AI chips manage to process data in certain areas faster and more efficiently than the processors from Intel or AMD known from PCs and notebooks.
(Photo: Bloomberg)
Munich, Tokyo Chips are more popular than ever. One category is particularly in demand: semiconductors for artificial intelligence (AI). According to a forecast by the market research institute Gartner, sales will increase by more than half to almost 36 billion dollars in 2021. For the coming year, the market researchers expect an increase of a good 20 percent to around 43 billion dollars. Business would have more than tripled in just four years.
No wonder there has been a run for this new type of chip. According to Gartner expert Alan Priestley, around 80 start-ups worldwide are hoping for a breakthrough. Most of them will not survive on their own two feet and merge into corporations, the expert predicts. They lack market access.
At the same time, tech companies are rushing to the AI chips. But not all of them have the patience to build up the know-how themselves. For example, world market leader Intel took over the Israeli start-up Habana two years ago for two billion dollars.
AI chips are a special form of application-specific integrated circuits, so-called Asics. They manage to process data in certain areas faster and more efficiently than the processors from Intel or AMD known from PCs and notebooks.
Top jobs of the day
Find the best jobs now and
be notified by email.
Venture capitalists are speculating on more deals. The most important European start-up in this field is Graphcore. The British firm has raised more than $ 700 million from investors and was valued at just under $ 2.8 billion in the last financing round.
The US competitor Sambanova collected 676 million dollars this spring and can therefore fall back on a total of one billion dollars. Californian competitor Groq raised $ 300 million in April.
Sony brings artificial intelligence to end devices
More and more established corporations are also developing AI chips. Take Sony, for example: Last year, the Japanese jumped on the trend of using AI directly in devices. Experts speak of the “edge”. The company presented its smart IMX500 image sensor, which is the first in the world to be able to evaluate the photo pixels directly on an integrated AI chip.
Sony is promoting this image analysis on the semiconductor primarily as an AI application that protects privacy. If the customer does not want it, no image data would have to leave the chip in order to be evaluated in a data center as before.
For example, Sony suggests so-called “heatmaps” in retail, where the supermarkets only receive metadata, in this case anonymous movement information from consumers, and no video images of the customers.
The traffic planning system was tested in Rome. Cameras could detect free parking spaces or unoccupied seats in the bus. The information can be used by motorists or transport companies.
But it’s not just about privacy: Since the chip only transmits meta data and no video data, the data stream is reduced to up to a ten-thousandth. This reduces power consumption, the time delay decreases, while cybersecurity is supposed to increase.
For Sony, the chip was the first step towards becoming a platform provider for AI applications. Eita Yanagisawa, Managing Director of Sony’s System Solutions division, recently presented the Aitrios platform. It takes the image sensor as the basis. “It’s a one-stop platform that simplifies the introduction of AI at the edge,” promises the manager.
The service allows customers to develop their own applications. Sony also offers an app marketplace and a cloud service. If possible, Sony wants to expand the range to other sensors that measure movement or humidity, for example, says Yanagisawa.
While Sony focuses on the end device, others focus on AI in data centers. This is also the case with the cloud provider Amazon Web Services (AWS). With the in-house developed AI chip series “Inferentia”, AWS promises more efficient data analysis in machine learning. “Trainium” is intended to significantly improve the training of neural networks – this is associated with the processing of large amounts of data. So far, Nvidia has led the way with its graphics processors.
Most AI chips are in smartphones
The companies currently generate around two thirds of sales with AI chips with smartphone manufacturers. You use the components, for example, for voice and face recognition or in the camera. The largest providers include Qualcomm and Mediatek.
In the future, AI chips should be found everywhere, says Gartner expert Priestley: from traffic lights to surveillance cameras to washing machines. By the middle of the decade, sales will rise to just under $ 73 billion, doubling compared to 2021.
More: Expensive catch-up: Intel loses market share to the competition – share price collapses