Weebit Nano takes aim at costly AI challenges of more speed and less power

• AI’s power hunger puts memory in the spotlight
• Weebit’s ReRAM shows promising AI efficiency gains
• Weebit customers target 2027 production as commercialisation advances

Artificial intelligence has developed a rather expensive habit.

The smarter it gets, the more electricity it wants. And apparently, feeding it billions of dollars’ worth of chips isn’t quite enough anymore.

Now, one of the world’s biggest memory chipmakers, South Korea’s SK hynix, reckons the industry’s next big challenge isn’t simply making computers faster.

It’s stopping them from guzzling so much power.

Speaking at a technology conference in Seoul this week, SK hynix HBM executive Choi Jun-yong warned that AI’s bottleneck was shifting from computing performance to memory, and increasingly, electricity.

The company is developing next-generation memory technology that reduces the energy wasted moving information around inside AI systems.

It’s a pretty significant shift in thinking.

For years, the semiconductor industry has been obsessed with squeezing more computing muscle into increasingly powerful chips.

But all that extra horsepower isn’t much use if the electricity bill becomes astronomical.

And for investors, it raises an interesting question.

What if the next big AI winners aren’t necessarily making the biggest processors, but helping them work smarter?

AI has a memory problem

Here’s something that doesn’t get nearly enough attention in the AI investment frenzy.

A surprising amount of energy isn’t spent doing calculations. It’s spent shuffling information between memory and processors.

Think of a chef who has to walk across the kitchen every time they need an ingredient. Even the fastest chef would spend half the evening wandering around.

Traditional computing works somewhat similarly, constantly moving data back and forth before processing it.

McKinsey recently estimated that by 2030, around 60% of AI computing demand could come from inference, rather than training.

Inference is when AI actually gets to work, answering questions, recognising images or making decisions.

And because inference happens repeatedly, potentially billions of times, even modest efficiency improvements could translate into substantial savings.

That’s why semiconductor companies are exploring ways to perform calculations directly inside memory.

Which brings us to an interesting ASX-listed contender.

Weebit tackling the problem inside the chip

Weebit Nano (ASX:WBT) has spent years developing a type of semiconductor memory called resistive random-access memory, or ReRAM.

Unlike conventional flash memory, ReRAM stores information by changing the electrical resistance of a material.

It retains information when power is switched off, while offering advantages in speed, energy consumption and manufacturing efficiency.

And there’s a potentially bigger opportunity emerging.

Weebit is exploring “in-memory computing”, where certain calculations happen directly within memory rather than requiring information to travel backwards and forwards to a processor.

It’s an approach that could become increasingly valuable as AI spreads into cars, industrial equipment, robots and battery-powered devices.

In October, Weebit reported results from a US customer’s study of its ReRAM technology for an AI accelerator designed for harsh environments.

For the particular AI task assessed, projections indicated more than 80 times lower power consumption than a GPU-based implementation – approximately 280 times higher compute efficiency than an FPGA-based accelerator, and inference latency falling from tens of milliseconds to less than one millisecond.

Those are research projections, not commercial product results, but they illustrate the potential.

Weebit has also signed agreements with two new AI customers to explore in-memory computing applications.

CEO Coby Hanoch acknowledged the work still ahead, saying:

“While significant R&D work remains before we have a fully commercial IMC solution, these results are an encouraging indication of ReRAM’s potential in next-generation AI computing.”

And while that AI opportunity is still being developed, Weebit’s core embedded memory technology is already much further along commercially.

Weebit closes in on commercial production

According to industry forecasts cited in Weebit’s October investor presentation, the global semiconductor market is on track to reach US$1 trillion four years earlier than previously expected.

That’s an enormous amount of silicon heading into everything from AI servers to cars and industrial equipment.

For investors, this is where things get interesting.

It’s one thing to develop impressive technology in a laboratory. Getting semiconductor manufacturers to put it inside actual chips is an entirely different challenge.

Weebit’s ReRAM has completed qualification at South Korean foundry DB HiTek, while US semiconductor giant onsemi has begun qualification work. Its collaboration with Texas Instruments is also progressing.

And the commercial progress is becoming real, with two product customers running software on working silicon samples featuring Weebit’s ReRAM.

Both are targeting production in 2027, subject to completing validation.

“Having two customers successfully testing working samples is an important step towards mass production of our technology,” said Hanoch.

One of those customers, Overlord Labs, is incorporating Weebit’s memory into a battery intelligence chip for edge-AI devices, with production targeted for 2027.

And Weebit doesn’t need to build expensive semiconductor factories itself.

Its business model revolves around licensing intellectual property to chipmakers, with potential royalties as customers move into production.

The company recorded $15.3 million in FY26 revenue and received $6.1m from customers in the September quarter.

It also finished September with $164.2m in cash, giving it financial flexibility as it works towards commercial production.

Its inclusion in the S&P/ASX 200 in September was another milestone.

Long Shortz with Weebit Nano: Moving to mass production

The next AI trade could be efficiency

Weebit cited Yole Group research forecasting the emerging embedded non-volatile memory market could grow from around US$300m in 2025 to US$4.1bn by 2031, with ReRAM accounting for half.

But Weebit isn’t the only ASX semiconductor player trying to make AI smarter without sending electricity bills through the roof.

BrainChip Holdings (ASX:BRN) is developing neuromorphic processors that mimic aspects of how the brain processes information, allowing AI to operate on devices while consuming less power.

Nanoveu (ASX:NVU) , through its EMASS subsidiary, is developing ultra-low-power chips for edge AI applications, including drones and other battery-powered devices.

Meanwhile, Adisyn (ASX:AI1) is tackling another headache inside semiconductor chips.

It’s developing graphene-based interconnect technology to potentially replace conventional copper connections, which become less efficient and generate more heat as chips shrink.

For investors looking beyond the usual AI heavyweights, that makes these companies worth watching.

Stockhead has not provided, endorsed or otherwise assumed responsibility for any financial product advice contained in this article.

At Stockhead we tell it like it is. While Weebit Nano is a Stockhead advertiser, it did not sponsor this article.

The post Weebit Nano takes aim at costly AI challenges of more speed and less power appeared first on Stockhead.