Speridlabs
AI lab building spatial foundation models for understanding and generating 3D worlds.
Website: https://speridlabs.com/
Cover Block
From the public record
| Field | Value |
|---|---|
| Name | Speridlabs |
| Tagline | AI lab building spatial foundation models for understanding and generating 3D worlds. [Speridlabs] |
| Headquarters | Madrid, Spain [LinkedIn] |
| Stage | Seed |
| Business model | API / Developer Platform |
| Industry | Deeptech |
| Technology | AI / Machine Learning |
| Growth profile | Venture Scale |
| Founding team | Co-Founders (2): Chema Garabito, Guillermo Garabito [LinkedIn] |
| Funding label | Undisclosed |
| Investors | Pear VC, Base10 Partners [Speridlabs, April 2026] |
| Accelerator | PearX [TechCrunch, October 2026] |
Links
From the public record
- Website: https://speridlabs.com/
- LinkedIn: https://www.linkedin.com/company/speridlabs
- X / Twitter: https://x.com/speridlabs
The Short Version
PUBLIC Speridlabs is an early-stage spatial AI company building foundation models for understanding and generating 3D worlds, and it merits investor attention now because it has begun to surface in credible startup ecosystems with a concrete product concept and named backers, but still sits before broad commercial proof [Speridlabs, April 2026] [TechCrunch, October 2026]. The public record is thin on the company's origin story, but Speridlabs emerged from stealth in April 2026 and later appeared at PearX demo day in October 2026, which gives at least a dated sequence for its public launch and early investor visibility [Speridlabs, April 2026] [TechCrunch, October 2026].
The product thesis is straightforward: Speridlabs says it is building spatial foundation models that can understand the 3D world, generate within it, and eventually reason over dynamic environments, with an initial application focus in robotics, gaming, and special effects [Speridlabs, April 2026] [TechCrunch, October 2026] [Mezha, October 2026]. Its first disclosed product, Mundus, is described in secondary coverage as a "3D Midjourney" that lets users modify one part of a scene while preserving the geometry of the rest, which, if borne out in product use, would make editability rather than raw generation the key wedge to watch [TechCrunch, October 2026] [Mezha, October 2026].
On team, Chema Garabito is publicly identified as founder, CEO, and chief scientist, while public LinkedIn-based sourcing also ties Guillermo Garabito to the founding group, though verified detail on prior operating history, research pedigree, and commercial track record remains limited in the available record [LinkedIn] [Prospeo]. That leaves the current investment question less about founder visibility and more about whether the team can convert a technically ambitious thesis into repeatable developer adoption.
The commercial posture appears to be an API or developer-platform model aimed at developers, researchers, creators, and companies, while funding disclosure remains notably sparse [Speridlabs, April 2026]. Speridlabs has said it is backed by Pear VC and Base10 and participated in PearX, but has not publicly disclosed round size, valuation, or lead structure, so the next 12 to 18 months likely hinge on evidence of product usability, named design partners or customers, and clearer signals that the company can move from category promise into measurable adoption [Speridlabs, April 2026] [TechCrunch, October 2026].
Unconfirmed -- This section relies partly on company materials and limited third-party coverage, with some team details only partially corroborated by LinkedIn and directory data.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Seed |
| Business Model | API / Developer Platform |
| Industry / Vertical | Deeptech |
| Technology Type | AI / Machine Learning |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (2) |
| Funding | Undisclosed |
The Company in Brief
PUBLIC
Speridlabs is presenting itself first as a research lab, not a packaged software company, and that framing matters because the public record is still thin on corporate detail. On its website, the company describes itself as an AI lab building spatial foundation models for understanding and generating 3D worlds, with Madrid, Spain listed as its headquarters and Pear VC and Base10 named as backers in its April 2026 launch post [Speridlabs, Unknown] [Speridlabs, April 2026]. The same materials identify Chema Garabito and Guillermo Garabito as founders, though the available company sources do not specify a founding date or legal entity name [Speridlabs, April 2026].
The clearest public milestone is the April 2026 blog post, "The Shape of Intelligence," which functions as the company coming-out statement. In that post, Speridlabs laid out its thesis around "Spatial Intelligence," said it was building models intended for developers, researchers, creators, and companies, and disclosed backing from Pear VC and Base10 without naming a round size or date [Speridlabs, April 2026]. The website and company LinkedIn presence together suggest an early-stage organization still defining its external footprint rather than a business with a mature set of public operating disclosures [Speridlabs, Unknown] [LinkedIn].
