SANTA CLARA, California, January 21, 2026 — Upscale AI has closed an oversubscribed $200 million Series A financing to move its open AI networking platform from architecture into commercial deployment. Tiger Global, Premji Invest and Xora Innovation led the round, which brought total funding received by the company to more than $300 million.

The financing supports a full-stack product spanning silicon, systems and software. Upscale AI plans to expand engineering, sales and operations while preparing its first networking solutions for shipment during 2026. The company is addressing a specific infrastructure problem: expensive accelerators, memory and storage cannot operate as one productive system when communication becomes the constraint.

Series A follows a $100 million seed round

The new financing arrived about four months after Upscale AI emerged with more than $100 million of seed capital. Alongside the three lead investors, participants in the Series A included Maverick Silicon, StepStone Group, Mayfield, Prosperity7 Ventures, Intel Capital and Qualcomm Ventures.

Bloomberg reported that the transaction valued the company at more than $1 billion. Upscale AI did not publish an exact valuation in its release, so that figure should be understood as reported transaction context rather than a precise company-disclosed metric.

The announcement establishes four operating priorities

  • complete development across networking silicon, systems and software;
  • expand engineering capacity as products move toward commercial deployment;
  • build sales and operational teams for hyperscalers and AI infrastructure operators;
  • support open standards so customers can combine different compute technologies.

SkyHammer is designed for scale-up networking

Upscale AI describes SkyHammer as its purpose-built scale-up architecture. Scale-up communication joins accelerators, memory and storage closely enough to behave as a coordinated computing domain. This differs from simply connecting conventional endpoints across a general-purpose data-centre network.

Training and inference workloads exchange large amounts of data under tight timing requirements. If communication latency, bandwidth or congestion prevents processors from receiving work, adding more accelerators can increase capital cost without producing proportional useful computing. Networking therefore affects utilisation and economics as well as technical performance.

An exploded unbranded networking assembly connects a large switch chip, circuit board, optical modules and copper links to two dense AI server racks
AI infrastructure becomes a coordinated engine only when processors, memory, storage, switches and links work within the same performance envelope.

An open stack is part of the product claim

Upscale AI's release catalogue places the Series A alongside its earlier launch and architecture milestones. The company says its platform advances technologies including ESUN, Ultra Accelerator Link, Ultra Ethernet, SONiC and the Switch Abstraction Interface.

These technologies address different layers and scopes. A named standard does not by itself prove interoperability. Commercial customers will need validated combinations of switch silicon, optics, cables, software, management tools and accelerators, together with repeatable installation and support.

Upscale AI also lists participation in the Ultra Accelerator Link Consortium, Ultra Ethernet Consortium, Open Compute Project and SONiC Foundation. Participation can provide access to ecosystem requirements and implementation work; customer choice will depend on shipping products, conformance and sustained compatibility.

Funding moves the test from design to delivery

The company said its networking solutions were slated to ship during 2026. That statement is a schedule objective, not evidence that volume delivery had already begun on January 21. Commercial deployment requires production silicon, system qualification, software stability, supply availability and field support to converge.

Customers operating large AI clusters also evaluate failure recovery, observability, power, cooling and upgrade paths. A high-performance switch can become an operational bottleneck if faults are difficult to isolate or if a software change interrupts a large computing domain. The announced expansion of engineering and operations is therefore as relevant as the underlying chip.

Management sees networking as the next AI constraint

Chief executive Barun Kar said the investment would accelerate the company's effort to re-architect networking for the AI era. Executive chairman Rajiv Khemani pointed to demand for open, scalable alternatives. These statements express management's strategy and expectations; adoption will be determined through customer qualification and deployment.

The company profile identifies a leadership team with experience across networking, silicon, systems and software. Upscale AI is positioning that integrated expertise against established suppliers with broad product portfolios, mature distribution and long customer relationships.

About Upscale AI

Upscale AI develops high-performance networking infrastructure for AI training, inference, generative AI, edge computing and cloud-scale deployments. Its stated approach combines open-standard silicon, systems and software to provide ultra-low-latency communication across heterogeneous AI infrastructure.

Company contact

Upscale AI, Inc.
Press and media: press@upscaleai.com
Office: 3101 Jay Street, Santa Clara, CA 95054
Website: https://upscale.com