Tensordyne Announces Breakthrough Inference System to End AI’s Speed vs. Cost Trade-Off

Tensordyne Napier (TDN) has successfully taped out, delivering up to 17x more tokens per watt and 13x higher throughput than NVIDIA Blackwell systems

‍

SUNNYVALE, CALIFORNIA & MUNICH, GERMANY, June 15th, 2026 – AI systems company Tensordyne today announced Tensordyne Napier (TDN), a breakthrough AI inference system designed to eliminate the industry’s longstanding trade-off between inference speed and cost efficiency.

Built in partnership with Broadcom and HPE Juniper Networks, the Tensordyne Napier platform combines novel logarithmic AI math, tightly integrated memory architecture, and a high-performance scale-up interconnect to deliver substantially higher throughput, lower power consumption, and improved infrastructure economics for large-scale AI inference workloads.

Tensordyne has successfully completed tape-out of the Napier processor, which is now in production at TSMC on its 3nm process node.

Over the past year, Tensordyne has built a growing pipeline of hyperscaler, neocloud, and sovereign AI infrastructure operators evaluating the platform for next-generation inference deployments, including more than a dozen Letters of Intent tied to system demos and early access engagement. With tape-out now complete and production underway at TSMC, Tensordyne is advancing these engagements toward benchmarking, beta deployment, and broader infrastructure planning opportunities representing more than $200 million in forecasted Napier system demand.

This launch comes ahead of an anticipated Series D financing later this year.

‍

Re-Architecting AI Inference from First Principles

Demand for high-speed inference has accelerated dramatically as AI adoption scales globally. However, today’s inference infrastructure remains constrained by rising power consumption, infrastructure density limitations, and escalating operational costs.

The world’s largest Generative AI companies now spend more than 50% of current revenues on inference infrastructure, making inference economics one of the defining challenges facing the industry.

To address these limitations, Tensordyne redesigned core components of the AI inference stack across math, compute, memory, and networking:

TDN Math (Logarithmic Mathematics)

TDN replaces large-scale multiplication operations with simplified addition-based computation, significantly improving performance-per-watt efficiency across frontier AI models.

TDN AIP (Artificial Intelligence Processor)

Each TDN processor tightly integrates substantial fast SRAM alongside HBM memory, minimizing idle compute cycles and supporting efficient execution of the industry’s largest models.

TDN Link (Any-to-Any Scale-Up Interconnect)

Tensordyne’s proprietary scale-up fabric delivers sub-microsecond communication latency between processors, maximizing compute utilization and minimizing interconnect bottlenecks.

TDN72 Inference Pod and Rack System

Tensordyne is also announcing the TDN72, a 72-chip inference pod designed to outperform a full NVIDIA NVL72 rack while consuming substantially less infrastructure capacity.

A full Tensordyne Napier rack combines four TDN72 pods and delivers:

• 17x more tokens per watt
• 13x more tokens per second
• Up to $33 million more annual revenue per rack

The system supports disaggregated serving and multi-trillion parameter model execution at greater than 1,000 tokens per second per user within a single rack configuration. It would require at least nine NVIDIA Rubin + Groq LPX racks to perform the same throughput.

Addressing the Industry’s Infrastructure Shift

Hyperscalers are expected to spend more than $700 billion on infrastructure this year as demand for inference accelerates globally. Yet current systems continue to force trade-offs between inference speed, deployment density, and operating economics.

Tensordyne Napier was designed specifically to address these constraints by delivering high-speed inference with substantially improved infrastructure efficiency and profitability potential.

“The market is hungry for fast AI; customers want speed, but achieving it has always meant accepting prohibitive costs,” said Marc Bolitho, Tensordyne’s CEO. “By optimising math, compute, memory, and networking from first principles, Napier delivers affordable inference without compromising on speed. We’re delivering fast, cost-effective AI and eliminating the need for bolted-together systems that require multiple racks.”

Frank Ostojic, Senior Vice President and General Manager, ASIC Products Division at Broadcom, says, “We are happy to support the Tensordyne team as our partner & customer. They are leveraging Broadcom's fundamental IP Platform, advanced Silicon and Packaging design technologies to bring their 3nm AI Inference Accelerator to market—focused on achieving industry-leading performance and power efficiencies."

“Cirrascale is looking forward to introducing the Tensordyne Napier system to our customers to unlock a new level of energy-efficient, high-performance AI infrastructure at scale. What Tensordyne is building directly addresses the pent-up demand for alternative technologies that provide fast, cost-efficient inference,” adds Dave Driggers, CEO of Cirrascale Cloud Services, a Tensordyne customer.

“Tensordyne’s innovations across the hardware inference stack unlock a new paradigm in performance, energy efficiency, and cost optimization—directly advancing BlueSky’s mission to deploy a new generation of ultra-efficient AI infrastructure at global scale.”
~ Ian Hartley, BlueSky Compute

Kevin Johnson, Tensordyne investor and board member, adds, "The AI era has unleashed the largest capital investment cycle in history. In pursuit of scalable, profitable AI inference infrastructure, Tensordyne delivers significant improvements in compute power efficiency and speed, which is a force multiplier for hyperscalers and data center operators.” Johnson is the former CEO of Starbucks and a current Goldman Sachs board member.

About Tensordyne

Tensordyne builds AI chips and rack-level systems for data centers to solve the trade-off between speed and cost in inference. In partnership with HPE Juniper Networks, Tensordyne developed Tensordyne Napier (TDN), an AI inference platform designed to deliver higher throughput, faster user response times, and improved infrastructure economics for hyperscalers, neoclouds, and enterprise AI operators.

Tensordyne is headquartered in Sunnyvale, California and Munich, Germany.

‍

For media enquiries, please contact Nara Communications

Frankie McGovern – frankie@naracommunications.com
Yeona Choi – yeona@naracommunications.com