A second milestone arrived later in 2026, when Speridlabs appeared at PearX demo day, an event that linked the company more clearly to the venture ecosystem around Pear VC. That appearance is relevant as a chronology marker, but the company website remains the more direct source for what Speridlabs says it is building and who it is building for [Speridlabs, April 2026]. For now, the investable question is less about corporate history, which is only lightly documented in public, and more about whether the lab can translate an ambitious technical thesis into durable product evidence.
Single-source, plausible -- Confirmed primarily by Speridlabs' website and April 2026 company blog post, with partial corroboration from LinkedIn.
What They Have Built
MIXED
Speridlabs is making a narrow but interesting product claim: that its models can work inside a 3D scene without rewriting the whole scene each time. The company describes itself as an AI lab building spatial foundation models that understand the 3D world, generate within it, and eventually reason over dynamic worlds that change through time [Speridlabs, April 2026]. In named-publisher coverage, the first application is framed more concretely around robotics, gaming, and special effects, which gives the effort a clearer commercial shape than the broader "spatial intelligence" language on the company site [TechCrunch, October 2026] [Mezha, October 2026].
The product surface that is publicly described is Mundus. According to TechCrunch and Mezha, Mundus lets users modify part of a 3D image or scene while preserving the geometry of the rest of the environment, a capability the company has also summarized informally as a "3D Midjourney" [TechCrunch, October 2026] [Mezha, October 2026]. That matters because most of the public differentiation rests on editability and scene persistence, not on a disclosed model architecture, benchmark result, or deployment footprint. The intended users named by the company are developers, researchers, creators, and companies, which fits an API or developer-platform posture, but the public record does not yet show technical documentation, pricing, or verified production usage at the time of writing [Speridlabs, April 2026] [Speridlabs].
Single-source, plausible -- Product direction is supported by one company primary source and two named-publisher reports, but technical specifics remain limited and some framing originates with the company.
Market Size and Demand
PUBLIC
The market matters now because Speridlabs is aiming at a point where generative AI, robotics simulation, and 3D content creation are starting to overlap in public discourse, but the underlying commercial category is still being defined [TechCrunch, October 2026] [Speridlabs, April 2026].
The public record does not support a clean TAM, SAM, or SOM for "spatial foundation models" as a standalone category, so the more careful approach is to anchor on adjacent markets rather than force precision that the sources do not offer. Speridlabs itself frames the initial application set as robotics, gaming, and special effects, and that framing is repeated in third-party coverage from TechCrunch and Mezha [Speridlabs, April 2026] [TechCrunch, October 2026] [Mezha, October 2026]. That suggests the company is not entering a single software budget line so much as trying to sit across several existing ones: developer tooling for 3D creation, model infrastructure for embodied AI, and visual production workflows.
The demand signal in the sources is qualitative, but it is coherent. The company describes its work as building models that understand the 3D world, generate within it, and eventually reason over dynamic worlds that change through time, while TechCrunch highlights a more immediate wedge: the ability to edit one part of a 3D scene without changing the rest of the geometry [Speridlabs, April 2026] [TechCrunch, October 2026]. If that claim holds in production, the commercial pull would likely come from workflows where persistence matters more than one-shot image generation, especially simulation, game asset iteration, and effects pipelines. That is still an inference from product positioning rather than disclosed customer adoption.
Adjacent markets matter here because buyers may evaluate Speridlabs against existing tools before they evaluate it as a new category. In robotics, the substitute is often simulation and scene-understanding tooling rather than media-generation software; in gaming and special effects, the substitute can be conventional 3D creation software or newer generative systems that operate on video or images [TechCrunch, October 2026] [Mezha, October 2026]. The company itself has argued that systems such as Runway, Odyssey, and Google's Genie are less editable in the way Mundus is intended to be, but that remains a company claim rather than an independently established market consensus [Speridlabs, April 2026].
Macro and regulatory forces are present mostly by implication in the available material. Robotics and simulation markets continue to benefit from broader AI investment cycles, while gaming and visual effects remain sensitive to production budgets and toolchain switching costs; both conditions can help or slow adoption depending on how much workflow change Speridlabs requires [TechCrunch, October 2026]. The public sources used here do not identify company-specific regulatory approvals or constraints, but any product serving generative media and model infrastructure will likely face the familiar questions around training data provenance, enterprise reliability, and IP-sensitive deployment, which investors would need to diligence separately because the current record does not answer them [Speridlabs, April 2026].
| Market lens | Public claim | Interpretation |
|---|---|---|
| Core category | Speridlabs describes itself as an AI lab building spatial foundation models that understand and generate 3D worlds [Speridlabs, April 2026] | The company is positioning into an emerging category, not a mature budget line |
| Initial sectors | Robotics, gaming, and special effects are cited as the first application areas [TechCrunch, October 2026] [Mezha, October 2026] | Early demand is likely to be fragmented across several buyer groups |
| Product wedge | Mundus is presented as an editable 3D generation tool that preserves scene geometry when one part changes [TechCrunch, October 2026] [Mezha, October 2026] | The commercial test is whether persistence solves a costly workflow problem better than existing 3D or generative tools |
The table shows a market story built on adjacency and workflow pain points, not on published market-size math. That makes the opportunity interesting, but it also means category formation risk is still high.
