A new market report by Bright Data titled “Data for AI: 2025 A Year in Review,” revealed a staggering acceleration in capital allocation toward data acquisition. In just the first 10 months of 2025, businesses increased their investment in data for AI by 262%.
This isn’t just about feeding Large Language Models (LLMs) more text. The report outlines a massive structural shift toward multimodal training and real-time agentic workflows, separating the true innovators from the rest of the pack.
The 61X Enterprise Advantage
The most striking divergence in the 2025 landscape is the widening gap between enterprise giants and small-to-medium enterprises (SMEs). According to the report, enterprises are now out-investing SMEs by a factor of 61X when it comes to data for AI.
What are they buying with this capital? They aren’t just scraping the surface; they are building deep, proprietary knowledge bases. The investment is pouring into three key areas that feed a “competitive edge”:
- Multimodal training data: High-definition video, images, and audio.
- Real-time pipelines: Essential for agents that need to act on live information.
- Custom data delivery infrastructure: Tailored systems to handle the massive influx of information.
Leading AI labs have radically increased their appetite for this multimodal data, recently peaking at 10 petabytes per day, equivalent to one million hours of 4K video daily.
The Rise of “Agentic” Retrieval
The buzzword of 2025 is clearly “Agentic AI”, autonomous agents that don’t just answer questions but perform tasks. To do this, they need a connection to the live web, not just a static archive of the past.
The report highlights a critical trend: Two-thirds of AI companies are now using a specific combination of data collection tools to power these agentic retrieval workflows.
This “dual mandate” of modern AI development, combining historical context with real-time access, has driven a 104% increase in the combined usage of SERP (Search Engine Results Page) APIs and Web Unlocker tools in just ten months.
Or Lenchner, CEO of Bright Data, summarizes the stakes: “The world’s leading AI companies are now racing to build proprietary knowledge bases to train models and enable agentic retrieval. That race is accelerating a dramatic surge in global data consumption”.
The Death of Manual Scraping
Perhaps the most dramatic statistic in the report is the 1,631% growth in usage from Web Scraper APIs between January and October 2025.
This four-digit surge signals a fundamental maturity in the market. The industry is shifting away from manual data engineering teams and toward API-first infrastructure. The report notes that tasks which previously took a team of six engineers three months to complete are now being handled by a single API call.
This efficiency is driving a massive spike in investment across specific toolsets:
- Web Scraper API: +1,631% (Driven by social media, e-commerce, and LLM-generated content tracking)
- Data (Static): +841% (The foundational knowledge for training models)
- SERP API: +734% (The connection to the “live web” for agentic search)
The Market Hierarchy
The report classifies the current data landscape into four distinct tiers:
- Tier 1: Giants (Google, AWS)
- Tier 2: Enterprise Specialists (Bright Data)
- Tier 3: Tech Challengers (Zyte, Apify)
- Tier 4: Specialized Innovators (Diffbot, NetNut)
Bright Data positions itself as a “stand-out leader” in Tier 2, citing recognition from Forbes for its massive 50% year-over-year revenue growth, currently exceeding $300M annually. Their “Web Unlocker” technology has become a critical asset for training data, boasting a 99.95% success rate against sophisticated anti-bot defenses.
Conclusion: The Era of Real-Time Intelligence
As we close 2025, the data suggests that the “training phase” of AI is evolving into the “acting phase.” The explosive growth in SERP APIs and real-time extraction tools proves that the next generation of AI models won’t just be learned scholars; they will be active participants in the live economy.
For enterprises, the message is clear: The moat is no longer just the model, it’s the infrastructure that feeds it.
