Anders Hammarbäck had watched enough AI companies stumble into courtrooms to know there was a better way. While Anthropic negotiated a $1.5 billion preliminary settlement with authors and OpenAI cut checks to News Corp, the former McKinsey consultant saw an opening: What if you could just... pay upfront?
It's not the kind of pitch that makes headlines at a product launch. But in September 2025, after a year operating quietly in Stockholm, RedpineAI went public with precisely that unglamorous proposition—a licensed data marketplace that promises AI builders what they need most right now: plausible deniability when regulators come knocking.
The company calls it "the Knowledge Layer for Agentic AI." Strip away the jargon and you get something simpler: a two-sided exchange where content owners rent their archives to AI developers without anyone ending up on a witness stand. Hammarbäck, the CEO, frames it almost defensively: "We never scrape data and always attribute back to the content owners."
In an industry built on the free-for-all scraping of the open web, that counts as radical.
The End of Free Data, Maybe
Timing, as they say in venture capital, is everything. RedpineAI's emergence tracks suspiciously well with a broader industry reckoning. Cloudflare rolled out "Pay-Per-Crawl" in mid-2025, letting websites charge AI bots or block them entirely. The Really Simple Licensing standard gave publishers a robots.txt equivalent for expressing AI terms. Getty Images and Perplexity signed a multi-year licensing pact. Even the Associated Press got into the game.
The message was clear, if belated: the era of treating the web as an all-you-can-eat buffet had ended. Or was ending. Or should end, depending on who you asked.
RedpineAI's backers certainly believe it. Colin M. Evans from OpenAI led a $1.3 million angel round alongside Feng Hong, co-founder of Xiaomi, and Gustav Lindqvist from Perplexity. Anna Nordell Westling, who co-founded Sana, and Daniel Langkilde from Kognic also put in money. A cluster of Spotify alumni invested through Greens. Peter Sarlin, CEO of Silo AI, joined shortly after launch.
It's the kind of investor lineup that suggests RedpineAI isn't so much predicting a shift as betting on one already underway.
Lindqvist told Swedish tech outlet Techtidningen that the startup addresses "the biggest bottleneck in the AI world." Evans emphasized compliance as regulation tightens globally. Whether either man would have said the same thing two years ago—when web scraping remained the industry's unspoken norm—is an open question.
What You Get for Your Money
RedpineAI positions itself as the anti-scraper, though that's partly marketing and partly necessity. The European regulatory environment doesn't leave much room for gray areas. The EU AI Act and GDPR loom over any data play coming out of Stockholm.
The platform brokers access to what the company describes as multi-modal datasets: text, images, audio, video, code, scientific journals, even motion capture footage and drone imagery. All of it supposedly cleared for AI use. At launch, RedpineAI claimed 100 billion-plus tokens of premium data already available—a number that sounds impressive until you remember frontier labs chew through multiples of that during pre-training.
Distribution happens through what the company calls an "agent-first API," meaning autonomous systems can query and pull data programmatically. It supports multiple stages of model development: pre-training, fine-tuning, reinforcement learning, evaluation, and retrieval-augmented generation. The technical details remain sparse—documentation isn't public—but the pitch is straightforward enough. If you're building an AI agent that needs domain-specific knowledge without legal exposure, RedpineAI wants to be your middleman.
The company emphasizes depth over breadth, a strategic choice that shows up in its hiring. Open roles include "GTM—Health" and "GTM—Legal," signaling vertical plays in healthcare and law. Robotics and finance round out the launch messaging. It's a practical approach, if not particularly sexy. Better to own a few categories well than promise everything to everyone.
Spotify Refugees and CERN Scientists
Hammarbäck comes out of McKinsey and later Antler, where he ran the Nordic region as partner and managing partner. His co-founder, David Österdahl, built products at Spotify and iZettle—Swedish pedigree through and through. The third core team member, Leonora Vesterbacka, holds a PhD from ETH Zürich and did time at CERN before leading KB-Whisper at Sweden's National Library, a project that earned her finalist recognition for "AI Swede of 2025."
