NVIDIA Invest Baseten With $150 Million to Boost AI Inference Infrastructure
The Wall Street Journal (WSJ) reported that NVIDIA had invested in Baseten, a startup which was developing an AI-inference optimisation platform in the U.S., amounting to 150 million USD (or about 220.7 billion Korean won). This figure is half of the amount that Baseten collected in this round, which was $300 million. As this investment was closed, the corporate value of Baseten hit the $5 billion mark (equivalent to around 7.4 trillion Korean won) and doubled. According to WSJ, NVIDIA is busy pouring funds into technologies that enable AI inference to be more efficient and stated that the AI industry doesn’t focus on training large-scale models but on inference.
Assuming that the stage of developing a smart AI brain is called AI training, the next stage is inference, which is the use of the brain that is already created successfully. The trick is to consume less power and pay less in terms of usage fees by achieving results at a faster rate using limited resources. As large-scale model training (which demands astronomical start-up costs) dominated the competition of big tech companies, several start-ups are currently defining the market in the inference space. NVIDIA, the leader in the AI semiconductor market in terms of training, is preemptively investing in gaining a presence in the inference market, which is projected to have a place to occupy in the future.
NVIDIA also invested $20 billion (or about 29 trillion Korean won) in Groq in December of the previous year, a firm that is creating Language Processing Unit (LPUs) to provide faster inference in lieu of Graphics Processing Unit (GPUs). Although formally the two companies entered into an exclusive use of the technology agreement that was technically termed as a technology agreement, analysts opine that this is actually an acquisition of Groq by NVIDIA. NVIDIA is developing into a full-fledged provider of AI systems, integrating GPUs, inference-specific chips, and inference-efficiency software by acquiring other non-GPU products, such as the software strengths of Groq and Baseten.
It is true that AI inference startups have experienced huge valuation growth over the recent past. Cerebras Systems, a company specialising in cloud services based on large language model (LLM) inference, also declared a 10-billion-dollar computing alliance with OpenAI last year and is now negotiating a 1-billion-dollar funding round with a valuation of $22 billion. Fireworks AI, a company that offers AI inference infrastructure to developers, raised $250 million in October last year, valuing it at 4 billion.
