Rivos made headlines in 2022 after Apple filed a trade secrets suit against it, which accused Rivos of hiring away dozens of Apple engineers and using confidential info to develop chips to rival the iPhone maker’s own.
The company denied the allegations and countersued Apple for unfair competition. Apple ended up settling its lawsuit in February. Around the same time, it ended separate litigation with several of the Apple engineers Rivos had hired.
Now, with the courtroom drama behind it, Rivos is redoubling its efforts to bring its chipset tech to market, CEO Puneet Kumar told TechCrunch.
“Rivos was founded with the mission of building industry-leading, power-efficient, high-performance chips,” Kumar said. “We’re excited to be targeting customers who are building data-driven solutions.”
A substantial new funding tranche will help to finance those efforts.
On Tuesday, Rivos announced that it raised over $250 million in an oversubscribed, extended Series A led by Matrix Capital Management with participation from chip giants, including Intel (via its corporate VC division) and MediaTek. Other backers included Cambium Capital, Hotung Venture Group, Walden Catalyst, Dell Technologies Capital and Koch Disruptive Technologies.
It’s quite the turnaround for Rivos, which was founded in 2021 and roughly a year ago was struggling to raise funds from investors and recruit employees under the shadow of the Apple suit. In August, Rivos laid off nearly two dozen employees, or 6% of its workforce at the time, and was forced to delay a planned $400 billion Series A fundraising round, The Information reported at the time.
A custom server chip
The long-term goal with Rivos, Kumar said, is to build chips primarily for servers that can handle intensive data analytics and AI workloads, including generative AI workloads.
“We’re targeting customers building data-driven solutions (e.g., those utilizing generative AI and data analytics to drive decisions),” Kumar said. “There’re many companies targeting such markets; Rivos supports the intense hardware requirements of the AI models and analytics that will remake the enterprise.”
Rivos’ first chipset is built on RISC-V, the open standard instruction set architecture (ISA).
ISAs are a technical spec at the foundation of every chip, describing how software controls the chip’s hardware. For general-purpose computing, chip design teams typically license an existing ISA from an incumbent (e.g., Arm or Intel). But RISC-V presents an open, no-royalties-attached alternative.
Rivos’ chip features what Kumar describes as a “data parallel accelerator” to speed up AI- and big data–related computations, essentially a GPU designed for purposes beyond graphics processing. It was made using TSMC’s 3nm fabrication process. In chip manufacturing, “process” refers to the size of the smallest component that can be embedded on a chip.
That 3nm is considered close to the cutting edge. While Qualcomm, MediaTek, Nvidia and AMD among others are expected to employ TSMC’s process for their upcoming chip families, Apple was the only company to use it in 2024 in its M3 chipset series.
In addition to building the chip, Rivos is working on self-contained data center hardware based on the Open Compute Project modular standard, which will effectively serve as plug-and-play chip housing. And it’s creating a “firmware-to-app” software stack for programming the chip, Kumar said.
“Customer workloads can be easily deployed on our more efficient hardware, but still using their existing models and databases, giving them an immediate benefit,” Kumar added.
Rivos, which is pre-revenue at the moment, plans to make money by charging customers — chiefly large data center operators — for its hardware and complementary software solutions. David Goel, an early investor, said that Rivos’ “low-friction” adoption pipeline is a key differentiator in the cutthroat chip market.
“The Rivos team has adeptly integrated the groundbreaking new RISC-V architecture with an inventive accelerator, effectively bringing this vision to life,” Goel told TechCrunch. “Their prototype chip serves as a compelling demonstration of their unique capability.”
But is it differentiating enough?
Stiff competition
Big Tech firms — one of Rivos’ potential customer segments — are racing to develop their own in-house chips for AI and big data analytics as the generative AI boom continues.
Google’s on its fifth-gen TPU and recently revealed Axion, its first dedicated chip for running models. Amazon has several custom chip families under its belt. Microsoft last year jumped into the fray with the Azure Maia AI Accelerator and the Azure Cobalt 100 CPU. And Meta’s inching along with its own designs.
Meanwhile, startups by the dozens are angling for a slice of a custom data center chip market that could reach $10 billion this year and double by 2025.
Groq, a company developing chips to run AI models faster than conventional hardware, recently formed a new business unit geared toward enterprise applications and use cases. AI hardware startup Tenstorrent, helmed by engineering luminary Jim Keller, is looking to build its chipsets into data centers. And Rebellions, a South Korean fabless AI chip firm, has raised hundreds of millions of dollars in capital to ramp up production of its data center–focused chip, Atom.
But Nvidia, the dominant force in chips right now, is proving to be a tough one to topple.
Nvidia briefly became a $2 trillion company this year, riding high on the demand for its GPUs for AI training. Wells Fargo Equity Research estimates that Nvidia has a 98% market share in data center GPUs, and the company’s data center business was up more than 400% in Q4 2023 as Nvidia builds a new unit to design bespoke chips for cloud computing firms and others.
Given the fierceness of the competition — and the chilling effect Nvidia’s supremacy has had on funding for would-be rivals — it’s been rough going for some custom server chip upstarts.
A few months ago, Graphcore, which reportedly had its valuation slashed by $1 billion after a deal with Microsoft fell through, said that it was planning job cuts due to the “extremely challenging” macroeconomic environment. Habana Labs, the Intel-owned AI chip company, laid off an estimated 10% of its workforce last year. Also last year, SiFive — like Rivos, an RISC-V startup — let go of 20% of its workforce and discontinued its core product line.
So will Rivos fare better? Maybe.
Kumar wouldn’t talk about customers, and Rivos’ chip isn’t anticipated to reach mass production until sometime next year. But with 375 employees and hundreds of millions of dollars in the bank, Kumar said that Rivos is well-positioned to expand manufacturing and double down on platform and software engineering.
“The rapid changes in generative AI and the merger with the data analytics stack makes it vital that accelerators be easy to program and debug, and that data can seamlessly move between CPU and accelerator,” Kumar said. “Rivos addresses this need through our ‘recompile-not-redesign’ approach.”
Comment