Tenstorrent has closed a $1.5 billion Series E equity round at a $6.2 billion post-money valuation, the AI computing startup said in an October 19 announcement. Trace Capital and Innovation X led the financing.
The company plans to use the money for volume manufacturing, wider commercial expansion and large-scale deployment of its Galaxy servers. It will also develop the AI and CPU intellectual property that it licenses to customers designing their own chips.
Participating investors include ARK Invest, Fidelity Management & Research Company, Cadence Ventures and Sandisk. The announcement describes a closed equity raise, not a future fundraising target. It does not give an India-specific allocation or a timetable for deploying the new capital.
How the financing has changed
The earlier Series D announcement, dated December 2, 2024, reported more than $693 million raised, led by Samsung Securities and AFW Partners. That release quoted a $2 billion pre-money valuation. The old and new headline valuations therefore use different bases: the earlier figure excluded the incoming investment, while the latest includes it.
The 2024 round was intended to support open-source AI software, developer hiring, expansion of design and development centres, and systems and cloud infrastructure. Tenstorrent already combined hardware sales with licensing AI and RISC-V designs. The new financing continues that business model while putting more emphasis on manufacturing and commercial scale.
Servers, rather than just chip designs
The company's current Galaxy specifications show the physical infrastructure behind that expansion. Its Blackhole system lists 32 accelerator chips and one terabyte of GDDR6 accelerator memory in a six-unit, air-cooled rack chassis. It also offers four-server superclusters and describes configurations extending to 36 or more Galaxy systems.
Those systems require data-centre infrastructure, not simply space beside a desktop. The Blackhole specification lists average system power use of eight to 10 kilowatts and a 119-kilogram system weight. The manufacturer also specifies an operating temperature of 10 to 35 degrees Celsius and lists a dedicated management-network connection. Power, cooling and equipment management are therefore part of the deployment requirements.
The company says its Tensix Neo architecture is designed to use the same programming model across chips and Ethernet-connected racks. Its compiler-to-kernel software is open source, so deployment includes a software layer as well as specialised processors.
An existing engineering link to India
An October 8 partnership statement provides a more specific Indian connection. Vayavya Labs, headquartered in Belagavi with a Bengaluru development centre, has built virtual models of Tenstorrent hardware subsystems since 2024. Its work includes security processors, system-management logic and direct-memory-access functions.
These virtual platforms let software and validation teams develop and debug against a representation of the hardware before physical chips are ready. The companies say this helps identify integration problems earlier. The models cover boot and fuse infrastructure, secure memory zeroing and a hardware security module for cryptographic operations, giving teams several subsystems to validate together.
Vayavya said it had most recently delivered a phased release of a security-enclave virtual platform on two independent simulation targets, then added system-management capabilities. The collaboration predates the Series E. The models use Vayavya's SystemC modelling library, with an Accellera-compatible path and a commercial version tuned to Tenstorrent's simulation environment.
Voice software offers another deployment example
A separate October 2 announcement with Smallest AI describes an application for the server platform: running the Lightning V2 text-to-speech model natively on Galaxy Blackhole machines. The companies presented it as an on-premises voice-AI system for high-volume applications, including financial services, telecom and healthcare.
The stated design keeps voice-agent workloads on the customer's own infrastructure, giving organisations control over sensitive data instead of requiring those workloads to be hosted elsewhere. The announcement says Lightning V2 was already available on Tenstorrent hardware with usage-based pricing and no upfront commitments.
Smallest AI develops speech recognition, speech generation and speech-to-speech systems. The joint announcement identifies low latency and high concurrency as requirements for enterprise voice applications. This connects Tenstorrent's server expansion to a specific software workload: generating speech quickly enough for real-time conversations while handling many requests.
Dalimss News has separately covered RapidCanvas's enterprise AI financing and Tamr's data-management investment. Tenstorrent's round concerns the computing hardware and chip-design business underlying AI applications, rather than either of those software companies.
