Investing.com -- The U.S. decision to tighten export controls on Nvidia’s H20 chips has introduced fresh complications for China’s AI sector, but the near-term disruption appears limited as domestic firms pivot to alternatives and deepen their investment in local ecosystems.

Analysts at Bernstein say the ban may even hasten China’s shift away from reliance on American semiconductors, a transition already underway across both hardware and software fronts.

The Nvidia (NASDAQ: NVDA ) H20 chip, considered one of the lower-tier accelerators tailored for the Chinese market, offered significantly less computational power compared to other high-end Nvidia offerings.

Analysts argue that its ban is largely symbolic, as performance comparisons showed it was already lagging behind Chinese alternatives like Huawei’s Ascend 910C.

Huawei’s recently unveiled AI CloudMatrix 384 platform, for instance, delivers 1.7 times the computing power of Nvidia’s NVL72 node, albeit at nearly four times the power consumption.

Bernstein analysts believe the greater impact lies in the signal the ban sends and the supply dynamics it intensifies.

Companies with foundational AI models at their core, especially those in the media-focused tech sector, will feel the constraints most acutely.

However, a large portion of the Chinese AI industry has proactively diversified its compute strategies to reduce reliance on Nvidia, employing a combination of substitutes for inferencing, smaller edge-trained models, and hybrid platforms that interlink domestic chips with Nvidia’s hardware.

Short-term operations are expected to remain largely unaffected. Most AI firms Bernstein surveyed reported they could continue using previous-generation chips like the Nvidia 3090 or V30, or switch to Huawei and other local alternatives.

The cloud sector, however, may face rising pressures. As compute power becomes scarcer and demand persists, pricing for cloud-based GPU acceleration is climbing.

Bernstein notes that GPU rental prices across various providers have risen steadily since early 2023.

In terms of longer-term outlook, the H20 ban is likely to catalyze a deeper domestic push toward AI independence.

Huawei is expected to close the performance gap further, helped by innovations in software that improve cluster bandwidth and model distillation techniques.

A domestic open-source initiative, Deepseek, has demonstrated promising results in developing efficient small-scale models optimized for performance without requiring massive computing power.

Bernstein flags three main technical approaches China is leveraging to reduce dependence on Nvidia.

First is the manual re-engineering of CUDA-trained models to run on Huawei chips, an effort that, while resource-intensive, has yielded up to 90% performance parity in some cases like iFlytek’s adaptation of its iSpark model.

Second, compiler-based porting of pre-trained models offers an automated but still-imperfect path.

Third, a middle-layer compute harmonization approach is being pursued to allow training and inference across heterogeneous chip environments.

TMT and software companies are also expected to shift from private to public cloud infrastructure, according to the brokerage.

State-owned enterprises are favored in the cloud sector due to their close ties to domestic chip suppliers like Huawei.