Replacing Nvidia is a tall order. While Chinese competitors are years behind the company’s cutting-edge technology, many analysts and insiders warn they are catching up, thanks to U.S. export restrictions.

U.S. chip restrictions on the sale of advanced semiconductor technology, especially those used in artificial intelligence, have been rolled out over several years, with the initial aim of curbing China’s military advancement and protecting US dominance in the AI industry.

However, according to Nvidia CEO Jensen Huang, U.S. semiconductor export controls on China have been “a failure,” causing more harm to American businesses than to China.

While the goals of cutting back the Chinese military’s access to advanced U.S. technology and maintaining U.S. leadership in AI appear to have had some success on paper, loopholes and existing semiconductor stockpiles in China have complicated these aims, said Ray Wang, an independent tech and chip analyst with a focus on U.S.-China competition.

“That’s partly why we are seeing a closing of the gap between Chinese and U.S. AI capabilities,” added Wang.

While Washington’s most comprehensive export controls were passed during former U.S. President Joe Biden’s term in the White House, curbs on Huawei and SMIC, China’s largest chipmaker, go back to Donald Trump’s first term in office.

On April 15, Nvidia disclosed that new controls, which restricted sales of its H20 graphics processing units to China, had led to a $5.5 billion charge against its revenue.

The restrictions are expected to be a boon for the demand and development of local Nvidia alternatives like Huawei, which is working on its own AI chips. They also come against the background of Beijing mobilizing billions as part of its chip self-sufficiency campaign. 

“The bottom line is, the controls have incentivized China to become self-sufficient across these supply chains in a way they never would have contemplated before,” Triolo said. 

Chinese AI-related achievements, such as DeepSeek’s R1 model and news of Huawei chip progress, have led observers to question the effectiveness of chip controls. 

According Wang, the independent analyst, China’s semiconductor and AI space has seen an acceleration of startups, market opportunities, and AI talent alongside the restrictions, which has clearly resulted in domestic innovations. 

“I think the arguments that export controls accelerate innovation is quite valid,” Wang said. 

Nivida’s Haung also noted these trends in April, telling lawmakers in Washington that the country has made enormous progress in the last several years and is right behind the U.S. 

Nvidia’s H20 chip was designed specifically to comply with existing chip controls prior to the clampdown on exports.

“We are not just talking about one export control, we are talking about a series of export controls that originate from all the way back in 2019,” said Wang, noting that the evolving policies have had a couple of different objectives. 

Meanwhile, in what DGA’s Paul Trilio calls a “moving of the goalposts,” it seems that the aims of the restrictions have shifted to an intention to slow down and contain Chinese AI and semiconductor developments. 

“The continued expansion of the controls, and the lack of an articulation of what the clear end game here is, has really created a lot of issues, and created a lot of collateral damage,” Trilio said, adding that it has led more people to question the policy. 

In a statement earlier this month, the Information Technology & Innovation Foundation, a U.S. think tank which has received funding from various technology companies, said in a post that “the Biden administration’s export control policy for AI chips has largely been a failure since day one. Yet, year after year, it has doubled down, attempting to plug various loopholes.”

“While [the U.S. government] is certainly right to prevent U.S. companies from selling advanced AI technology to the Chinese military, cutting U.S. companies off from the entire commercial Chinese market is a cure worse than the disease,” Stephen Ezell of ITIF told CNBC in an email.

“U.S. export controls have cost NVIDIA at least $15 billion in sales, and those are revenues the company needs to be able to earn to invest in future generations of innovation.”

Bitcoin continued its rally on Thursday, hitting a brand new record high above $111,000.

Bitcoin hit $111,886.41 in early trading hours in London, according to Coin Metrics, before paring some of those gains to trade at around $111,012.00 at 07:03 a.m. London.

Bitcoin’s move has been “driven by a mix of positive momentum, growing optimism around U.S. crypto regulation, and continued interest from institutional buyers,” James Butterfill,  head of research for crypto-focused asset manager CoinShares, told CNBC by email.

The price rise in world’s largest cryptocurrency is taking place despite a drop in U.S. stock markets on Wednesday.

Bitcoin has typically correlated with equity markets, particularly the tech-heavy Nasdaq.

The diverging movements of bitcoin and stocks could be the result of investors looking for alternative stores of value.

“The rally was also helped along by broader macro concerns, including Moody’s recent downgrade of U.S. sovereign debt, which added to the narrative of Bitcoin as a hedge against fiat instability,” Butterfill noted.

Ratings agency Moody’s cut the United States’ sovereign credit rating last week.

There have been some positive developments for the crypto space on the regulatory front in the U.S. too. The GENIUS Act — a bill to regulate stablecoins — cleared a key procedural vote in the Senate.

U.S. President Donald Trump and his AI and crypto czar David Sacks have pushed forward a pro-crypto agenda in the U.S., which has helped support the market.

Adding to upbeat news for crypto, JPMorgan CEO Jamie Dimon, a notable bitcoin skeptic, said that the bank will allow clients to buy the digital currency.

– CNBC’s MacKenzie Sigalos contributed to this story.

The artificial intelligence boom has sent energy demand soaring. Some of the supercomputers sucking up all that power are helping to find new energy sources.

Fusion energy is the process of forcing two hydrogen atoms to combine and form one helium atom, which releases huge amounts of power. It uses a stellarator, a type of fusion reactor invented in the 1950’s that produces heat.

Until now, the technology was too difficult to deploy commercially.

But this old concept has brand new potential. Type One Energy, a startup based in Tennessee, claims to have proven that fusion energy will be able to produce electricity in the next decade.

“It’s going to create heat that’s going to boil water, make steam, run a turbine and put fusion electrons on the power grid on a 24/7 reliable basis,” said Type One Christofer Mowry.

AI has made it all practical.

“Things have really accelerated remarkably over the last five or six years,” Mowry said. “The supercomputers have allowed industry, academia and large institutions to develop now and actually test at large scale the science machines that demonstrate the process.”

Dozens of other companies are working on different approaches to fusion energy, but Mowry said Type One is so far the only one with the proven stellarator technology to implement at existing power plants. It will soon be tested with the Tennessee Valley Authority.

TDK Ventures is betting that Mowry is right.

“With Type One Energy solutions, we expect outsized return potential,” said Nicola Sauvage, president of TDK Ventures. “Fusion is no longer science fiction, and Type One Energy’s technology is catching up fast to the vision of this low-cost, continuous green energy.”

Type One is also backed by Breakthrough Energy Ventures, Centaurus Capital, GD1, Foxglove Capital, and SeaX Ventures, and has raised a total of $82.4 million.

Fusion energy is different from nuclear power, and there’s no risk of a nuclear accident. The power source has no long-term radioactive waste, and, according to Mowry, can’t be weaponized.

But for handling AI, it could be a critical solution. Fusion energy can be deployed anywhere, whether it’s next to a data center or near a large industrial park that needs clean, reliable energy.