such a strong demand for these new features, so the motivation for change is lacking.
As for Pyth, its advantage is ultra-low latency, but its data sources are 'first-party data' provided by exchanges and market makers, not 'third-party data' scraped from public networks. Asking Pyth to scrape flight delay information or parse audit reports is simply beyond its architectural support.
Thus, the collaboration between APRO and Nubila is essentially filling a market gap: possessing both AI processing capabilities and oracle on-chain capabilities while ensuring data verifiability.
What is the most challenging part of this? It's not the AI technology itself (as large models are already quite powerful now), but how to make on-chain contracts trust the conclusions provided by AI.
You see, if AI says 'this audit report shows the company has sufficient reserves', why should the smart contract trust it? AI could be hacked, fed false data, or the model itself could have bugs. The traditional solution is to have multiple AI nodes vote and reach a consensus. But this is not enough, because if everyone is using the same contaminated data source, the vote is useless.
APRO's solution is a three-layer architecture of 'multi-source aggregation + AI verification + manual review':
The first layer, scraping raw data from multiple independent sources (official websites, news media, regulatory platforms).
The second layer involves using AI models to process each data source separately, extracting key information and calculating credibility scores.
The third layer introduces a manual review mechanism for high-value scenarios (such as RWA settlements in the millions of dollars) as a final safeguard.
Although this design increases complexity, it greatly enhances security. Moreover, as AI models become increasingly accurate, the manual review in the third layer can gradually be reduced, ultimately achieving full automation.
Moreover, there is one point that many people have overlooked: the cost advantage of AI oracles.
You might think that processing data with AI is more expensive, requiring model runs, computing power, and storage. However, in certain scenarios, AI oracles can actually be cheaper.
For example, an RWA project needs to verify the reserve situation of 100 tokenized assets every day. The traditional approach is to hire an audit firm, charging $1,000 per asset, resulting in a daily cost of $100,000. In contrast, AI oracles can automatically scrape data and validate it, possibly costing only a few hundred dollars (mainly for computing and node fees).
This cost difference will become more apparent after scaling. When APRO's AI oracles handle thousands of verification tasks daily, the marginal cost approaches zero, whereas traditional auditing costs grow linearly.
This is why I say that AI oracles need to 'revolutionize' not traditional oracles like Chainlink and Pyth (which are still solid in price feeds), but centralized intermediaries like audit firms, data providers, and manual verification services.
Imagine a future RWA platform where all asset verification, compliance checks, and risk assessments are completed automatically by AI oracles. Audit reports are not generated every three months, but are updated in real time; data is not manually organized, but automatically scraped by AI; anomalies are not detected after the fact, but are warned in advance.
This efficiency improvement will completely change the cost structure and operational model of the RWA industry.
Of course, some people may worry: will AI oracles make mistakes? Is the data generated by AI trustworthy?
This is a reasonable concern. AI can indeed make mistakes, especially when dealing with ambiguous information (for example, a news report saying 'the company may face regulatory scrutiny'—is that 'may' a 50% probability or a 90% probability?).
However, APRO's design philosophy is: not to pursue 100% accuracy, but to pursue 'better than existing solutions'.
What are existing solutions? They involve manual audits, and auditors can also make mistakes; centralized data providers can be compromised; community voting can involve unprofessional voters. AI oracles do not need to be perfect; they only need to be more accurate, faster, cheaper, and more transparent than these solutions.
And as technology advances, the accuracy of AI will only keep increasing. Now, GPT-4 has already outperformed the human average in many tasks; what about in two years? It is possible that even professional auditors will be outperformed by AI.
Therefore, I believe that the collaboration between APRO and Nubila is not just to ride the wave but is genuinely addressing an impending market demand. When the fields of RWA, AI Agents, and prediction markets take off, whoever can provide reliable AI data services will control the pricing power of the infrastructure.
From this perspective, APRO's choice to bet on AI-native oracles at this moment is a very smart strategic decision. They are not competing head-on with Chainlink in the traditional DeFi market but are opening up a new battlefield—one that Chainlink cannot enter in the short term and one that Pyth does not even want to enter.
By the time the cake of this battlefield is enlarged, APRO will have become the de facto standard setter. At that time, others wanting to enter will find that APRO's technological accumulation, client cases, and data advantages have already formed an insurmountable moat.
This is the revolution that 'AI-native oracles' aim to bring: not to replace existing oracles, but to disrupt all reliance on manual and centralized data verification processes.