Single-source, plausible -- Based primarily on company materials, with partial corroboration from TechCrunch and Mezha; no independent third-party market-sizing report was identified in the provided sources.
Who Else Is Fighting for This
Competitive Set
MIXED Speridlabs is positioning itself against a mix of video-generation platforms, interactive world-model efforts, and broader generative AI labs, with the clearest public claim centered on editable 3D scenes rather than 2D media output [Speridlabs, April 2026] [TechCrunch, October 2026].
| Company | Positioning | Stage / Funding | Notable Differentiator | Source |
|---|---|---|---|---|
| Speridlabs | Spatial AI lab building foundation models to understand and generate 3D worlds for robotics, gaming, and special effects | Seed, backing from Pear VC and Base10 disclosed, amount undisclosed | Mundus is described as enabling edits to one part of a 3D scene while preserving the rest of the scene's geometry | [Speridlabs, April 2026] [TechCrunch, October 2026] |
| Runway | Named by Speridlabs as a competing system in generative media | Not established in the provided sources | Publicly treated here only as a reference point for comparison, not independently benchmarked in the available materials | [Speridlabs, April 2026] |
| Odyssey | Named by Speridlabs as a competing system in world or media generation | Not established in the provided sources | Publicly treated here only as a reference point for comparison, not independently benchmarked in the available materials | [Speridlabs, April 2026] |
| Google's Genie | Named by Speridlabs as a competing system in generated interactive worlds | Not established in the provided sources | Publicly treated here only as a reference point for comparison, not independently benchmarked in the available materials | [Speridlabs, April 2026] |
The public map breaks into three segments. First are adjacent incumbents in generative media, represented here by Runway, where the user expectation is fast visual output and broad creator awareness, even if the available source set does not independently document feature parity with Speridlabs [Speridlabs, April 2026]. Second are frontier world-model or interactive-environment efforts such as Odyssey and Google's Genie, which matter because they frame the competitive standard around persistent environments rather than single images [Speridlabs, April 2026]. Third are substitutes from internal tooling in robotics, gaming, and special effects, where a buyer may choose existing simulation, asset, or pipeline software instead of adopting a new model layer, although the current source set does not identify specific vendors in those categories [TechCrunch, October 2026].
The edge Speridlabs can argue today is narrow but coherent. Its own materials, echoed by TechCrunch and Mezha, point to a specific product behavior: modifying one part of a scene while preserving the geometry of the rest [Speridlabs, April 2026] [TechCrunch, October 2026] [Mezha, October 2026]. That is a more concrete wedge than a general claim to "3D AI," because it speaks to workflow integrity in domains where scene consistency matters. The problem is durability. Public evidence does not yet show a proprietary dataset, customer lock-in, developer ecosystem, or distribution channel that would make this advantage hard for better-capitalized peers to copy [Speridlabs, April 2026].
Exposure is easier to see than insulation. Google has the strongest named advantage if the contest shifts toward large-scale world models that require extensive compute, research depth, and integration with existing developer platforms, though the provided sources do not establish Google's current product scope beyond the Genie reference [Speridlabs, April 2026]. Runway, by contrast, represents a different risk: if customer demand settles on creator-friendly interfaces and rapid content workflows instead of deep spatial consistency, then a company built around a more specialized 3D stack could struggle to own the primary distribution surface [Speridlabs, April 2026]. Speridlabs also does not appear, from public materials, to own a downstream channel in robotics, gaming, or VFX, which makes platform adoption contingent on product quality rather than captive demand [TechCrunch, October 2026].
The most plausible 18-month scenario is a sorting of the category by end use rather than a single winner-take-all outcome. Winner if X: Google's Genie, if the market rewards research scale and broad developer reach over specialized editing behavior [Speridlabs, April 2026]. Loser if Y: Speridlabs, if editable persistent geometry remains a compelling demo but does not convert into a repeatable developer platform with named users or production deployments [TechCrunch, October 2026] [Speridlabs, April 2026]. The counter-scenario is still credible: Speridlabs could carve out a defensible niche if robotics, gaming, or effects teams prove willing to adopt a purpose-built spatial model rather than retrofit 2D-first generative tools, but public evidence for that adoption is not yet visible [TechCrunch, October 2026] [Mezha, October 2026].
