Gather AI Raises $40M for Physical AI Infrastructure Beyond LLMs
Gather AI has secured a $40 million Series B led by Smith Point Capital, the firm of former Salesforce co-CEO Keith Block, bringing its total funding to $74 million. The round validates the Carnegie Mellon-founded startup’s Physical AI platform, which uses classical Bayesian techniques—not LLMs—to power autonomous systems that actively map and monitor warehouse operations.
This investment marks a critical inflection point for AI infrastructure, shifting focus from purely digital, text-generating agents to systems that must operate reliably in the physical world. As enterprises seek to scale automation beyond the data center, a new class of Physical AI infrastructure is emerging to solve the critical bottlenecks of real-world deployment across logistics, manufacturing, and field services.
The Warehouse Intelligence Bottleneck
A fundamental visibility crisis cripples modern warehouse operations. Despite heavy investment in automation, logistics facilities are plagued by inefficient inventory tracking, safety blind spots, and reactive workflow management. This “last mile” of intelligence relies on manual processes that cannot scale.
Traditional warehouse management depends on periodic barcode scanning and manual audits, leaving massive gaps in real-time operational data. The resulting inefficiencies—time spent locating lost pallets, correcting inventory mismatches, and identifying damaged goods—create a significant drag on productivity and profitability that intensifies as supply chain complexity grows.
Existing automation hardware can move and sort goods, but it lacks the continuous, ambient intelligence needed for predictive optimization. This creates a persistent bottleneck where operational efficiency is capped by the limits of human oversight.
A Bayesian Architecture for Physical Space
Gather AI’s platform transforms off-the-shelf cameras on forklifts and autonomous drones into a continuous, facility-wide intelligence layer. The system’s core differentiator is its goal-directed behavior; rather than passively scanning, its agents actively seek specific operational data based on real-time priorities.
The technical foundation combines classical Bayesian probability methods with targeted neural networks, a deliberate move away from the LLM-first trend. This architecture provides the determinism and reliability essential for physical tasks, eliminating the risk of hallucinations while enabling real-time decisions about what information to gather next.
The system is trained to identify barcodes, lot codes, case counts, and signs of damage or non-compliance. By maintaining a contextual model of the warehouse’s state, it becomes “curious” about specific data points needed to resolve discrepancies or optimize workflows. Integration with existing Warehouse Management Systems (WMS) allows for immediate visibility without costly infrastructure replacement.
Enterprise Adoption Evidence
Gather AI’s platform is deployed with logistics leaders like Kwik Trip, Axon, GEODIS, and NFI Industries, where real-time visibility directly impacts financial performance. This customer roster validates the solution’s fit for high-throughput, mission-critical environments.
The startup’s technical credibility is rooted in its Carnegie Mellon origins, where the founding team developed autonomous helicopter systems for the FBI. This experience in real-world robotics informs their approach to warehouse intelligence. The recent 2025 Nebius Robotics award for Vision AI further demonstrates industry recognition.
The leadership of Keith Block, who executed the enterprise go-to-market playbook at Salesforce, signals strong confidence in Gather AI’s ability to scale and meet the rigorous demands of Fortune 500 customers.
The Emergence of Physical AI Infrastructure
Gather AI exemplifies the convergence of robotics, computer vision, and enterprise software that defines the emerging Physical AI infrastructure category. Unlike digital agents, these systems must navigate environmental variables, hardware integration, and safety protocols that are non-existent in software-only deployments.
The turn to classical AI highlights a key technical divergence. Physical AI requires architectural foundations optimized for reliability, real-time decision-making, and environmental adaptation—qualities where probabilistic LLMs often fall short.
This trend mirrors the evolution of cloud computing, where specialized platforms (e.g., for databases, serverless) emerged to solve problems that general-purpose infrastructure handled inefficiently. The warehouse market is the proving ground for this new infrastructure class before it expands into manufacturing, agriculture, and other physical domains.
Looking Forward
The next 12-18 months will be a crucial test for the Physical AI market, determining if it can match the adoption velocity of software-based AI. Key indicators will be the expansion of customer deployments, the depth of WMS and robotics integrations, and clear evidence of ROI in diverse operational settings.
Technical evolution will likely focus on multi-agent coordination between drones and ground-based systems, alongside deeper integration with other robotics platforms. The determinism of Gather AI’s Bayesian approach provides a distinct advantage in regulated industries and for any operation where explainability is paramount.
As the market matures, expect established automation vendors and major cloud platforms to enter the space. Gather AI’s technical head start and enterprise validation provide a strong, defensible position in this rapidly expanding category.
The rise of Physical AI platforms like Gather AI exposes the next layer of enterprise automation challenges: coordinating autonomous agents in complex, real-world environments. As these systems move from isolated tasks to interconnected workflows, orchestration platforms like Overclock become essential for managing the interplay between physical agents, digital systems like a WMS, and human operators, ensuring reliable, end-to-end process execution.