Hot French startup ZML releases free product to speed inference across lots of AI chips

Nvidia’s Dominance Faces Challenges from European AI Startups

The days of Nvidia’s unparalleled market dominance in the AI chip space are being disrupted by a new wave of European startups, including ZML, a hot French AI startup endorsed by Turing Award winner Yann LeCun. ZML has released inference-performance software that allows a variety of open-source large language models to run on a variety of chips, including Nvidia’s, AMD’s, Google’s TPU, Apple Metal, and Intel Arc. This development mirrors what happened to Intel in the 2000s when ARM-based chips started to gain traction, threatening Intel’s dominance in the mobile space.

With ZML/LLMD, the newly launched LLM inference server, the company’s ambition is to break existing silos and make different chips available for AI use cases at their maximum available speed. This could be a market disruptor, amid mounting fears over AI-related costs. ZML hopes to provide enterprises and clouds with the option to use a mix of chips, some of which might be less costly or consume less energy.

The promise of achieving peak performance across a variety of chips is a technological feat, but it also raises questions about the business model of AI chipmakers. As AI becomes integrated into our work and everyday lives, optimizing inference has become a key area of investment, with Nvidia gearing up for the rise of inference. However, the trend has also been hailed the “inference gold rush,” with multiple startups competing in the space.

ZML’s Decision Logic and Mechanics

What ZML is not saying publicly is that its software assist could potentially disrupt the business model of AI chipmakers like Nvidia. By providing a software solution that allows different chips to run at maximum speed, ZML is effectively reducing the need for customers to purchase specific chips for specific use cases. This could lead to a decrease in revenue for chipmakers, who have traditionally relied on the sale of specialized chips for AI applications.

From a technical perspective, ZML’s software solution is designed to optimize inference performance across a variety of chips. This requires a deep understanding of the underlying architecture of each chip, as well as the ability to optimize the software for each specific use case. ZML’s team of 20 people has been working closely with chipmakers to achieve this goal, with more releases planned in the coming months.

The operational mechanics of ZML’s software solution are complex, involving the use of open-source large language models and custom-built software to optimize inference performance. This requires significant expertise in AI and chip architecture, as well as a deep understanding of the business model of AI chipmakers. ZML’s founder, Steeve Morin, has a track record of success in the AI space, having previously served as VP of engineering at Zenly, which was acquired by Snapchat for nine figures in 2017.

Winners, Losers, and Disrupted Parties

The winners in this scenario are likely to be enterprises and clouds that can take advantage of ZML’s software solution to reduce their AI-related costs. By providing the option to use a mix of chips, some of which might be less costly or consume less energy, ZML is effectively reducing the barriers to entry for AI adoption. This could lead to an increase in demand for AI solutions, particularly in industries where cost is a significant factor.

The losers in this scenario are likely to be AI chipmakers like Nvidia, who may see a decrease in revenue as a result of ZML’s software solution. However, it’s worth noting that Nvidia has been gearing up for the rise of inference, and may be able to adapt to this new reality. Other chipmakers, such as AMD and Intel, may also be affected by ZML’s software solution.

The disrupted parties in this scenario are likely to be the traditional business models of AI chipmakers. By providing a software solution that allows different chips to run at maximum speed, ZML is effectively disrupting the traditional business model of chipmakers, who have traditionally relied on the sale of specialized chips for AI applications. This could lead to a significant shift in the way that chipmakers operate, with a greater focus on software solutions and less on hardware sales.

The Skeptical Case

One potential skeptical view of ZML’s software solution is that it may not be able to achieve the same level of performance as specialized chips. While ZML’s software solution may be able to optimize inference performance across a variety of chips, it may not be able to match the performance of specialized chips that are designed specifically for AI applications.

Another potential skeptical view is that ZML’s business model may not be sustainable in the long term. By providing a free software solution, ZML is effectively giving away its intellectual property, which may make it difficult to generate revenue in the future. This is a concern that has been raised by other startups in the AI space, who have struggled to generate revenue despite having a strong product.

The Signal to Watch Next

The next signal to watch is the adoption rate of ZML’s software solution. As more enterprises and clouds begin to use ZML’s software, we will get a better sense of whether it is able to achieve the same level of performance as specialized chips. We will also get a better sense of whether ZML’s business model is sustainable in the long term.

Another signal to watch is the response of AI chipmakers to ZML’s software solution. As chipmakers begin to adapt to this new reality, we will get a better sense of whether they are able to innovate and stay ahead of the competition. This will be an important indicator of the health of the AI chip market, and will have significant implications for the future of AI adoption.

Bookmark this one — it will matter to your business decisions this week.

By Priya Nair, AI & Startup Reporter at TrendFlashy