Qualcomm to Acquire Chip Legend Jim Keller's Company for $10 Billion

According to foreign media The Information, Qualcomm, a global mobile chip giant, is in acquisition talks with AI chip startup Tenstorrent. The valuation of this transaction is approximately $8 billion to $10 billion. Compared with the company's valuation of $2.6 billion at the end of last year, the premium is nearly four times, and the acquisition cost is extremely high. Once finalized, it will become one of the largest acquisition cases in the global AI chip field in the past three years.

As a rising star in the chip industry led by Jim Keller, a legendary chip designer in the semiconductor industry known as the "Silicon Sage", Tenstorrent has gained a great reputation in the industry for its RISC-V architecture and AI accelerator design. This also shows Qualcomm's strategic determination to continue to break away from smartphone chips.

01

The primary motivation for Qualcomm to suddenly initiate this acquisition is the urgency to seek a diversified business transformation. For a long time, Qualcomm's main revenue has been highly tied to smartphone chips.

However, in recent years, the global smartphone market has reached its peak in terms of dividends, and major terminal manufacturers have also launched their own chip R & D projects to replace Qualcomm's Snapdragon chips.

In order to explore new growth curves, Qualcomm has actively turned its attention to fields such as the automotive industry, data centers, and cloud AI computing in recent years.

Although Qualcomm has excellent AI processing capabilities at the edge, it has always lacked a strong technical product matrix in the computing power center market for large - model training and cloud inference. If it starts to develop server - level AI chips from scratch, the long R & D cycle may cause it to miss the golden window period of the AI era. Therefore, acquiring Tenstorrent at a high premium can enable Qualcomm to make up for its technical shortcomings in the cloud and data centers, directly skip the self - R & D stage, and quickly enter the high - end AI computing power market.

Secondly, Tenstorrent's unique "anti - NVIDIA" technology route and cost - effective system - level architecture can provide Qualcomm with a differentiated competitive weapon.

Currently, NVIDIA's dominance in the AI computing power field highly depends on expensive HBM memory and proprietary NVLink interconnection technology. However, Tenstorrent, led by Jim Keller, does the opposite. As early as 2024, it publicly stated that the key to the future AI chip revolution lies in "abandoning HBM".

In order to break through the high - cost barrier, Tenstorrent chooses smaller and lower - cost chips in hardware, using ordinary GDDR6 video memory and a large amount of on - chip SRAM to replace the expensive HBM with limited packaging production capacity.

In terms of cluster interconnection, Tenstorrent relies on standard high - speed Ethernet to connect accelerators into a distributed cluster. Taking the Galaxy Blackhole AI computing platform as an example, the single - platform price of a platform containing 32 accelerators is only $110,000. Although its absolute speed is not as fast as NVIDIA's DGX, its price is only one - third to one - fifth of the latter, and its efficiency has increased several times.

If Qualcomm can acquire this low - cost, high - throughput architecture, it can naturally provide an attractive alternative solution to customers suffering from NVIDIA's high costs.

In addition to the AI acceleration platform, Tenstorrent's technical accumulation in the field of high - performance RISC - V CPUs is also a key factor attracting Qualcomm.

In the past, the RISC - V ecosystem has always lacked high - performance products that can be truly applied to high - end servers. The TT - Ascalon high - performance RISC - V CPU launched by Tenstorrent fills this gap. Its top - end Ascalon - X core is powerful enough to directly compete with Arm's Neoverse V2 and V3 cores. More importantly, Tenstorrent adheres to a completely open strategy, providing an open - source software stack and licensing its AI core and RISC - V CPU IP externally.

For Qualcomm, it has long been restricted by lawsuits and licensing terms with Arm during the process of self - R & D of micro - architectures. If it acquires Tenstorrent's top - notch RISC - V IP, it means that Qualcomm can also choose a new path.

02

Interestingly, Tenstorrent's layout in the scenarios of intelligent vehicles and edge computing also matches Qualcomm's strategy of building the "Snapdragon Digital Chassis" very well. Currently, Tenstorrent is promoting the automotive version "Alexandria" of its RISC - V CPU. This version adds a core feature set for automotive functional safety, targeting the advanced driver - assistance system (ADAS) and automotive high - performance computing markets. If Qualcomm can combine Tenstorrent's existing products with its existing smart cockpit and autonomous driving chips, it can further expand into niche markets such as industry and healthcare, where Arm rarely participates, and achieve a more extensive edge AI ecosystem coverage.

Of course, this billion - dollar bet has also triggered some cautious sentiment in the capital market. After the acquisition news spread, Qualcomm's after - hours stock price in the US market fell slightly by about 1%.

Most investment institutions believe that the valuation is too high, and the technology in the AI chip industry is changing rapidly. Whether the high acquisition cost can be smoothly converted into revenue still faces numerous challenges such as technical integration, team retention, and commercialization.

In addition, due to Tenstorrent's strong open - source gene and high team independence, it is a major test to integrate it into Qualcomm's huge business empire. To balance risks, Qualcomm is very likely to introduce a "performance milestone payment" mechanism with floating payments in the final transaction structure, paying in installments according to Tenstorrent's subsequent technology implementation and revenue targets.

This article is from the WeChat official account "Mkeji.com". The author is Mkeji.com. It is published by 36Kr with authorization.