In 2022, Rivos gained notoriety when Apple sued it for trade secrets, claiming that Rivos had hired off numerous Apple engineers and had exploited private information to create processors that might compete with the iPhone maker’s own.
Rivos filed a countersuit against Apple for unfair competition, refuting the accusations. In the end, Apple settled the legal dispute in February. Simultaneously, it resolved individual legal disputes with multiple Apple engineers that Rivos had employed.
CEO Puneet Kumar told TechCrunch that Rivos is now stepping up its efforts to bring its chipset technology to market, putting the trial drama behind it.
“Rivos was founded with the goal of creating the most high-performance, power-efficient chips in the industry,” Kumar stated. “We are thrilled to be focusing on clients who are developing data-driven solutions.”
A significant new Financing tranches will assist in funding those initiatives.
On Tuesday, Rivos reported that it had acquired more than $250 million in an extended Series A that was oversubscribed and headed by Matrix Capital Management. Chip giants MediaTek and Intel (via its corporate VC branch) were among the participants in the round. Koch Disruptive Technologies, Dell Technologies Capital, Walden Catalyst, Cambium Capital, and Hotung Venture Group were among the other funders.
It’s quite the comeback for Rivos, which was established in 2021 and was having trouble hiring staff members and obtaining investment capital around a year ago due to the Apple lawsuit. Almost two dozen workers, or 6% of Rivos’s total employment at the time, were let go in August, and the company was forced to postpone a $400 billion Series An informational fundraising campaign, The Information Β reported at the time.
A specially designed server chip
Building chips primarily for servers that can perform demanding data analytics and AI workloads, especially generative AI workloads, is the long-term objective of Rivos, according to Kumar.
“Our goal is to serve clients who are developing data-driven solutions, such as those who use data analytics and generative AI to inform choices,” Kumar stated. “Many businesses are vying for these kinds of markets; Rivos supports the demanding hardware needs of the AI models and analytics that will completely transform the enterprise.”
RISC-V, an open standard instruction set architecture, serves as the foundation for Rivos’ initial chipset (ISA).
Every semiconductor has an ISA, a technical specification that describes how software manages the hardware of the device. Chip design teams usually licence an existing ISA from an incumbent (e.g. Arm or Intel) for general-purpose computing. However, RISC-V offers a free, open-source substitute with no strings attached.
To expedite AI and big data-related computations, Rivos’ chip has what Kumar refers to as a “data parallel accelerator,” which is essentially a GPU tailored for uses other than graphics processing. It was created utilising TSMC’s 3nm manufacturing technique. “Process” in chip manufacturing refers to the smallest possible component size that can be integrated into a chip.
That 3 nm is regarded as being near the state of the art. Apple was the only business to adopt TSMC’s process in 2024 with its M3 chipset series, but other companies including Qualcomm, MediaTek, Nvidia, and AMD are all anticipated to use it for their upcoming chip families.
Apart from creating the chip, Rivos is developing gear for standalone data centres that adheres to the Open Compute Project’s modular specification. This will essentially function as chip casing that is plug-and-play. Additionally, according to Kumar, it’s developing a “firmware-to-app” software stack for chip programming.
“Workloads from customers can be quickly deployed on our more energy-efficient hardware while utilising their current models and databases, providing them with an instant advantage,” Kumar continued.
Currently in pre-revenue status, Rivos intends to generate revenue by charging clients, primarily big data centre operators, for its hardware and ancillary software solutions. Rivos’ “low-friction” adoption pipeline, according to early investor David Goel, is a crucial advantage in the very competitive chip business.
Goel told TechCrunch, “The Rivos team has skillfully integrated the revolutionary new RISC-V architecture with an innovative accelerator, effectively bringing this vision to life.” “Their prototype chip provides an engaging example of
of their unique capability.β
But is it differentiating enough?
Stern rivalry
As the generative AI boom continues, huge IT corporations, one of Rivos’ prospective customer categories, are racing to develop their own in-house chips for AI and big data analytics.
Googleβs on its fifth-genΒ TPUΒ and recently revealedΒ Axion, its first dedicated chip for running models. Amazon hasΒ severalΒ customΒ chipΒ families under its belt. Microsoft last yearΒ jumped into the frayΒ with theΒ Azure MaiaΒ AI Accelerator and the Azure Cobalt 100 CPU. And MetaβsΒ inching alongΒ with its own designs.
Startups by the dozens, meanwhile, areΒ angling for a slice of a custom data center chip marketΒ that could reach $10 billion this year and double by 2025.
Recently, Groq, a startup that creates chips to run AI models faster than traditional hardware, established a new business unit focused on enterprise use cases and applications. Leading engineer Jim Keller’s AI hardware business Tenstorrent wants to integrate its chipsets inside data centres. Additionally, to increase production of its Atom chip, which is targeted at data centres, Rebellions, a South Korean fabless AI chip company, has raised hundreds of millions of dollars in funding.
However, Nvidia is currently the dominant power in semiconductors and is proving to be difficult to unseat.
Riding high on the demand for its GPUs for AI training this year, Nvidia momentarily became a $2 trillion corporation. According to Wells Fargo Equity Research, Nvidia holds a 98% market share in the data centre.
centre GPUs, and as Nvidia constructs a new division to create customised chips for cloud computing companies and others, the company’s data centre revenue increased by more than 400% in Q4 2023.
Some bespoke server chip upstarts have found it difficult to get funding due to the intense rivalry and the impact of Nvidia’s dominance.
Following the collapse of a proposed merger with Microsoft, Graphcore reportedly saw a $1 billion reduction in its worth. A few months ago, the company said that it was preparing to lay off employees due to the “extremely challenging” macroeconomic situation. 10% of the employees at Intel-owned AI chip maker Habana Labs were let go last year. Like the RISC-V firm Rivos, SiFive also let rid of 20% of its workforce last year.
of its employees and ended its main product line.
Will Rivos do any better, then? Perhaps.
Customers were not discussed by Kumar, and it is not expected that Rivos’ chip will be produced in large quantities until later in the coming year. Nonetheless, Kumar claimed that Rivos is well-positioned to increase manufacturing and focus more on platform and software innovation given its 375 workers and hundreds of millions of dollars in cash on hand.
“It is critical that accelerators be easy to programme and debug, and that data can move between CPU and accelerator seamlessly, given the rapid changes in generative AI and the merger with the data analytics stack,” Kumar stated. “Rivos uses our’recompile-not-redesign’ approach to address this need.”
