The Paris-based AI-native real-estate company, founded by Entrepreneurs First alumni Mehdi Rais and Amine Chraibi, has Heartcore and Balderton co-leading the round, an unusual cap-table for a pre-seed. The technical pitch is more interesting than the headline.

There is a particular shape of European AI seed round that has, in 2026, become harder to land than the headline figures imply: a pre-seed in which two of the continent’s most experienced firms agree to co-lead. On Tuesday, Paris-based Davis announced a $5.5m pre-seed with Heartcore Capital and Balderton Capital both leading.

The participation list adds Evantic, Yellow VC, and Entrepreneurs First on the institutional side, with an angel roster that includes operators and researchers from Meta, Black Forest Labs, Hugging Face, Supabase, Spore.Bio, and members of the founding team behind SpaceMaker, the architectural-AI startup Autodesk acquired in 2020.

That last detail is the most informative one. Davis is, on paper, a real estate technology company. On the technical specifics it has chosen to disclose, it is something more interesting than that.

What Davis actually does?

The proposition is, in plain terms, the elimination of the architectural feasibility gap. Real-estate developers and investors evaluating a site currently spend weeks or months moving through a fragmented stack of tools, consultants, and software to get from raw land data to a usable feasibility study and a credible architectural concept.

Davis takes that process and replaces most of the calendar with a single integrated workflow: regulatory, technical, and market data go in as constraints; feasibility studies and architectural designs (volumetrics, floor plans, space planning, ROI estimates) come out within days. Human architects review every output before it is delivered to clients.

The company is not selling software. It is, deliberately, selling outcomes. Developers and investors receive finished feasibility studies and design concepts directly, in the same way they would receive them from a traditional architectural consultancy, except that the underlying production process is AI-led with an architect-in-the-loop validation layer.

The hybrid model is structurally important: it lets Davis price as a service rather than as a SaaS product, capture a higher share of value per project, and avoid the tooling-adoption friction that has historically slowed PropTech penetration into traditional development workflows.

Davis’s bet is that architectural AI should not generate buildings like images. Most recent tools in the category rely on continuous diffusion models, the same family of systems used for image and video generation. Davis is taking a different route: its models generate buildings as structured compositions of rooms, walls, layouts, and architectural elements.

That distinction matters. It follows the technical lineage of research such as HouseDiffusion, which showed that discrete and continuous denoising can produce better floorplan geometry than pixel-space generation, especially on relationships architects actually care about: parallelism, orthogonality, shared corners, and clean incident geometry.

The company is launching its first proprietary model, Gaudi-1, alongside the round. Davis claims state-of-the-art results on RPLAN and MSD, the Swiss Dwellings dataset of 5,372 detailed floor plans, across IoU, FID, and KID metrics. If those claims hold under independent evaluation, the positioning is meaningful. Davis is trying to solve the part of architectural AI that has kept many tools trapped as demos: producing outputs that can satisfy real-world design, regulatory, and financial constraints.

Why the founding team and cap table matter

Davis was founded in 2025 by Mehdi Rais, its CEO, and Amine Chraibi, its CTO. Rais grew up in a family of architects, giving the company a practitioner-side understanding of the problem. Chraibi is an AI researcher from École Polytechnique, with work focused on generative models for structured data, the same technical territory Davis is now commercialising.

The two met through Entrepreneur First’s Paris cohort and, according to EF, were among the fastest teams in the cohort to raise. The cap table reflects that momentum. Heartcore and Balderton co-leading a pre-seed is an unusually strong signal in European AI. Heartcore’s Max Niederhofer pointed to the combination of a discrete architectural model, regulatory constraints, architect-in-the-loop validation, and the promise of compressing feasibility work from months to days. Balderton’s Rob Moffat framed the attraction more simply: few AI startups are both fast to market and building proprietary models.

The angel slate adds another layer. Members of the founding team behind Spacemaker, acquired by Autodesk in 2020 for $240m, have backed Davis directly. That matters because Spacemaker remains the cleanest exit precedent in architectural AI.

Spacemaker sold AI tools to architects and urban planners, helping them generate and optimise early site designs against constraints like wind, lighting, terrain, traffic, and zoning. Davis is aiming at a different customer and a different business model.

Its customer is the real-estate developer. Its deliverable is not a tool, but a finished feasibility study. Its value capture comes through project-level service contracts rather than SaaS subscriptions.

That shift is central to the story. Selling software to architects is a narrower market. Selling outcomes to developers opens a larger one, measured in projects rather than firms. But it also raises the bar: Davis’s AI outputs must be good enough that human architects can validate them, not rebuild them. The discrete architectural-element approach is the technical bet behind that claim.

The first risk is execution. Davis’s economics depend on keeping the human validation layer narrow. If architects spend too much time correcting the AI’s work, the promised compression from months to days breaks down.

The second risk is regulatory complexity. Architecture and development rules vary across countries, cities, and municipalities. Davis says its system can adapt to local regulations as input data, but supporting many regimes at once is operationally difficult.

The third risk is competition. Autodesk, Bentley, Trimble, and a long list of AI-native entrants are moving toward generative design. Davis’s technical approach may be differentiated, but it is entering a category where incumbents have distribution, capital, and existing customer relationships.

Davis is one of the more credible European AI pre-seeds of 2026 because the story has a clear red line: a founder pair with domain and technical depth, a heavyweight cap table, and a model architecture aimed at the real bottleneck in development, feasibility work.

The bet is not simply that AI can draw floor plans faster. It is that structured architectural generation can compress one of real estate’s slowest and most expensive workflows. If that holds across asset classes and jurisdictions, Davis will have something more durable than a clever design tool. It will have a technical moat inside one of the world’s largest asset classes.