# Tensordyne

> Tensordyne is a Sunnyvale- and Munich-based AI hardware company building inference systems that use a hardware-native logarithmic number system to cut the cost and power of running large AI models. Formerly the computer-vision startup Recogni, it rebranded in September 2025 and in June 2026 unveiled Napier (TDN), a 3nm, air-cooled inference platform it claims delivers roughly 13x the throughput and 17x the efficiency of Nvidia's GB300 NVL72 rack. The company positions itself as a direct challenger to Nvidia in the market for profitable, high-throughput generative-AI inference.

- **Founded:** 2017
- **Headquarters:** Sunnyvale, California, United States
- **Founders:** R.K. Anand (Co-Founder & Chief Product Officer), Gilles Backhus (Co-Founder & Head of AI), Emily Stuart (Co-Founder)
- **Team size:** ~150 employees
- **Products:** Tensordyne Napier (TDN), TDN AIP (Napier accelerator), TDN72 pod / rack, TDN LINK, TDN Math (Logarithmic Number System)
- **Notable:** Rebranded from Recogni to Tensordyne in September 2025 to reflect a pivot from computer vision to generative-AI inference systems., Completed tape-out of the Napier processor on TSMC's 3nm node (June 2026)., Claims 13x throughput and 17x energy efficiency versus Nvidia's GB300 NVL72 rack using logarithmic math.

## Products & services

- **Tensordyne Napier (TDN)** — A next-generation AI inference system announced June 2026, built on TSMC's 3nm process and co-designed with Broadcom. Named after John Napier, inventor of logarithms. Uses a hardware logarithmic number system to deliver high per-user speed and rack throughput with air cooling.
- **TDN AIP (Napier accelerator)** — The Napier processor: a 3nm chip with roughly 138 billion transistors, ~2.1 petaflops per die, a 1.33GHz accelerator core, 1.5GHz CPU, 256MB SRAM, and 144GB of HBM3E.
- **TDN72 pod / rack** — A system uniting 72 accelerators for ~68 petaflops of compute and ~42TB of HBM per pod; four pods form a single inference rack delivering 608 PFLOPS of dense compute.
- **TDN LINK** — An ultra-low-latency scale-up interconnect enabling linear scaling across a 72-node pod.
- **TDN Math (Logarithmic Number System)** — The company's patented logarithmic math architecture (filed 2019) that converts multiplication into addition in hardware, claimed to cut energy per operation by up to ~22x versus 16-bit floating point.

## Achievements

- Rebranded from Recogni to Tensordyne in September 2025 to reflect a pivot from computer vision to generative-AI inference systems.
- Completed tape-out of the Napier processor on TSMC's 3nm node (June 2026).
- Claims 13x throughput and 17x energy efficiency versus Nvidia's GB300 NVL72 rack using logarithmic math.
- Raised approximately $176M across rounds including a $102M Series C.
- Expects more than $200M in orders for the Napier inference system.

## Latest updates

- **2026-06** — Announced Tensordyne Napier (TDN), claiming to end AI's speed-vs-cost trade-off; disclosed 3nm tape-out at TSMC and partnerships with Broadcom and HPE Juniper Networks.
- **2026-06** — Said it expects more than $200M in orders for Napier and is preparing a Series D round; positioned as a direct challenger to Nvidia.
- **2025-09** — Rebranded from Recogni to Tensordyne, reflecting expansion from vision recognition into full-stack generative-AI compute.

## Links

- Website: https://tensordyne.ai
- LinkedIn: https://www.linkedin.com/company/tensordyne
- Twitter/X: https://twitter.com/Recogni_Inc/
- YouTube: https://www.youtube.com/@Tensordyne
- Instagram: https://www.instagram.com/TensordyneInc
- Facebook: https://facebook.com/recogniai

---

Profile page: https://yespress.io/tensordyne
Published by YesPress — https://yespress.io
Last updated: 2026-07-10