Can AI Save Science?

AI Processors Using Dataflow Look Very Promising

The Genesis Mission is DOE’s big new story: a national push to use AI and world-class supercomputers to crack science’s toughest mysteries and supercharge US research. It consists of 26 national importance challenges and a cast of tech giants—AMD, Nvidia, Microsoft, and others—plus scrappy innovators betting on fresh ideas. Instead of relying only on classic CPUs and GPUs, newcomers like NextSilicon and other dataflow chip designers are reinventing how computers think, so tomorrow’s breakthroughs can finally escape today’s bottlenecks.

Source: Jon Peddie Research

Genesis is a DOE national mission to accelerate science through artificial intelligence and build the world’s most powerful scientific platform to accelerate discovery science, strengthen national security, and drive energy innovation.

The DOE has unveiled 26 science and technology challenges, billed as “of national importance,” to advance the Genesis Mission and speed innovation and discovery through AI.

The DOE says the Genesis Mission will develop an integrated platform that connects the world’s best supercomputers, experimental facilities, AI systems, and unique datasets across every major scientific domain to double the productivity and impact of American research and innovation within a decade.

The list of companies signed up to help reads like the who’s who of tech: AMD, Amazon, Google, IBM, OpenAI, Microsoft, Nvidia, etc.

But these big problems won’t easily yield their secrets to conventional processors or AI models, or they already would have, and there’d be no need for the project—so something new is needed to crack the code, the secrets, and get the genie.

Interestingly, it might be that the little guys have the secret sauce to really tackle the big problems. And what is that secret sauce? Dataflow, emerging around 1974–1975. Jack Dennis and his team at MIT are credited with pioneering this field as a radical alternative to the traditional von Neumann control-flow architecture.

An older start-up (founded in 2017), Israeli-based NextSilicon, developed a dataflow processor aimed at the AI training and inference market. They targeted DOE, and the tech got them in the door and some early trial contracts. Then Sandia National Laboratories launched a new supercomputer, Spectra, that uses Maverick-2 accelerators developed by NextSilicon. NextSilicon is also collaborating with partners such as Dell Technologies and Penguin Solutions to facilitate early-adopter programs.

They didn’t invent it, and they aren’t the only ones employing dataflow.

Company	CIM	Neuromorphic	RISC-V	AI Accelerator
Esperanto.AI			Yes	Yes
Flex Logix				Yes
GrAI Matter Labs		Yes		Yes
Graphcore				Yes
Groq				Yes
Untether AI	Yes			Yes
Applied Brain Research		Yes		Yes
Axelera AI	Yes			Yes
Cambricon				Yes
Cerebras				Yes
D-Matrix	Yes			Yes
EdgeCortix				Yes
Hailo Technologies Ltd				Yes
Kinara (pre-NXP)				Yes
MemryX	Yes	Yes		Yes
Morphing Machines				Yes
Mythic	Yes	Yes		Yes
NextSilicon				Yes
Quadric.io				Yes
SambaNova Systems				Yes
SiFive			Yes	Yes
Tenstorrent				Yes

Table 1. AI processors based on dataflow. (Source: Jon Peddie Research)

Included in the list is Groq, whose technology is not part of Nvidia, another clear demonstration of the path forward and Nvidia’s intention to be part of it. Kinara is another dataflow acquisition (by NXP), and Intel is in advanced negotiations to acquire dataflow start-up SambaNova Systems for approximately $1.6 billion, while Esperanto’s dataflow IP has been acquired by Ainekko (sometimes referred to as Nekko.ai). That leaves AMD to either craft its own or buy someone with dataflow.

So, in addition to the DOE’s use of AI models to tease out the secrets of science, they will also have to employ the newest AI processor types to get the best and fastest results. That’s not to say that classic machines will not also be employed; they just won’t be the star of the show. And don’t be surprised to find those crazy quantum machines in the mix.

We think the cloud training and inference AI processor market was worth over $46 billion in Q4’25 and is growing.

Epilogue

What’s the difference between the classic grand challenges and the Genesis 26 challenges?

Classic “grand challenges” are broad, aspirational agendas for science and technology, while the Genesis 26 are a single, tightly scoped, AI-centric challenge list owned by DOE. Those challenges are defined as ambitious but achievable goals that mobilize diverse researchers and sectors to tackle major national or global problems (health, climate, space, etc.). Typically high-level and open-ended (e.g., eradicate a disease, land humans on the moon, sequence the human genome), they are not tied to one agency, technology, or a fixed 26‑item menu.

A 26‑item, DOE-authored list specifically for the Genesis Mission, the Genesis 26 challenges are focused on using AI to accelerate work in energy, discovery science, and national security. Each challenge is written as a concrete problem with defined AI approaches, justification, and expected impact (for example, accelerating fusion licensing, scaling biotechnologies, and modernizing grid planning).

Classic grand challenges span many domains and often multiple agencies; Genesis 26 are confined to DOE’s mission space and AI‑enabled use cases. Classic grand challenges are thematic “North Stars,” whereas Genesis 26 are operational, implementation-ready targets with specific AI and infrastructure hooks.

About the Author

Dr. Jon Peddie is a recognized pioneer in the graphics industry, president of Jon Peddie Research, and named one of the world’s most influential analysts. Dr. Peddie is an ACM Distinguished Speaker and is an IEEE Distinguished Visitor and named an IEEE Computer Society Distinguished Contributor and Charter member. He lectures at numerous conferences and universities on topics about graphics technology and the emerging trends in digital media technology. Contact him at [email protected].