The global economic downturn, coupled with a shrinking investment landscape and a shrinking market, is expected to bring a full-blown crisis to AI chip startups, which had once been on the rise like rising stars, in 2023. Reports suggest that at least two promising AI chip startups facing this challenge could be sold.

According to Omdia's Top AI Hardware Startups Market Radar, more than 100 venture capital (VC) firms have invested more than $6 billion (approximately 7.8 trillion won) in the top 25 artificial intelligence (AI) chip startups since 2018.

Compared to 2021, when the COVID-19 pandemic was at its peak, the current funding environment is a complete turnaround. Omdia forecasts that financing will become more difficult due to the shift from a global chip shortage to an oversupply, a shift in monetary policy, and an economic downturn starting in 2022.

Experts predict that the management pressure surrounding AI chip startups will intensify.

“The most well-funded AI chip startups are under pressure to deliver software that rivals the market leader, Nvidia,” said Alexander Harrowell, senior analyst for Advanced Computing at Omdia. “This is the biggest challenge in bringing new AI chip technologies to market.”

Omdia predicts that at least two major startups will be acquired through trade sales this year, either by hyperscale cloud providers or major chipmakers.

“The most likely scenario is probably a sale to a major supplier,” said Alexander Harrowell, senior analyst at CNBC. “Apple has $23 billion in cash on its balance sheet, Amazon has $35 billion, and Intel, Nvidia, and AMD have around $10 billion.” He noted that these hyperscalers are very keen to adopt custom AI silicon and can afford to maintain the technology.

The analysis also found that half of the $6 billion in VC funding during this period went to a single technology: large-die and CGRA accelerators designed to load entire AI models onto a chip. However, given the continued growth of AI models, there are also questions about this approach.

"In 2018 and 2019, the idea of bringing entire models to on-chip memory was a good fit because it offered very low latency and addressed the input/output challenges of large AI models," explained Alexander Harrowell, a senior analyst at Intel. However, as these models have continued to evolve since then, scalability has become a critical issue. "More structured and internally complex models mean AI processors need to offer more general-purpose programmability," he said. "Thus, the future of AI processors may lie in a different direction."