(TNS) — Fueled by a global artificial intelligence boom, the global race for computing power is in full swing. OpenAI's Sam Altman is trying to raise $7 trillion for his chip manufacturing venture. Major technology companies such as Microsoft and Amazon are developing their own AI chips. The need for more computing power to train and use AI models will drive the quest for everything from cutting-edge chips to huge data sets, but It is not only a source of current geopolitical influence (such as curbing chip exports); It's also shaping how countries will grow and compete in the future, with governments from India to the UK developing national strategies and stockpiling Nvidia graphics processing units.
I believe it's time for America to develop its own national computing strategy: the Apollo program for the AI era.
In January, pursuant to President Biden's executive order on AI, the National Science Foundation launched the National AI Research Resource (NAIRR) pilot program. The program is envisioned as a “shared research infrastructure” that provides AI computing power and access to open government and non-government data. set, and training resources for students and his AI researchers.
The NAIRR pilot is critical, but it's just the first step. His NAIRR task force's final report, released last year, outlined the final $2.6 billion budget needed to operate his NAIRR over six years. That's far from enough, but it still remains to be seen whether Congress will authorize NAIRR beyond the pilot.
Meanwhile, more needs to be done to expand government access to computing power and bring AI to national services. Advanced computing is now central to our nation's security and prosperity. We need it to optimize our nation's intelligence, pursue scientific breakthroughs like nuclear fusion, accelerate the discovery of advanced materials, ensure the cybersecurity of our financial markets and critical infrastructure, and more. The federal government was instrumental in enabling major technological advances of the last century by providing core research infrastructure such as particle accelerators for high-energy physics in the 1960s and supercomputing centers in the 1980s. played a very important role.
With other countries around the world now making sustained and ambitious government investments in high-performance AI computing, we cannot risk falling behind. It's a race to develop the most world-changing technology in human history.
First, we need to build more dedicated government AI supercomputers for a range of missions, from sensitive information processing to advanced biological computing. In modern times, computing power and technological advances move hand in hand.
Over the past decade, the United States has taken classic scientific computing to exascale with Frontier, Aurora, and the upcoming El Capitan machine, a massive computer capable of performing more than 1 quintillion (1 billion) operations per second. succeeded in pushing the era. Over the next 10 years, the power of AI models is predicted to increase by 1,000x to 10,000x, with major computing architectures potentially able to train 500 trillion parameter AI models in a week (for comparison , GPT-3 has 175 billion parameters). Supporting research of this scale will require a more powerful and dedicated AI research infrastructure, significantly better algorithms, and more investment.
Currently, the United States still leads in advanced computing, but other countries are approaching parity and trying to overtake the United States. For example, China aims to increase its total computing power by more than 50% by 2025, and is reported to be deploying 10 exascale systems by 2025. We cannot risk acting too slowly.
Second, while some may argue that we should use existing commercial cloud platforms instead of building a high-performance federal computing infrastructure, I believe we need a hybrid model. Masu. Research shows that using federal computing instead of commercial cloud services can significantly reduce long-term costs. In the short term, scaling up cloud computing provides a basic level of access to projects that is quick and streamlined. This is the approach the NAIRR pilot is taking, with contributions from both industry and federal agencies. But in the long term, acquiring and operating powerful government-owned AI supercomputers with a dedicated mission to support the needs of the U.S. public sector will help make AI more widespread and improve national security. and preparations will be made for an era in which the world will become the center of prosperity.
This expanded federal infrastructure could also benefit the public. The life cycle of government computing clusters has traditionally been approximately seven years, after which new systems are built and old systems are retired. Inevitably, as newer, cutting-edge GPUs emerge, hardware updates will phase out older supercomputers and chips, allowing them to be recycled for low-intensity research and non-commercial use, making them cost-effective for consumer use. Adds high computing resources. So far, most advances in AI have been driven by universities and the private sector, but fully distributed models will increasingly face computing constraints as demand soars. In a survey of the nation's largest computing users conducted by MIT and the nonprofit American Competitiveness Council, 84% of respondents said they faced computing bottlenecks in running key programs. did. If the United States is to stay ahead, it will need massive investment from the federal government.
Third, the national computing strategy must be aligned with the talent strategy. Governments can better compete with the private sector for AI talent by providing workers with opportunities to tackle national security challenges using world-class computational infrastructure. To ensure that we can provide a large and sophisticated workforce for these highly technical and specialized roles in the development and implementation of AI, America will also recruit and recruit talented global students. need to be maintained. A key part of this effort is establishing clear migration routes. For example, it would exempt people with Ph.D.s in related technical fields from the current H-1B visa cap. Fundamentally rethinking how computation is done, shaping AI for the public good, and spearheading a new paradigm that can push the boundaries of technology and bring its benefits to everyone. requires the brightest minds.
America has long benefited from its position as a global driver of innovation in advanced computing. Just as the Apollo program galvanized our nation to win the space race, setting national goals for computing will not only make AI more competitive in the coming decades, but also make it more accessible. It will foster breakthroughs in research and development in virtually every field. Advanced computing architectures cannot be built overnight. Start laying the foundation now.
© Copyright 2024 Technology Review, Inc. Distributed by: TRIBUNE CONTENT AGENCY, LLC.
