According to a report by The Information, global mobile chip giant Qualcomm is in acquisition talks with AI chip startup Tenstorrent. The deal is valued at approximately $8 to $10 billion, representing a nearly fourfold premium compared to the company's $2.6 billion valuation at the end of last year. The acquisition cost is extremely high, and if finalized, it would be one of the largest M&A deals in the global AI chip sector in nearly three years.
Led by legendary chip designer "Silicon Sage" Jim Keller, Tenstorrent has gained significant recognition in the industry for its RISC-V architecture and AI accelerator designs. This move also highlights Qualcomm's determination to further diversify beyond smartphone chips.
01
Qualcomm's sudden pursuit of this acquisition is primarily driven by the urgency of seeking business diversification. For a long time, Qualcomm's main revenue has been heavily tied to smartphone chips.
However, in recent years, global smartphone market growth has peaked, and major terminal manufacturers have also launched their own in-house chip projects aiming to replace Snapdragon chips.
To explore new growth avenues, Qualcomm has actively turned its focus in recent years to areas such as automotive, data centers, and cloud AI computing.
Although Qualcomm possesses strong AI processing capabilities at the edge, it has lacked a powerful technical product portfolio in the data center market for large model training and cloud inference. Starting from scratch to develop server-grade AI chips would involve a lengthy R&D cycle, potentially causing it to miss the golden window of the AI era. Therefore, acquiring Tenstorrent at a high premium would allow Qualcomm to fill its own technological gaps in cloud and data centers, directly bypassing the in-house development phase and quickly entering the high-end AI computing market.
Secondly, Tenstorrent's unique "anti-NVIDIA" technical approach and cost-effective system-level architecture can provide Qualcomm with differentiated competitive weapons.
Currently, NVIDIA's dominance in the AI computing field heavily relies on expensive HBM memory and proprietary NVLink interconnect technology. In contrast, Jim Keller's Tenstorrent has taken the opposite path, publicly stating as early as 2024 that the key to the future AI chip revolution lies in "abandoning HBM."
To break down high-cost barriers, Tenstorrent opts for smaller, lower-cost chips in hardware, utilizing standard GDDR6 memory and large amounts of on-chip SRAM to replace expensive and packaging-capacity-constrained HBM.
For cluster interconnection, Tenstorrent relies on standard high-speed Ethernet to connect accelerators into distributed clusters. Taking the Galaxy Blackhole AI computing platform as an example, a single platform containing 32 accelerators is priced at only $110,000. While its absolute speed may not match NVIDIA's DGX, its price is only one-third to one-fifth of the latter, yet its efficiency is several times higher.
If Qualcomm can incorporate this low-cost, high-throughput architecture, it would naturally be able to offer a highly attractive alternative to customers struggling with NVIDIA's high costs.
Beyond the AI acceleration platform, Tenstorrent's technological accumulation in high-performance RISC-V CPUs is also a key asset attracting Qualcomm.
In the past, the RISC-V ecosystem lacked truly high-performance products applicable to high-end servers. Tenstorrent's TT-Ascalon high-performance RISC-V CPU fills this gap, with its highest-end Ascalon-X core boasting performance directly comparable to Arm's Neoverse V2 and V3 cores. More importantly, Tenstorrent adheres to a fully open strategy, providing open-source software stacks and licensing its AI core and RISC-V CPU IP externally.
For Qualcomm, which has long faced constraints from lawsuits and licensing terms with Arm during its own microarchitecture development, acquiring Tenstorrent's top-tier RISC-V IP would mean gaining the option for a new path.
02
Interestingly, Tenstorrent's layout in intelligent automotive and edge computing scenarios also aligns well with Qualcomm's strategic push for the "Snapdragon Digital Chassis." Currently, Tenstorrent is advancing its automotive version of the RISC-V CPU, "Alexandria," which incorporates core functional safety features targeting the advanced driver-assistance systems (ADAS) and automotive high-performance computing markets. If Qualcomm can integrate Tenstorrent's existing products with its current intelligent cockpit and autonomous driving chips, it could further expand into industrial and medical niche markets where Arm has little presence, achieving broader edge AI ecosystem coverage.
Of course, this $10-billion-level gamble has also sparked cautious sentiment in the capital market. After the acquisition news broke, Qualcomm's after-hours stock price dipped slightly by about 1%.
Most investment institutions consider the high valuation excessive. Moreover, given the rapid technological iteration in the AI chip industry, whether the substantial acquisition cost can be smoothly translated into revenue still faces numerous challenges, including technology integration, team retention, and commercialization.
Additionally, as Tenstorrent possesses a strong open-source culture and high team independence, integrating it into Qualcomm's vast commercial empire poses a significant test. To balance risks, Qualcomm is highly likely to incorporate a floating payment mechanism based on "earn-out milestones" in the final transaction structure, making payments in installments based on Tenstorrent's subsequent technology deployment and revenue targets.
This article is from the WeChat public account "镁客网," author: 镁客网
