Graphics processing units (GPUs) — the workhorses behind most AI models — are undeniably powerful, but they come with a glaring downside: they consume massive amounts of energy. Goldman Sachs projects that as GPUs proliferate in data centers to power AI applications, electricity demand will soar by 160% by 2030. This trajectory poses a significant sustainability challenge.
Enter Sagence AI, a trailblazing startup founded by veteran chip designer Vishal Sarin. Sarin, an analog and memory circuit expert with over a decade of industry experience, envisions a future where AI computing is both powerful and energy-efficient. Sagence, formerly known as Analog Inference, is pioneering analog chips that aim to replace energy-hungry GPUs for specific AI workloads.
A Radical Approach to AI Hardware
“The applications that could make practical AI truly pervasive are restricted because today’s systems can’t deliver the required performance,” Sarin explains. “Our mission is to overcome these limitations — economically, efficiently, and responsibly.”
While numerous companies are creating custom AI hardware, Sagence stands apart with its unique analog chip design. Unlike traditional digital chips, which store data as binary ones and zeros, analog chips use a continuous range of values to represent data. This fundamental difference enables significant advantages in energy efficiency and data density.
Analog technology isn’t new; it powered engineering marvels like modeling North America’s electrical grid during its heyday from the 1930s to the 1980s. However, digital chips eventually took over due to their scalability and precision. But the growing energy and performance bottlenecks in digital systems are reigniting interest in analog computing.
Breaking the Bottlenecks
Traditional digital chips, including GPUs, require hundreds of components to perform calculations that analog chips can achieve with just a few modules. Moreover, digital chips often shuttle data back and forth between memory and processors, creating bottlenecks that slow performance and increase power consumption.
Sagence’s analog chips take a different approach. As “in-memory” chips, they store and process data directly in memory, eliminating the need for constant data transfer. This innovation allows them to complete tasks faster and with significantly less power. Additionally, their ability to store data in a continuous range of values boosts data density, meaning more information can be processed in the same physical space.
“All the leading suppliers of AI silicon are using outdated architectures,” Sarin argues. “These legacy approaches are blocking progress in AI adoption. Sagence’s products are designed to overcome these limitations by delivering high performance with lower power and cost.”
Challenges of Analog
Despite its promise, analog technology isn’t without challenges. Achieving high precision with analog chips is harder because they require more accurate manufacturing processes. Programming these chips is also more complex, which can slow adoption.
However, Sarin sees Sagence’s analog chips as complementary to, rather than outright replacements for, digital chips. They’re particularly well-suited for specialized applications in servers and mobile devices, where power and latency are critical concerns.
“Our goal isn’t to completely disrupt the digital ecosystem but to enhance it with analog solutions that address its shortcomings,” Sarin says.
From Lab to Market
Sagence plans to bring its first chips to market in 2025. The company is already working with multiple customers to refine its technology and integrate it into existing infrastructure. Unlike competitors like EnCharge and Mythic, Sagence is focused on creating system-level products that seamlessly fit into current deployment scenarios.
The company’s efforts have garnered strong backing. Sagence has raised $58 million in funding from high-profile investors, including Vinod Khosla, TDK Ventures, Cambium Capital, Blue Ivy Ventures, Aramco Ventures, and New Science Ventures. Now, with a 75-person team, Sagence is preparing for its next fundraising round to scale production and expand its workforce.
“Our cost structure is a significant advantage,” Sarin notes. “We’re not chasing performance by adopting the latest and most expensive manufacturing processes. This keeps our production costs manageable.”
Timing and Market Potential
Sagence’s timing could be fortuitous. After a slow 2023, semiconductor startups are seeing renewed interest from venture capital. From January to July 2024, chip startups raised nearly $5.3 billion, well ahead of last year’s total of $8.8 billion.
That said, chipmaking remains an expensive and competitive endeavor. International sanctions, tariffs, and entrenched ecosystems like Nvidia’s present significant hurdles. Even well-funded startups like Graphcore, which raised nearly $700 million, have struggled to carve out a market share, with Graphcore filing for insolvency in 2023.
To succeed, Sagence must prove its analog chips can deliver dramatically lower power consumption and superior efficiency at scale. It will also need to navigate the capital-intensive process of manufacturing and win over customers locked into competing ecosystems.
The Future of AI Hardware
Despite the challenges, Sarin remains optimistic. “We’re packaging our core technology into products that meet real-world demands,” he says. “Our chips are designed to redefine what’s possible in AI hardware while addressing the pressing issues of power, cost, and latency.”
If Sagence can deliver on its promises, it has the potential to reshape the AI hardware landscape, offering a greener, faster, and more efficient alternative to GPUs. With its unique analog approach and a growing roster of investors, Sagence could be the company to watch as the demand for sustainable AI computing continues to rise.
