Mathpix Expands AI Training and Inference Infrastructure at DataVerge, Deploying Nvidia B300 GPUs to Power Real-Time AI Document Processing

Key Takeaways

• Mathpix adds Nvidia B300 GPUs for real‑time document processing
• Colocation at DataVerge reduces latency for New York enterprise users
• Facility provides 1.5 MW now, 3 MW more by Q2 2027
• High‑density pods and 35 carriers ensure scalable, low‑latency AI workloads
• On‑site remote‑hands support accelerates incident response and uptime

Pulse Analysis

The surge in AI‑driven document automation has created a demand for ultra‑low latency processing, especially in finance, research, and enterprise settings where seconds translate to competitive advantage. Traditional public‑cloud routes often add network hops and variable performance, prompting firms like Mathpix to seek edge‑proximate infrastructure. DataVerge’s carrier‑neutral Brooklyn hub offers direct access to over 35 network providers, enabling deterministic routing and minimal jitter for API‑driven workloads, a critical factor for real‑time PDF and handwritten note conversion.

Nvidia’s B300 GPUs represent a generational leap in AI compute, delivering higher memory bandwidth and tensor performance than previous generations. For Mathpix, this translates into the ability to fine‑tune larger language‑vision models and serve more concurrent inference requests without sacrificing latency. The B300’s architecture also supports mixed‑precision workloads, reducing power draw while maintaining throughput—an essential attribute for dense colocation environments where cooling and power budgets are at a premium. By housing these servers in DataVerge’s high‑density cold‑aisle pods, Mathpix can sustain consistent performance even under peak demand.

DataVerge’s strategy of pairing massive power capacity (1.5 MW now, expanding to 4.5 MW) with robust remote‑hands support positions it as a go‑to venue for AI‑intensive startups and established firms alike. The facility’s scalability ensures that companies can grow in place, avoiding costly migrations. As more AI applications migrate from cloud‑only to hybrid edge models, providers that combine carrier diversity, low‑latency interconnects, and next‑gen GPU readiness will capture a growing slice of the market, driving further investment in metropolitan data center ecosystems.