Qualcomm has been working for several years on squeezing AI (artificial intelligence) capabilities into small devices. Now it has established the $100m Qualcomm AI Ventures Fund to support AI start-ups and encourage applications that will drive device-based usage (and uptake of Qualcomm chips) rather than being cloud-based.
The chip designer said the fund would complement and build on its own research into low power processing and connectivity, which it says are essential to AI in small footprint devices which may not always be linked to the cloud.
Particular areas of interest for the fund are AI for autonomous cars and robotics, plus specialized machine learning platforms.
The first company to receive financing from the new initiative is AnyVision from Israel, which is developing AI systems that can identify people and objects, but can allay privacy concerns about centrally stored personal data, by keeping everything on the device.
Qualcomm – a powerhouse in device chips, but with its data center processor development on the back burner – naturally argues that central, cloud-based AI applications are too computationally intensive to be cost-effective.
“As a pioneer of on-device AI, we strongly believe intelligence is moving from the cloud to the edge,” said CEO Steve Mollenkopf. “Qualcomm’s AI strategy couples leading 5G connectivity with our R&D, fuelling AI to transform industries, business models and experiences.”
Other companies are on the same track. Huawei launched its own AI chips last month, under its Ascend brand – with one family member, the 910, targeted at the data center, but another, the 310, designed for smartphones.
Ken Hu, Huawei’s rotating chairman, at the recent Mobile Broadband Forum event, echoed Qualcomm’s concerns about the cost of AI. “We are trying to make computing power more affordable,” he said. “Nowadays it is still very expensive and that is a big bottleneck for AI.”
In May, Qualcomm announced its AI Research organization, which places the company’s AI projects under one roof and enabling cross-function cooperation in this area.