[News] U.S. Three-Tier AI Export Rules May Hit Israel’s AI Development

In its final days, the Biden administration has introduced a three-tiered system for AI chip export controls, a move that could significantly impact global AI development. According to a report from MoneyDJ, citing Tom’s Hardware, Israel—a key U.S. ally—has been placed in the second tier of these restrictions. Consequently, Israel may now need to obtain U.S. licenses to purchase AI chips developed by American companies.

Impact on Israel’s AI Industry

The report from Tom’s Hardware, referencing Israeli media outlet CalcalisTech, highlights serious concerns raised by Israeli tech companies. They warn that the new U.S. export controls could severely disrupt Israel’s AI sector by delaying the construction of data centers and hindering progress in cloud services within the country.

As noted by Tom’s Hardware, if the new U.S. AI chip export controls are fully implemented, Israeli entities may need to apply for U.S. licenses even to purchase Intel’s Gaudi series AI processors, which are locally developed in Israel.

The Three-Tiered AI Export System

The Biden administration’s AI chip export regulations divide countries into three tiers. According to Tom’s Hardware, Tier 1 includes 18 of the U.S.’ most trusted allies, such as Taiwan, Australia, Canada, Japan, and most Western European nations. These countries can generally access U.S. AI technology without restrictions.

Tier 3, on the other hand, encompasses adversarial nations, including Russia, China, and other countries under U.S. sanctions, which are entirely barred from obtaining advanced AI processors.

Tier 2 sits in the middle, encompassing allies like Israel, the Baltic states, Poland, Saudi Arabia, and the United Arab Emirates. These countries face restrictions on accessing U.S. AI technology, as detailed in the report.

Specific Restrictions on Tier 2 Countries

As highlighted in the report from Tom’s Hardware, entities in Tier 2 countries can procure up to 1,700 NVIDIA H100 GPUs (or equivalent compute power) for non-sensitive purposes without needing export licenses. These orders, particularly those from universities, healthcare institutions, and research organizations, are exempt from counting toward a country’s overall AI chip limits.

However, businesses requiring significant AI processing power must secure National Validated End User (NVEU) status to purchase larger quantities by complying with high-security standards. With this designation, entities can procure up to 320,000 NVIDIA H100 GPUs (or equivalent compute power) over a two-year period, as stated by Council on Foreign Relations.

For entities without NVEU status, purchases are capped at 50,000 GPUs. Some Tier 2 countries may negotiate special agreements with the U.S. government to double this cap to 100,000 Nvidia’s H100 GPUs, according to the report.

Challenges for Large AI Developers

Despite these provisions, the report notes that large enterprises working on AI models with hundreds of billions of parameters may find that, even with the ability to procure up to 320,000 GPUs over two years, these quantities may still fall short of their needs.

Furthermore, the process of applying for export licenses for every GPU batch could lead to significant delays, as emphasized in the report. Given the rapid pace of AI development, such delays could pose substantial challenges to the industry.

The report also mentions that the proposed export controls include a 120-day public comment period. The final decision on these regulations will likely be made by the incoming Trump administration.

Read more

• [News] Why the U.S. Bars India from Unrestricted Access to AI Chips
• [News] Biden’s AI Chip Export Curbs in the Works: Key Points of Three-Tier System at a Glance

(Photo credit: NVIDIA)