Qualcomm Looks to Tenstorrent to Accelerate Its Move into Data Centers

Qualcomm wants to stop being seen only as the leading mobile chip company. Its entry into laptops, automotive, edge AI, and now data centers points to a broader strategy: bringing its expertise in efficient computing to markets where NVIDIA, AMD, Broadcom, Google, and other hyperscalers are shaping the new AI infrastructure.

In this context, the possible acquisition of Tenstorrent, the chip startup led by Jim Keller, fits. According to reports from financial media, Qualcomm is in talks to acquire the company for a valuation between $8 billion and $10 billion. No agreement has been announced, and negotiations could change or not reach closure, but the deal clearly has an obvious industrial logic.

Tenstorrent is not just any startup. Its proposal combines AI accelerators based on its Tensix Cores, RISC-V architecture, open software, and intellectual property licenses for clients who want to design their own silicon. For Qualcomm, acquiring it would mean more than just adding another product. It would bring in team expertise, architecture, IP, software, and a technical brand strongly associated with Jim Keller, one of the most renowned engineers in the industry.

From smartphones to AI racks

Qualcomm has been trying for years to reduce its dependence on the mobile market. Snapdragon remains its most recognizable business, but the company has sought growth in automotive, Windows on Arm PCs, extended reality, connectivity, Internet of Things, and more recently, AI infrastructure.

The clearest step so far has been with AI200 and AI250, their inference solutions for data centers. AI200 is expected for commercial deployment in 2026 and AI250 in 2027. Qualcomm presents these platforms as rack-scale systems, not just isolated cards. AI200 integrates accelerators, memory, interconnection, and management software to run large models with a focus on efficiency and operational cost.

The challenge is enormous. Entering the data center market isn’t just about having an efficient chip. It requires a complete platform, framework support, developer tools, integration with orchestrators, operational reliability, supply chains, and customers willing to choose an alternative to NVIDIA. This is where Tenstorrent could contribute something Qualcomm cannot build overnight: a specialized team in AI architectures and a culture more aligned with open hardware and programmable designs.

Timing also plays a role. If Qualcomm aims to be part of the inference conversation by 2026 and 2027, an acquisition could accelerate capabilities that would otherwise take longer to develop organically. The company already has financial strength, expertise in SoC design, connectivity, and energy efficiency. What it needs is credibility in a market where software and roadmap continuity are scrutinized closely by buyers.

What Tenstorrent brings to the table

In December 2024, Tenstorrent closed a Series D round of over $693 million, with a pre-money valuation of $2 billion. The round was led by Samsung Securities and AFW Partners, with investors including LG Technology Ventures, Hyundai Motor Group, Fidelity, Baillie Gifford, and Bezos Expeditions, among others.

At that time, the company announced it had secured agreements totaling around $150 million, a noteworthy figure indicating commercial traction beyond just technological innovation. It also stated that the funds would be used to build its open software stack, hire developers, expand R&D centers, and develop systems and clouds for AI developers.

Its most visible product on the cards market is Blackhole. The Blackhole p150a features 120 Tensix Cores, 32 GB of GDDR6 memory, 16 large RISC-V cores, 180 MB of SRAM, up to 300 W power consumption, and four QSFP-DD 800G ports for linking multiple cards, grouping memory, and enhancing performance in larger configurations.

Tenstorrent does not compete by copying the dominant GPU model exactly. Its approach is to offer accelerators and licenses for clients seeking more control over their technology. This strategy could appeal to companies looking for alternatives to closed solutions, especially if they need custom silicon for inference, edge computing, automotive, servers, or proprietary products.

Jim Keller as a strategic asset

Jim Keller’s name carries significant weight in this operation. Keller has worked at AMD, Apple, Tesla, and Intel, and is associated with projects such as AMD Zen, Apple’s A4 and A5 chips, and Tesla’s autonomous driving hardware. He became CEO of Tenstorrent in 2023, after initially joining as CTO.

