Vessl AI raised its hourly rate for Nvidia H100 graphics processing units by 24.7% starting this month, following a string of price increases by U.S. neocloud providers such as CoreWeave and Nebius amid a surge in demand for artificial intelligence computing. Supply has tightened not only for the newest chips but also for older models as demand for high-performance GPUs jumps. Other Korean cloud providers also expect to raise GPU rates depending on supply conditions, raising concerns that higher costs of securing GPUs will weigh on service development at Korean AI companies and on government AI transformation, or AX, projects.
Vessl AI on the 1st raised the hourly rate for H100 chips on its VESSL Cloud service to $2.98 from $2.39, according to AI industry sources on the 21st. The neocloud company counts leading Korean AI firms including Upstage and Motif Technologies among its clients and currently operates several thousand GPUs. It has secured about 7,000 GPUs, most of them leased rather than purchased outright. "It is hard to secure the volume we need for H100s — effectively, the price is whatever the seller names," a Vessl AI official said. "As AI demand grows, the number of customers wanting to use GPUs keeps rising, so a price adjustment was unavoidable." The company plans to expand its GPU fleet to about 10,000 units by the end of the year, but intends to keep adjusting rates flexibly to reflect supply conditions and procurement costs.
Other Korean AI cloud companies, including Elice, have also raised prices recently. In the past, rental rates for older AI GPUs fell when new models arrived, but prices for previous-generation chips are now rising as well. Hourly cloud rental rates for the H100 climbed 21.9% over the past month to $3.28 from $2.69, according to Ornn, an AI compute trading platform. That is about 70% of the price of the higher-end H200. Analysts say demand is spilling over to relatively older GPUs because supply of the newest chips cannot keep pace with surging computing demand.
The trend is affecting Korea's AI industry and research community. Some companies have given up bidding for government AX projects because they could not secure the GPUs they needed. Universities and research institutes, unable to obtain GPUs reliably, are scaling back research or searching for new funding programs. "What determines success or failure in business and research now is not just whether you have good AI technology, but whether you can secure the GPUs you need in time," an AI industry official said. "The ability to secure computing resources is becoming a key element of AI competitiveness."
