Fabless semiconductor company Astera Labs raises $50M for AI hardware designs

Astera Labs, a fabless semiconductor company headquartered in Silicon Valley, today announced that it raised $50 million in series C funding at a $950 million post-money valuation. Fidelity led the round with participation from Atreides Management, Valor Equity Partners, Avigdor Willenz Group, GlobalLink1 Capital, Intel Capital, Sutter Hill Ventures, and VentureTech Alliance.

CEO Jitendra Mohan says that the new capital will enable Astera to scale its operations and support the development of products that target chipset challenges in datacenters. Seventy-employee Astera is already generating “several millions” of dollars in revenue monthly, Mohan claims, and the gross margin profile is at or above competing semiconductor companies.

“We are driving innovation across several different vectors to support the explosive growth in the need of data-intensive workloads in the cloud and servers. Our competitors range from traditional semiconductor companies to smaller niche players depending on the specific product categories that we are addressing,” Mohan told VentureBeat via email. “At this point, we have established a decisive lead ahead of our competition with majority market share and are driving the industry with our purpose-built solutions.”

Astera Labs was founded in 2017 by Mohan, Casey Morrison, and Gajendra Akkasalamakki with the goal of addressing performance bottlenecks in datacenters with purpose-built “smart” connectivity solutions. The founding team previously worked together for more than 15 years at Texas Instruments, brainstorming various data path and connectivity problems in datacenters and clouds.

“We saw a massive opportunity to enable lower-latency interconnects to support the exploding data-intensive needs for AI and machine learning in the cloud. These specialized workloads have given rise to heterogeneous computing, where GPUs, AI processors, and other accelerators work side-by-side with general-purpose CPUs — a rapidly growing trend that’s redefining the connectivity backbone in the next generation of data centers,” Mohan explained. “We built Astera Labs’ product portfolio based on the latest PCIe, CXL, and Ethernet protocols with unwavering focus to enable higher bandwidth and better resource utilization for cloud service providers.”

Smart connectivity

Enterprises accelerated their adoption of AI and machine learning in 2021, during the pandemic. Forbes reports that 76% of enterprises are now prioritizing machine learning in their IT budgets. But AI workloads require the right infrastructure to execute —  a hardware sweet spot that Astera intends to corner, whether in the cloud, on-premises, or a mix of both.

“[Our] system-on-chip is composed of multicore massively parallel processors, dedicated hardware machines for analytics processing, I/O accelerators, and memory accelerators,” Mohan said. “Tests with a prototype of the chips on real life workloads … demonstrated improved performance by 20 times to 100 times compared to a CPU.”

Astera’s products support “single-digit nanosecond” latency through its custom silicon, increasing the available bandwidth for AI and machine learning applications. Among other hardware, Nvidia’s new DGX platforms use Astera’s technology to interconnect large arrays of GPUs. 

“Our first product offering — the Aries Smart Retimer portfolio for PCIe and CXL connectivity — enables data centers and cloud service providers to fully deploy system architectures that run AI and machine learning applications while reducing power consumption by as much as two times and using two times fewer components than alternative solutions,” Mohan said. “Additionally, we have built in a powerful array of signal monitoring features to allow system operators to detect early signs that a signal is beginning to degrade, quickly identify the exact location and nature of the issue, and work to replace the faulty component before it negatively impacts overall system availability. These added diagnostics capabilities make Astera Labs’ Retimers ‘smart,’ allowing us to rise above the crowded field of competing connectivity solutions to capture a majority of the market share in this space.”

Astera Labs, which has raised $85 million in venture capital to date, expects to be cash flow positive “soon” and double its headcount in the next 9 to 12 months following design wins at CPU, GPU, and AI processor platforms and cloud customers. In the meantime, Astera plans to expand its focus with purpose-built application-specific integrated circuit (ASICs) and module solutions for big data analytics use cases.

Astera says it expects the chip to be on the market in “several quarters,” following tests at undisclosed Fortune 500 companies.