Chipping away

This week marked a significant shift in the Biden administration’s approach to tightening China’s ability to procure advanced AI chips.

In a notification letter to Nvidia and AMD — that likely reached other chip companies too — the U.S. said it will now require them to obtain a license before shipping chips that are more advanced than Nvidia’s A100 GPUs and AMD’s MI200 series. The new requirements could cost Nvidia as much as $400 million in quarterly sales, though they will have a negligible impact on AMD.

• Prior U.S. policy had focused on preventing China from gaining access to the tools necessary to produce the most advanced tech, chips or otherwise.
• But China’s civil-military-fusion doctrine made clear that all tech in China, whether owned by the military or civilian entities, is fair game for the government to use, said William Reinsch, senior adviser at Kelley Drye & Warren and Scholl Chair in International Business at the Center for Strategic and International Studies.
• “So what it means in practice is that, from an American point of view, a system based on determining the bona fides of a particular end user becomes a very difficult system to operate and enforce,” he said.

The goal of the U.S. effort is to block China’s military from gaining access to advanced chip tech, but also to prevent officials there from using that tech to commit human rights violations, according to two people familiar with the administration’s thinking.

• But that goal is increasingly difficult to achieve without broad restrictions, since any chips shipped to China could be repurposed for domestic uses such as surveillance or added to the military’s arsenal.
• “Because if the Chinese are saying all end users are eligible to have their technology used for military purposes, that means there is no reliable end user anymore in China,” Reinsch said.

The tech-based approach to blocking China’s access to advanced chips draws a clear line through the various layers of the chipmaking process.

• At the top of the process, the Commerce Department has issued an export control rule on design software that will hamper China’s ability to design next-generation advanced chips.
• The design software restrictions are an extension of a restriction on ASML from exporting next-generation EUV manufacturing tools to China, which are necessary for the most advanced manufacturing techniques.
• Other manufacturing systems are in the crosshairs too: U.S. officials are working on a tool-by-tool basis to determine which manufacturing equipment is necessary to produce FinFET transistors — a current generation of tech — and block those.
• And should a chip company build a product that the U.S. is uncomfortable with China getting its hands on, Wednesday’s new license requirements make clear it will restrict exports of those chips.

There is likely more to come. Reinsch said that in addition to the refinements the Commerce Department made to the recent notification to Nvidia, which the company disclosed early Thursday, the Biden administration will likely roll out more mechanisms and export controls in the coming weeks and months.

• The foreign direct product rule is an example of another level the administration could pull.
• Using the FDP rules, the U.S. could theoretically prevent chips from being exported that have major components designed or manufactured with U.S. technology or software.
• That means, even if a Chinese fabless chip company designs its own advanced processor and has it manufactured in Taiwan, the U.S. might be able to block its re-export back to China since many of the best chip manufacturing tools are made by U.S. companies.

A MESSAGE FROM VMWARE

VMware sits at the center of the multi-cloud universe and is focused on providing consistency across clouds, enabling choice of location and delivering best of breed capabilities.

Learn more

Billable hours: Can AI bring order to law documents?

There’s no shortage of lawyer tropes in pop culture. How about that one where the overworked legal associate pores over dusty books and files late into the night, a half-eaten container of cold Chinese takeout by her side?

AI is changing that, right? Yes … and no.

Today, tech-savvy law firms are hiring data scientists to build machine-learning models that use natural-language processing to organize unstructured, often esoteric legal documents. They might build a machine-learning model to decipher the work one lawyer has done and determine which legal issues that lawyer has dealt with most, for example.

“Historically, we’ve relied on a person for that,” said Sean Monahan, director in the Legal Transformation and Innovation practice at HBR Consulting. “You can actually show how legal issues within a business are growing or shrinking … What AI is letting us do is slice our legal work and apply much more rigid structures.”

But in general, tools designed to bring digital order to paper chaos are just now starting to figure out context of any kind (say, tax documents).

Not many NLP and analytics startups have tackled legal yet, said Martin Catania, president of legal IT consultancy Keno Kozie.

“The general perception is that it’s hard to do these things in legal,” Catania said. And when he’s evaluated startup software for analyzing unstructured data to see if it works in legal contexts, he said, “companies are overpromising.”

— Kate Kaye (email | twitter)

Around the enterprise

Regulators in the U.K. imposed new cybersecurity restrictions on telecommunications companies, arguing that they’re a unique target for cyberattacks.

Dell successfully challenged a cloud computing patent that it was accused of infringing, further evidence that cloud companies have escaped much of the patent trolling that accompanied the rise of mobile computing.

A MESSAGE FROM VMWARE

VMware sits at the center of the multi-cloud universe and is focused on providing consistency across clouds, enabling choice of location and delivering best of breed capabilities.

Learn more