Amazon released a microchip for so-called machine learning on Wednesday, entering a market where both Intel and Nvidia Corpare hope to increase their revenue in the next few years.
Amazon is one of the biggest buyers of Intel and Nvidia chips, and its semiconductors support Amazon’s thriving cloud computing division Amazon Web Services. But Amazon has begun to design its own chips.
Amazon’s so-called “Inferentia” chip, announced on Wednesday, will help researchers call reasoning, the process of using artificial intelligence algorithms and putting them into use, such as scanning incoming audio and converting it into text-based requests.
Amazon chips will not pose a direct threat to Intel and Nvidia’s business because it will not sell these chips. Amazon will begin selling services to cloud customers running on its chips next year. If Amazon relies on its own chips, it may deprive Nvidia and Intel’s major customers.
Intel’s processors currently dominate the machine learning reasoning market, and by 2021, Morningstar analysts believe its value will reach $11.8 billion (about RB.25 billion). In September, Nvidia launched its own reasoning chip to compete with Intel.
In addition to machine learning chips, Amazon also announced a cloud computing processor chip called Graviton on Monday. The chip is powered by Arm Holdings, a company controlled by SoftBank Group Corp. Arm-based chips currently power mobile phones, but many companies are working to make them suitable for data centers. Using the Arm chip in the data center may be Intel’s main challenge in this market.
When designing its own chips, Amazon is not just a cloud computing provider. In 2016, the letter-owned Google Cloud Computing Unit introduced an artificial intelligence chip designed to receive chips from Nvidia.
Custom chip design and production costs are high, and analysts point out that this investment drives research and capital expenditures for large technology companies.
Google Cloud executives said that customer demand for Google’s custom chip TPU has been strong. However, the cost of using the chip is high and requires software customization.
According to Google’s website, Google Cloud can charge $8 per hour for TPU chips and $2.48 per hour for accessing Nvidia chips.