The article reports that the united states government has approved exports of nvidia h200 data center grade accelerators to china, reversing an earlier decision that sought to block “full fat” h200 hardware from entering the chinese market. Nvidia engineers had reportedly been developing a heavily cut down “h20” alternative designed to comply with export sanctions, with details of this design emerging earlier in 2025. The policy shift is discussed in the context of rising competition from huawei’s ascend 910c accelerator, which is portrayed as a key rival to these newly approved, previous generation nvidia artificial intelligence processing solutions.
According to information attributed to bloomberg’s unnamed sources, white house officials have been examining the capabilities of the ascend 910c, particularly when deployed at scale in huawei’s cloudmatrix 384 super node system. While the ascend 910c is said to trail the h200 in both computing power and memory bandwidth, the article notes that an array of 384 such accelerators still represents a substantial computational resource. This capability, combined with broader concerns about china’s semiconductor trajectory, appears to have factored into united states deliberations about how strictly to limit high end accelerator exports.
The report further relays claims from industry figures that huawei and its foundry partners are working toward manufacturing 600,000 units of the ascend 910c over the coming year. White house staffers are also said to have considered a scenario in which huawei “would be capable, in 2026, of producing a few million of its ascend 910c accelerators,” underscoring fears of rapid domestic scaling. In parallel, a related reuters article cited by the piece states that chinese regulators are weighing measures to restrict local industry access to nvidia h200 systems, suggesting that export approvals may collide with china’s own efforts to manage dependence on foreign accelerators.
