Mathpix expands Brooklyn GPU deployment for AI workloads

Mathpix is expanding its deployment at DataVerge's Brooklyn data centre, adding NVIDIA B300 GPUs for AI training and real-time inference workloads.

The systems are being installed at DataVerge's Industry City facility as Mathpix builds out infrastructure for document conversion and structured data extraction services. The deployment expands a colocation relationship established last year.

Mathpix uses artificial intelligence to convert PDFs, handwritten notes, equations and scanned files into machine-readable text. According to the companies, the output is used in enterprise workflows, research applications and financial market tools.

The move reflects a broader push by some AI companies to place computing infrastructure closer to users rather than rely entirely on distant cloud regions. For Mathpix, the local deployment is intended to reduce latency for API-driven workloads serving customers in the New York area.

DataVerge describes its Industry City site as Brooklyn's largest carrier-neutral interconnection facility. The 50,000-square-foot data centre connects to more than 35 carriers and network providers.

The latest deployment centres on NVIDIA B300 GPU servers, which the companies said will help Mathpix run larger AI models and support more concurrent workloads. The hardware is also expected to speed up document processing tasks that require both model training and live inference.

That includes user uploads, batch processing jobs and API requests from customers integrating Mathpix's document recognition tools into their own systems. The company offers consumer and developer products, including its Snip app and MathpixOCR API.

Local compute

Proximity to users and customers was a key factor in the decision. By keeping systems in Brooklyn, Mathpix said it can process workloads closer to its client base instead of routing them through distant cloud regions.

"For Mathpix, AI training and inference performance are product features," said Nico Jimenez, Chief Executive Officer, Mathpix. "Our customers expect near-instantaneous document conversion, which means our infrastructure needs to be close to them and built for the demands of modern AI workloads. For both fine-tuning models and real-time inference, milliseconds matter when processing user uploads, enterprise batch jobs, and API-driven workflows. DataVerge gives us the ability to deploy the B300s with the power density, connectivity, and hands-on support we need at a cost structure that makes sense."

DataVerge said its facility supports GPU-intensive deployments with high-density cold aisle containment pods, around-the-clock remote hands support and what it described as rapid incident response. It also highlighted its Meet-Me Room, which connects customers to multiple carriers and network providers for routing and connectivity options.

Room to grow

DataVerge said the site has 1.5 MW of available capacity now, with another 3 MW due to come online later. That would give Mathpix room to expand within the same facility as demand rises rather than shifting workloads elsewhere.

DataVerge has been positioning the site as a location for businesses that need dense power, cooling and network access for AI systems. Growing demand for GPU infrastructure has sharpened the focus on data centre sites that can support power-hungry hardware while maintaining low-latency links to users.

"The B300 delivers a step change in AI inference capability, enabling larger models, more concurrent workloads, and faster response times at enterprise scale," said Ray Sidler, Chief Executive Officer and Cofounder, DataVerge. "But supporting that hardware in production requires high-density power and cooling, fast incident response, and the network architecture to keep latency low. AI inference companies don't just need somewhere to put GPUs; they need production-grade infrastructure that's close to their users and can scale without disruption."