Silicon Valley Should Carefully Assess Its AI and Semiconductor Dealings in the Middle East

Sam Altman, the controversial founder of OpenAI and an artificial intelligence (AI) evangelist, is reportedly on a global quest to raise trillions from Middle East governments to create a new, international network of AI chip factories. Nvidia CEO Jensen Huang recently traveled to the region and declared that every country should seek a sovereign AI capability by buying more chips and building their own AI computing infrastructure. Others are seeking to raise billions from the region for designing more AI chips. These are noble ideas, given that training future AI models will require more and faster AI chips, but doing so in a region with questionable ties to China raises possible national security concerns and will not increase U.S. supply chain resilience or long-term technology leadership. 

Silicon Valley knows well that ChatGPT is only made possible by powerful advances in chip technology, and that more silicon innovation is needed to push forward the boundaries of AI. However, prices for the most advanced AI chips have skyrocketed due to both increasing chip complexity and red-hot demand. A silicon arms race has already broken out between the world’s largest tech firms and governments to secure supply. Meta’s Mark Zuckerberg recently said he wants to amass 350,000 of the most advanced AI chips, and others are following suit. So, it is no surprise that customers of these chips are hoping to boost supply and that the chipmakers are hoping to boost demand with sovereign investors willing to invest at a high premium with a longer-term outlook. 

However, in the silicon space, money is not the only ingredient for success, and sovereign investors are looking for more than just a financial return. 

In fact, the chip industry is moving to meet demand, with chip factory investments expected to reach record highs in the coming years. This includes the AI silicon supply chain: the Taiwan Semiconductor Manufacturing Company (TSMC) is building a new 2-nanometer chip factory in Taiwan, memory chipmakers are expanding their high-bandwidth memory (HMB) production, and advanced packaging supply chains are also boosting output. But OpenAI and others believe this is not enough. It is not easy to boost supply immediately, as the entire supply chain for advanced chip design and production must move in unison to fuel rapid AI advances, and this takes time. Chip factories are massive investments bristling with complex precision machinery; it takes years to build and pump out chips, and the process requires an army of scientists and engineers. Today, only three companies in the world are able to produce the most advanced logic chips for AI.

Silicon Valley leaders would be wise to rethink their attempt to recreate the semiconductor ecosystem from scratch or expand supply so fast, as it could overstress a fragile industry. This would be a staggering and unnecessary investment that would drain valuable human talent from chipmakers who already face a severe talent shortage. On the off chance of success, it would create a supply glut, wreaking havoc on the chip industry. 

Furthermore, building an alternative AI silicon and computing ecosystem in an authoritarian region cozy with China and Russia could pose risks to U.S. national security. Already, prominent technology groups based in the Middle East, such as G42, have drawn scrutiny for apparent close ties to China. While G42 has reduced some of its investments in China, a larger concern is potential Chinese access to high performance computing (HPC) systems based in the region—access that would likely otherwise be restricted if built in China. This is not a hypothetical concern. For example, the King Abdullah University of Science & Technology (KAUST) in Saudi Arabia, which owns multiple powerful supercomputers and is building more, has close ties to China. KAUST's founder, who penned an op-ed proclaiming “America Can’t Stop China’s Rise,” recently openly advertised its supercomputers during a trip to China, and signed a memorandum of understanding with the Shenzhen government. KAUST also has extensive HPC research collaborations in China, including co-published scientific papers, and has hosted visiting fellows engaged in HPC research from multiple Chinese defense-related universities. Meanwhile, Chinese government technology researchers late last year visited and discussed biotech, quantum, and AI collaboration with the Abu Dhabi Technology Innovation Institute, which has recently purchased thousands of powerful Nvidia H100 AI chips that are restricted for sale in China. Blacklisted Chinese tech firms Huawei and Sensetime are also expanding their AI and computing footprint in the region. 

While Silicon Valley should be able to compete in the region with appropriate regulation and security mechanisms to curtail unlawful access to controlled technology, they can and should refocus their bolder silicon and AI investment ambitions by helping the United States and friendly partner nations reshore advanced chipmaking and expand AI data compute power. 

Big tech’s help is sorely needed in chipmaking. Major chipmakers Intel and TSMC have announced delays in building their new U.S. factories due to a lengthy rollout of U.S. government CHIPS Act funding and uncertainty if customers will pay for U.S.-made chips. OpenAI, Softbank, and other AI firms can help fix this problem by partnering jointly with chip designers, chip foundries, the federal government, and private investors to create a new super-sized U.S. chip foundry joint venture. This venture should exclusively produce the most advanced logic chips for AI and, by directly linking supply with demand, could give much-needed certainty to chipmakers. Government CHIPS Act funding helps buy down the risk, while customer co-investment ensures demand. This is not new or novel, and it works. Leading Taiwanese chip foundry TSMC is already using this model in both Germany and Japan and has seen initial success. As a result, TSMC’s projects in those regions are moving relatively quicker than their delay-plagued U.S. factory in Arizona.

In AI data center spending, the U.S. government is woefully behind the private sector and lacks any coherent plan. Instead of raising billions for massive data centers in the Middle East, Silicon Valley could partner with the U.S. government and Congress to develop and fund a National AI Computing Initiative that could grant billions to federal agencies, state governments, and leading research labs and universities to build out advanced AI data centers for cutting-edge research. This could be like the Eisenhower interstate highway initiative but for the generative AI era. We should also be making investments in chip design architecture and software improvements that can help develop faster and more efficient chips. Only by increasing investments in HPC chip architectures and systems can the United States remain a technology leader. 

Silicon Valley should take a harder look at the Middle East and think first about realizing its quixotic AI and silicon ambitions here in the United States. The U.S. government should also consider leveraging demand-side incentives such as government procurement and tax credits to ensure that the United States remains a global AI and silicon ecosystem leader. These efforts may be harder to do and take longer to realize, but with the right vision and focus, the long-term benefits to the United States would likely be tremendous. 

Jimmy Goodrich is non-resident senior associate with the Wadhwani Center for AI and Advanced Technologies at the Center for Strategic and International Studies in Washington, D.C.

James Goodrich
Senior Associate (Non-resident), Wadhwani Center for AI and Advanced Technologies