It's the kind of founding team that looks good in pitch decks: enterprise experience, consumer product chops, academic credibility. Whether that translates to market traction is another matter entirely.
The company claims 2 to 10 employees on LinkedIn, which means it's either staying lean or still building out infrastructure. Probably both. Hiring is active across legal, go-to-market, and design—roles that suggest RedpineAI is assembling the scaffolding even as it courts customers.
Early Bets and Unanswered Questions
By October 2025, RedpineAI had launched pilots with what Swedish outlet Tidningen Näringslivet described as "world-leading AI companies" in the U.S. and Asia. The company hasn't named them publicly, a choice that feels equal parts strategic discretion and contractual obligation.
The first announced partnership came in January 2026 with AsedaSciences, a drug discovery platform. Under the agreement, RedpineAI's licensed scientific and clinical literature feeds into AsedaSciences' 3RnD system for pharmaceutical research and clinical analysis. The pilot kicked off the following month.
It's a narrow deployment—targeted, vertical, proof-of-concept—but one that demonstrates the business model in practice. Not data at scale, but data with domain relevance. Whether that's enough to build a sustainable business remains to be seen.
RedpineAI operates as a two-sided platform, which means it needs to solve two problems simultaneously. On one side: AI builders who need licensed data without legal risk. On the other: publishers, research institutions, and content owners looking to monetize archives that have been sitting idle or, worse, scraped without permission.
The company offers revenue-share partnerships with data owners, helping them "carve out" subsets of their content for AI use through what it describes as legal and technical controls. Pricing details remain private—the platform is invite-only during beta—but the model appears flexible. Older website copy referenced raw data downloads, API access, and embeddings as distribution options, though specifics are hard to pin down.
For buyers, compliance is framed as a feature rather than a burden. RedpineAI's terms make clear that sample data shown on the platform isn't licensed for use; formal agreements are required. It's a sensible precaution in an era when one wrong dataset can trigger years of litigation.
What's less clear is how RedpineAI plans to compete in an increasingly crowded space. Competitors like Defined.ai, DataSales.ai, and Dappier have built similar marketplaces. Wirestock now offers premium training datasets via AWS Marketplace. Dow Jones expanded its Factiva platform to roughly 5,000 publishers exploring AI licensing deals. The race, in other words, is already on.
The Licensing Inflection Point
Hammarbäck has written about this moment before. In a company blog post titled "Why Smart Data Licensing Will Define the Next Era of AI," he argued that moves like Cloudflare's pay-per-crawl model signal the end of freely available web data. Compensated access, he suggested, is the new default.
Perhaps. The industry certainly appears to be moving in that direction, though not uniformly and not without resistance. Scraping remains cheaper and easier for those willing to risk it, and enforcement of AI-specific licensing terms is still evolving. Legal precedents are being written in real time.
But if the pattern holds—if regulation tightens, if litigation costs keep climbing, if major publishers continue cutting exclusive deals—then platforms like RedpineAI start to look less like insurance and more like infrastructure. Not glamorous, exactly. But necessary.
Whether 100 billion tokens proves sufficient is another question entirely. For context, that's a fraction of what frontier labs consume during pre-training. GPT-4 reportedly trained on trillions. The real test will be quality, relevance, and the depth of domain-specific coverage—not just token count.
Still, RedpineAI's backers and early partnerships suggest confidence in the underlying bet. If scraping becomes untenable—legally, reputationally, financially—then someone needs to build the pipes for compensated access. Hammarbäck and his team are wagering they can be that someone.
The company remains largely opaque on key details: specific pricing models, the full roster of data suppliers, the mechanics of revenue-share agreements with content owners, the technical documentation for its agent-first API. All of it is closed to all but early-access invitees.
Which means RedpineAI is still, in many ways, a bet on what the AI industry will look like in two years rather than what it looks like today. Licensing may not be glamorous, as Hammarbäck might say. But it beats a subpoena.
