- Vere Computing CEO says reversible computing is inevitable
- Reversible computing chip generates virtually no heat
- Of course, chips that don’t generate heat would deal a blow to the liquid cooling industry’s growth plans.
Longtime industry veterans know that critics have been warning for years that chipmakers would hit the limits of Moore’s Law, but while there’s no clear consensus on when that will happen, that day is approaching — and one startup, Vaire Computing, has its reversible computing technology ready and waiting.
In fact, the CEO of Vaire Computing is so convinced that reversible computing will be the driving force behind artificial intelligence (AI) in the future that he has actually hired almost an entire team of experts in the field. He told Fierce Network that the company aims to become the next Nvidia by gaining early exclusivity in the field.
“It may sound crazy, but the way we see it, in 15 years all computer chips will be reversible. It’s inevitable, there’s no escaping it,” Vale CEO Rodolfo Rosini said. “We figured it out, but I think everyone else, including Nvidia and Intel, got it wrong.”
On earth?
Let’s dig a little deeper to explain the logic behind Rosini’s statement.
Moore’s Law is an observation rather than a hard and fast law of physics that basically states that the number of transistors on an integrated circuit will double every two years without a dramatic rise in cost. Transistor count is important for chip companies like Nvidia as they seek to push the boundaries of computing for use cases like AI.
Contrary to what the headlines say, performance improvements are actually slowing (something I heard from an executive at Infinera, a Nokia acquisition, a while back) and the focus is starting to shift to energy efficiency.
This is where reversible computing comes into play.
In traditional chips, energy is used to change the state of gates on the silicon, and the wasted by-product is heat. But in reversible computing, the wasted energy from each gate change is recycled to power the next gate change. In practice, this means that reversible computing chips are slower than traditional chips, but they are very energy efficient and generate virtually no heat.
Reversible computing is not actually a new technology – its history dates back to the 1970s, when the first universal reversible gate was built in 1982 – but Vaire is optimizing the technology in multicore designs to make it viable in an AI-driven world.
If you’ve been paying attention, you know that heat and power are two major concerns for the computing industry, given increasing chip density and power grid constraints in the U.S. and abroad. Of course, chips that don’t generate heat would be a blow to the growth plans of the liquid cooling industry.
“In four years, we’re going to reset to a point where nobody’s going to get any performance unless they change their architecture,” Rosini said. “At that point, we’re going to have an opportunity to build a new trillion-dollar company over the course of a decade or so by taking full control of this technology.”
“So the bet is not to be acquired by Nvidia, There are “The next Nvidia,” he continued.
Vaire has already been in stealth mode for three years in research and development, and just raised $4 million in seed funding, bringing its total funding raised to date to $4.5 million, according to Rosini.
Vere is preparing to release an initial version of its silicon in early 2025. The CEO said the company will initially design the chips for edge deployments, but expects to have chips robust enough for use in data centers within four to five years.
Outside Perspective
So what does the analyst community make of Vaire and its claims?
Gaurav Gupta, vice president analyst at Gartner, told Fierce that he wasn’t familiar with Vaire, but that Rosini was right that Moore’s Law is unravelling. The doubling of transistors has slowed, and costs are no longer stable. In fact, Gupta said, the cost of leading-edge wafer chips is actually exploding.
Still, Gupta noted that chipmakers are making improvements in other areas, such as packaging techniques.
Gupta wasn’t able to speak to Vaire specifically, but he noted that there’s a big difference between the kind of research and development work Vaire has been doing for several years and commercializing a product at scale, which of course is much harder.
Fierce also quizzed AvidThink founder Roy Chua, one of the few analysts familiar with both the company and its technology, on whether Vaire’s plans are feasible. The verdict? He seems to think Vaire has big things going for it.
“Veir’s initial funding to achieve a relatively small but demonstrable milestone for a silicon business, especially one with such a potentially revolutionary potential, signals renewed interest in the space,” Chua said, adding that Veir’s target for early 2025 for its first silicon business is “bold but exciting if it happens.”
He added that replacing today’s computing architecture would take “a lot of effort and time,” as it is complex and would require careful rethinking of how logic gates are built, the manufacturing process, the EDA process for large-scale circuit design, system architecture, and more. And it’s not entirely clear that Vaire’s solution can be integrated directly into existing workflows, Chua added.
But if the veil teeth Can the company demonstrate the feasibility and commercial viability of its solution? If so, “we will see a lot more funding pumped into this space, given the need for more computing resources to run any kind of workload, including AI/ML. If successful, we expect the company’s next fundraise to be significantly larger to facilitate scale-up,” Chua concluded.
For the record, Fierce reached out to Nvidia for comment, but the company declined to comment on whether it has researched reversible computing, whether it is concerned about the limits of Moore’s Law, or whether it thinks anyone can unseat it as the dominant force in AI computing. Instead, a spokesperson directed Fierce to this blog post.
However, it’s worth noting that the company doesn’t just rely on transistors to improve the performance of its chips, but also tweaks the software to improve performance after each generation of chips is released. For more details, check out this article.
