Nvidia, the U.S.-based semiconductor giant, plans to introduce a new artificial intelligence (AI) chipset for the Chinese market, priced between $6,500 and $8,000, with mass production slated for June 2025, according to sources familiar with the matter.
This move, driven by stringent U.S. export restrictions, aims to maintain Nvidia’s foothold in China’s $50 billion data center market by offering a more affordable alternative to its recently banned H20 chip, which retailed for $10,000-$12,000.
The new chip, which insiders say will be priced between $6,500 and $8,000, is based on Nvidia’s latest Blackwell architecture the same core technology behind its most advanced AI processors. It will likely be dubbed the RTX Pro 6000D or B40, according to Chinese brokerage GF Securities, though an official name hasn’t been confirmed.
But unlike its more powerful predecessor, the H20 , which was effectively banned in China in April, this new version comes with weaker specifications and simplified manufacturing to comply with U.S. export rules. The H20 sold for around $10,000 to $12,000, a steep tag tied to its cutting-edge capabilities.
Nvidia’s new chip won’t use the more advanced high bandwidth memory (HBM) or TSMC’s Chip-on-Wafer-on-Substrate (CoWoS) packaging, both of which are key to supercharging AI performance. Instead, it will rely on GDDR7 memory, a more conventional (but slower) alternative, and skip the expensive advanced packaging tech that enhances data transfer speeds and energy efficiency.
China represents about 13% of Nvidia’s annual revenue, or an estimated $9 billion, according to company filings. For a tech giant riding the AI boom, that’s a market it can’t afford to lose , even if the chips have to be slower and cheaper.
Nvidia CEO Jensen Huang recently revealed that the company’s market share in China plunged from 95% to 50% since U.S. restrictions began taking effect in 2022. Even more devastating: the H20 ban forced Nvidia to write off $5.5 billion in inventory and walk away from $15 billion in potential sales.
Despite the downgrade in hardware, Nvidia retains a powerful advantage: CUDA, its proprietary programming platform. CUDA is the digital foundation most AI developers use to build models and applications for Nvidia chips.
“CUDA is sticky,” said Nicolas Gaudois, head of Asia tech research at UBS. “The ecosystem built around it makes it hard for developers to switch, even if the chips aren’t the best available.”
This means Nvidia may retain influence in China’s AI infrastructure even as its hardware is throttled, a critical point for maintaining long-term relevance.
In addition to the chip set for mass production as early as June, sources say Nvidia is already working on a second Blackwell-based GPU tailored for China. Production for that model could begin as soon as September, although details on specs remain scarce.
The company had initially considered tweaking the older Hopper architecture to comply with restrictions but ultimately abandoned that idea. Huang has since confirmed that Hopper can no longer be modified under current rules , making Blackwell the future, even if that future is being built under duress.
Nvidia’s new chip for China may not be a powerhouse, but it’s a calculated lifeline ,one designed to buy time, maintain relationships, and ride the razor’s edge of regulatory compliance.
With Huawei nipping at its heels and U.S. policy growing ever more aggressive, Nvidia’s future in China may hinge not just on how well it can innovate , but how well it can adapt.
–
–
