18 months. That's how old the company was when SAP bought it whole

On July 17, SAP announced it had officially completed its acquisition of German AI startup Prior Labs. Here's the wild part: the company was only founded in 2024. By the time the deal closed, Prior Labs was barely 18 months old. Europe's largest software company just swallowed a lab that's been around for a year and a half.

And it's not stopping at the acquisition. SAP committed to investing more than €1 billion over the next four years to scale Prior Labs into what it calls "a globally leading frontier AI research lab in Europe." SAP didn't officially disclose the purchase price itself, but TechCrunch reported the deal value at roughly $1.16 billion, and European startup outlets framed it as a "€1 billion exit." That's serious money riding on a single 18-month-old startup.

The key is what this company actually builds. Prior Labs does not build large language models like ChatGPT. Instead, it pioneered foundation models for structured (tabular) data — the rows and columns living in your spreadsheets, database tables, and ERP ledgers. SAP CTO Philipp Herzig summed up the whole thesis of this deal:

"Early on, SAP recognized that the greatest untapped opportunity in enterprise AI wasn't large language models; it was AI built for the structured data that runs the world's businesses."

Who is Prior Labs, and what is TabPFN?

Prior Labs was founded in 2024 in Germany, headquartered in Freiburg, by three cofounders: CEO Frank Hutter, CTO Noah Hollmann, and Sauraj Gambhir. Frank Hutter in particular is a globally recognized researcher in AutoML (automated machine learning). The company raised around €9 million in early funding in early 2024 — and then exited to SAP just 18 months later. As venture stories go, that's one of the fastest returns you'll see.

Its flagship is TabPFN. The name looks cryptic but breaks down cleanly: "Tab" for table, "PFN" for Prior-data Fitted Network. In plain English, it's a pre-trained foundation model built for table data. Where an LLM handles sentences and words, TabPFN ingests entire rows-and-columns tables and spits out predictions. It's open source, it has been downloaded over 3 million times, and it has effectively become a standard tool in the tabular-AI ecosystem.

The performance holds up too. According to Prior Labs, the latest TabPFN-2.6 ranks first on TabArena, the leading benchmark for tabular foundation models (TFMs). The more striking part is speed. Traditionally, building a prediction model from table data means running an automated machine learning pipeline — preprocessing, feature selection, hyperparameter tuning — for hours. TabPFN-2.6 reportedly matches the accuracy of a four-hour AutoML pipeline instantly, in a single model. On top of that sits a reasoning-focused variant, TabPFN-3-Thinking.

The credibility is stacked, too. Prior Labs' scientific advisory board includes Yann LeCun, Turing Award winner and a godfather of deep learning, and Bernhard Schölkopf, director of the Max Planck Institute for Intelligent Systems and president of the European AI research network ELLIS. This is a startup with serious academic gravity behind it.

The deal terms, dollars, and structure

Here's the structure of the deal based only on what's been disclosed. The purchase price is officially undisclosed, but the investment commitment and timeline are quite concrete.

Item What's disclosed
Deal announced May 4, 2026 (SAP official)
Deal completed July 17, 2026
Prior Labs founded 2024 (~18 months at close)
Investment pledge Over €1 billion across 4 years (to scale the lab)
Reported deal value ~$1.16 billion (TechCrunch); "€1B-class exit" (EU outlets)
Operating model Continues as an independent entity within SAP
Headquarters Freiburg, Germany
Core assets TabPFN (open source, 3M+ downloads), TabPFN-2.6, TabPFN-3-Thinking
Advisory board Yann LeCun, Bernhard Schölkopf
SAP tie-in Accelerates SAP's own SAP-RPT-1 model

Two things worth flagging here. First, the "€1 billion+" figure is not the purchase price — it's a four-year investment commitment. SAP never officially disclosed what it paid to buy the company; the $1.16 billion number comes from TechCrunch reporting. So separate from whatever SAP spent to acquire the lab, it's pledging to pour another €1 billion-plus over four years to grow it into a frontier-grade operation. The emphasis is on nurturing, not just buying.

Second, Prior Labs isn't being absorbed into SAP — it stays an independent entity. That matters a lot. When a giant buys a startup, the usual outcome is that it gets dissolved into the org and loses its identity. SAP chose to preserve research autonomy while bolting on everything it owns. Prior Labs gets access to SAP's vast enterprise data environment, its network of hundreds of thousands of customers, and a path to productization. Hutter's own words hit exactly that note: "Joining the SAP family gives us the resources, data environment and customer reach to take this category to its full potential."

What each side gets — for SAP, for Prior Labs

For SAP, three wins are clear. First, claiming enterprise AI's blind spot. For the past three years, AI investment has piled into LLMs. But most of the data enterprises actually run on isn't sentences — it's tables. Revenue ledgers, inventory tables, supply-chain data, churn metrics: all structured. SAP saw this entire market sitting empty and acquired the team that pioneered it. Second, accelerating its own model. SAP already had its own tabular foundation model, SAP-RPT-1, and adding one of the world's best research teams doubles down on that direction. Third, a killer weapon for its Business AI strategy. SAP is already pushing its Joule assistant; bolt on the ability to predict business outcomes — payment delays, supplier risk, customer churn, demand forecasting — and you get a moat rival ERPs will struggle to cross.

For Prior Labs, this is about the best landing a startup could dream of. Tabular-data AI is brilliant tech that hits a wall: "So where's the data and where are the customers?" Unlike LLMs, it's hard B2B infrastructure that won't produce a consumer hit. But SAP holds the ERP ledgers of large enterprises worldwide. The moment you attach yourself to the organization that touches more structured business data than anyone on Earth, your models get validated on real production data and get a straight path to becoming products. You keep research autonomy and get €1 billion of ammunition over four years on top.