Single-source, plausible -- Competitive names are confirmed in company materials and named-publisher coverage, but most comparative feature and stage detail for rivals is not independently corroborated in the provided public source set.
Opportunity
PUBLIC
The prize here is large if Speridlabs can turn a research thesis into production infrastructure, because a model layer that can understand, edit, and eventually reason over 3D worlds would sit upstream of robotics, game creation, and visual-effects workflows that still depend on labor-intensive tooling and fragmented pipelines [Speridlabs, April 2026] [TechCrunch, October 2026].
The headline opportunity is straightforward: become a default spatial-model API for developers and studios that need editable 3D world generation, not just one-off image or video outputs. That outcome is still early, but it is not purely aspirational on the public record. Speridlabs has articulated a product direction around spatial foundation models that "understand the 3D world, generate within it, and eventually reason over dynamic worlds that change through time," and TechCrunch's PearX coverage points to a specific wedge in Mundus, the ability to modify one part of a 3D scene while preserving the rest of the geometry [Speridlabs, April 2026] [TechCrunch, October 2026]. If that editing behavior holds up in real developer use, the company is pursuing a harder and more defensible layer than prompt-only media generation.
The upside paths are best thought of as a small set of concrete adoption routes rather than one monolithic market claim. Public evidence supports at least three plausible routes.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Developer infrastructure for 3D generation | Speridlabs becomes an API and model provider used by game, robotics, and simulation teams that need editable spatial outputs | A public product launch of Mundus or related APIs that proves persistent-geometry editing in production workflows | The company already frames developers and researchers as core users, and describes itself as building models rather than only consumer software [Speridlabs, April 2026] |
| Workflow layer for creative tooling | Mundus becomes embedded in content pipelines for studios and creators who need to revise scenes without rebuilding them from scratch | A visible integration, plugin, or design-partner win in gaming or VFX | TechCrunch and Mezha both highlight scene-level editing while preserving geometry as the differentiator, which maps more naturally to iterative creative workflows than to one-time asset generation [TechCrunch, October 2026] [Mezha, October 2026] |
| Spatial intelligence for robotics simulation | The models become useful in robotics data generation, environment understanding, or simulation authoring | A proof point that the same spatial model can support robotics use cases, not just media creation | Robotics is named in both company and third-party coverage as an initial application area, suggesting management is orienting the stack toward machine interaction with 3D environments rather than a narrow media niche [Speridlabs, April 2026] [TechCrunch, October 2026] |
The common thread in those scenarios is compounding through a difficult product surface. If developers build against a spatial API that preserves geometry and supports iterative scene editing, each use case can generate feedback on where models fail, which objects break, and which workflows recur most often. Over time, that can improve model performance in commercially relevant edge cases and make the platform more useful to the next cohort of builders. The beginnings of that loop are only lightly visible today, but the company's stated audience spans developers, researchers, creators, and companies, which is consistent with a platform strategy rather than a single-purpose app [Speridlabs, April 2026]. Pear VC, Base10 Partners, and the PearX demo-day slot do not validate product-market fit, but they do suggest the company has enough external support to pursue that loop past the lab stage [Speridlabs, April 2026] [TechCrunch, October 2026].
The size of the win is harder to anchor cleanly because the public record here does not include revenue, customer counts, or a disclosed financing round. Even so, the ambition can be stated with discipline. If Speridlabs were to become core infrastructure for spatial generation across a few high-value verticals, the outcome could resemble an infrastructure platform rather than a feature company. That would place the upside in venture-scale territory by definition, though any valuation translation remains a scenario, not a forecast. On current public evidence, the more defensible statement is that the company is aiming at a foundational layer in 3D world modeling, and foundational layers can matter disproportionately if they become embedded in developer workflows across robotics, gaming, and special effects [Speridlabs, April 2026] [TechCrunch, October 2026] [Mezha, October 2026].
Single-source, plausible -- Based primarily on company materials, with partial corroboration from TechCrunch and Mezha on product direction and initial application areas.
Sources
From the public record
[Speridlabs, April 2026] The Shape of Intelligence | https://speridlabs.com/blog/01-post
[TechCrunch, October 2026] 5 startups that caught VCs’ attention at the latest PearX demo day | https://techcrunch.com/2026/10/05/5-startups-that-caught-vcs-attention-at-the-latest-pearx-demo-day/
[Mezha, October 2026] PearX Demo Day Spotlights 16 Startups Building AI, Robotics and Finance Tools | https://mezha.net/eng/news/14f1c4ac_pearx_demo_day/
Articles about Speridlabs
- Speridlabs Builds a 3D Midjourney for Robots and Games — The Madrid AI lab, backed by Pear VC and Base10, is betting on editable spatial models as a wedge into robotics and special effects.