For Qualcomm, bringing Keller and the Tenstorrent team onboard could add talent in CPU architecture, RISC-V, AI acceleration, and system design. In an industry where top engineers are nearly as valuable as patents, this is no minor detail.

The acquisition would also serve a defensive purpose. If Qualcomm does not acquire Tenstorrent, another player might. Intel has already been mentioned as a potential interested party. For anyone looking to strengthen its position in AI accelerators and alternative architectures, Tenstorrent is an attractive piece.

Is the valuation of $8–10 billion justified?

The valuation being discussed, between $8 billion and $10 billion, represents a significant jump from the pre-money valuation of $2 billion in the December 2024 Series D. While it might seem aggressive, the AI market has changed drastically in just a year and a half. Talent, IP, roadmaps, and software have become more expensive, as all major players seek to reduce dependence on NVIDIA and capture a bigger share of inference spending.

The key question is whether Qualcomm is purchasing current revenues or future capacity. If only current business is considered, the figure seems very high. If viewed as a way to accelerate entry into the AI data center market, secure a unique team, and add RISC-V and accelerator IP to its portfolio, the deal aligns better.

Nonetheless, there are risks. Integrating a technical startup into a large corporation can slow down its pace. AI software is complex, and competing against CUDA, ROCm, TPU stacks, or in-house developments of hyperscalers requires years of effort. Moreover, the AI accelerator market is full of promises that often don’t reach mass production.

Why this deal matters to the market

A purchase of Tenstorrent by Qualcomm would signal clearly: the AI chip market won’t be limited to NVIDIA versus AMD. There will be a layer of competitors focused on inference, memory, efficiency, alternative architectures, RISC-V, and more customizable solutions.

Qualcomm’s expertise in energy efficiency, SoC integration, connectivity, and high-volume platforms combined with Tenstorrent’s AI architecture, open software, RISC-V, and a narrative of technological independence could create an alternative tailored for clients who don’t necessarily need to train enormous models but do require efficient, scalable inference at lower costs per query.

The deal would also have implications for RISC-V. Qualcomm is already familiar with Arm and has built for years on licensed architectures, but RISC-V offers a more open and customizable approach. Tenstorrent has made this combination of AI and RISC-V central to its identity. In an era where governments and companies seek greater control over their technological supply chain, this aspect could increase in value.

A deal still uncertain

For now, all of this should be seen as negotiation, not as a confirmed deal. Qualcomm and Tenstorrent have not announced any agreement. Valuation could change, terms might include milestone payments, and the transaction might not materialize. Moreover, any deal of this size will be scrutinized by investors and likely regulators.

If finalized, Qualcomm would make a significant step forward in AI infrastructure. If not, the news still highlights something important: Tenstorrent has become one of the most sought-after startups in the sector, and Qualcomm aims to position itself among the major providers of AI compute solutions.

The race is no longer just about who has the most powerful accelerator. It’s about who can deliver better cost per token, higher efficiency per watt, more memory, greater software control, and more freedom to customize architectures. Qualcomm understands that its future cannot rely solely on smartphones. Tenstorrent could be a fast, though costly, way to prove it.

Frequently Asked Questions

What is Qualcomm negotiating?

According to financial sources, Qualcomm is in talks to acquire Tenstorrent for between $8 billion and $10 billion. No official agreement has been announced.

What does Tenstorrent do?

Tenstorrent designs AI hardware based on Tensix Cores, sells systems and accelerator cards, develops open software, and licenses proprietary AI and RISC-V IP.

Why is Jim Keller important?

Jim Keller is one of the industry’s most renowned chip architects. He has worked at AMD, Apple, Tesla, and Intel, and has led projects like AMD Zen, Apple’s A4/A5 chips, and Tesla’s autonomous hardware. He became CEO of Tenstorrent in 2023, after serving as CTO.

How does Tenstorrent fit with Qualcomm?

Qualcomm aims to strengthen its presence in data center inference with AI200 and AI250. Tenstorrent could contribute architecture, software, RISC-V, technical team, and IP to accelerate this strategy.

via: The Information