Despite growing concerns about the AI bubble, venture capitalists remain highly enthusiastic about the chip sector. Just on Tuesday this week, three AI chip startups have already raised over $1.1 billion in funding, showing that the capital market still has confidence in emerging forces capable of challenging NVIDIA's dominance.

In this wave of funding, MatX, a company founded by former Google engineers, secured a major $500 million. MatX plans to launch its first chip, MatX One, later this year. Unlike companies like Groq, which focus on inference, MatX claims its chip can handle pre-training, reinforcement learning, and inference tasks. The chip features a unique architecture that combines ultra-fast SRAM with large-capacity HBM memory, aiming to achieve a perfect balance between throughput and processing speed.

Meanwhile, Dutch company Axelera also announced a $250 million funding round. Unlike MatX, which directly competes with NVIDIA, Axelera has a more practical goal, focusing on low-power edge computing applications such as computer vision and robotics. Its latest Europa chip offers computing performance comparable to NVIDIA A100 with only 45 watts of power consumption, achieving an extremely high energy efficiency ratio, just one sixth of A100's power consumption.

Additionally, SambaNova also received $300 million in funding during this round and announced a deep collaboration with Intel, integrating Xeon processors into AI servers. SambaNova also revealed its next-generation accelerator SN50, which is expected to be deployed first in SoftBank's data centers in Japan later this year. AIbase observed that this series of significant funding rounds indicate that customized, high-energy-efficiency AI silicon is becoming the new hot spot for capital investment.

Summary:

  • 💰 Capital Rush: On just Tuesday, $1.1 billion flowed into the AI chip sector, with MatX, SambaNova, and Axelera all receiving substantial support.

  • 🚀 Universal Chip Challenge: MatX, backed by Google, launched a general-purpose accelerator designed to handle both training and inference, pursuing the highest computing power per square millimeter through SRAM architecture.

  • 🔋 Edge Energy Efficiency Breakthrough: Dutch startup Axelera focuses on low-power edge AI, achieving performance equivalent to A100 with less than one sixth of the power consumption.