The shared upside is the banner of European AI sovereignty. In a world where America's OpenAI, Anthropic, and Google monopolize frontier AI, the story of a European company building a world-class AI lab on European soil (Freiburg) is politically potent. SAP leaned into it explicitly with the "globally leading frontier AI lab in Europe" framing. It's a tech acquisition and a symbolic asset for European industrial policy at once.

Precedents — the wins and the failures

Big companies buying AI talent and tech wholesale is a well-worn playbook, and the outcomes have split sharply. To judge this deal, look at the precedents.

Win — Google × DeepMind (2014, ~$500M). Google bought the four-year-old British AI lab and honored the condition that it stay an independent research org in London. The result was a run of world-leading research from AlphaGo to AlphaFold. Lesson: preserve an acquired AI lab's research autonomy and, long-term, you get explosive output. SAP keeping Prior Labs as an independent entity is a clear nod to this model.

Win — Microsoft × GitHub (2018, $7.5B). Microsoft bought the beating heart of the developer ecosystem yet kept its brand and independence — and later got it back as a massive asset in Copilot. Lesson: don't kill the ecosystem and open-source community you acquire, and it compounds. Given that Prior Labs' TabPFN is an open-source asset with 3 million downloads, whether SAP keeps that community alive is the whole game.

Failure — the "acquihire absorb" pattern. Conversely, when giants buy promising AI startups and dissolve them into the org, the core talent usually walks within two or three years, leaving an empty shell. Stripped of autonomy and buried in bureaucracy, the researchers leave. Lesson: the real asset in an AI acquisition isn't patents — it's people, and people leave without freedom. SAP promising both €1 billion over four years and independence is a design that reflects awareness of this risk.

The competitors' counter-moves

For Europe's homegrown AI players like Mistral and Aleph Alpha, this is a complicated signal. They've positioned themselves as the flag-bearers of European AI sovereignty — and now SAP has planted a "European frontier lab" flag on a different axis: structured data. It's less head-to-head competition than a division of territory. Mistral remains Europe's alternative in general-purpose LLMs, while SAP–Prior Labs specializes in enterprise tabular data. Still, they'll inevitably compete for the same pool of European government and enterprise funding and talent.

The US frontier labs (OpenAI, Anthropic, Google) will respond from a different angle. They dominate LLMs but are relatively weak at predicting structured business data. Since SAP is driving into exactly that gap, expect the US labs to either beef up table/enterprise-data tooling or partner with data-rich firms like SAP. The realignment of "those who hold the data vs. those who hold the models" is about to kick into gear.

The rival ERP and enterprise-software camp (Oracle, Microsoft, Salesforce) is the most important to watch. They each push their own AI assistants (Oracle AI, Microsoft Copilot, Salesforce Agentforce). Now that SAP holds a differentiating weapon in tabular foundation models, they'll feel pressure to bolt on comparable predictive-AI capability fast. Oracle in particular is SAP's biggest ERP rival, so a similar structured-data AI acquisition or an accelerated in-house build is very plausible.

The hyperscalers (AWS, Google Cloud, Azure) move from the infrastructure angle. If tabular foundation models take off, a market opens to sell them as a cloud service. Rather than picking a side, they'll respond by adding TFMs to their AI service catalogs. In the end, the Prior Labs deal elevated "structured-data AI" as a whole into an industry agenda item.

So what actually changes

For enterprise IT and data teams, the way predictive analytics gets done could shift within a few years. Today a data scientist spends weeks building churn or demand-forecasting models; if tabular foundation models like TabPFN get baked into SAP products, that work could collapse into a single model that runs instantly. Your ERP ledger could show "this supplier's risk is X%" or "this customer's churn probability is Y%" directly. That said, exactly when this lands in shipping products isn't spelled out in a concrete roadmap yet.

For AI researchers and developers, this is a signal that tabular-data AI has finally caught mainstream money. Structured-data research, long overshadowed by LLMs, just landed a €1 billion, four-year lab, so demand for talent and the open-source ecosystem around it should grow. Whether TabPFN stays open source or gets locked inside SAP products is the community's biggest question. Kill the open source here, and this tilts toward the failure column rather than the DeepMind one.

From a European industry and policy standpoint, the symbolism is large. With the US and China splitting frontier AI between them, Europe's largest software company building a world-class AI lab on European soil is a declaration that "Europe is in the game too." Ordinary users won't feel an immediate change, but going forward SAP–Prior Labs will become a frequently cited reference in European regulation and data-sovereignty debates. It's the opening experiment in how Europe answers the US frontier labs.

🥄 Three Things You're Probably Wondering

— So what does this mean for me? No direct impact right now. But if you work at a company that runs SAP, within a few years you might see churn, risk, and demand predictions surface automatically in your ERP screens. And if you do data work, tabular foundation models are a field worth watching starting now.

— Why now, and why table data instead of LLMs? The LLM race is crowded with US Big Tech and has turned into a red ocean, while AI for the structured data enterprises actually run on is still a wide-open blue ocean. SAP holds more enterprise ledger data than almost anyone, so it figures its odds of winning are higher here than in LLMs.

— Can Europe really catch the US frontier labs? Too early to call. €1 billion over four years is real money, but it's dwarfed by US Big Tech's AI spend. Still, SAP chose a different ring — "structured data" — instead of a head-on fight, and there data ownership is the weapon, so it sees a real shot. The next few years will tell.

Further Reading

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!