📚The TL;DR📝

1. “Attention is All You Need:” Research paper published by Google Brain in 2017, proposed the “transformer,” the underlying technology for modern artificial intelligence.

2. Since WWII, the U.S. has led scientific innovation through the National Science Foundation and the “triple-helix model”, a collaborative effort between government, universities, and private entities. 

3. However, in the last four decades, the triple helix model has declined, with private industry stepping in as the primary funder, innovator, and developer of AI.

4. The release of DeepSeek, a Chinese large-language model (LLM), prompted panic in U.S. policymakers, renewing the desire to solidify U.S. AI dominance.

5. Only private industry, rather than government, has the infrastructure funding required to compete with China in AI. This means that while the U.S. government can accelerate the technology, it cannot control the course of its development. 

📌How America Stopped Directing AI Innovation📌

On January 20, 2025, DeepSeek, a Chinese artificial intelligence (AI) startup, debuted to the public. President Donald Trump called it a “wake-up call” to U.S. companies; Marc Andreessen, a prominent venture capitalist, warned, “DeepSeek R1 is AI’s Sputnik moment.” The “Sputnik moment” refers to Washington’s urgency following the Soviet Union's launch of the Sputnik satellite during the Cold War.

Following the 1957 “Sputnik moment,” the U.S. government marshalled significant resources, doubling its research and development (R&D) budget, to compete with the Soviet Union’s innovation. This period represents the height of U.S. federal R&D funding and the “Triple-Helix Model,” which emphasized collaboration between the government, universities, and private industry. After the Cold War, U.S. federal R&D funding declined while private industry spending on R&D rose, filling the federal funding vacuum.

Over the last two decades, the private sector has led the funding, development, and deployment of American AI. During today’s “Sputnik moment,” private industry, rather than the U.S. government, is responsible for keeping pace with China. A series of federal policy decisions starting in the 1960s put AI technologies critical for national security in the hands of private corporations. 

How the Triple Helix Came Together

The arc of the U.S. AI policy, from leading innovation to watching private companies steer acceleration, began with Vannevar Bush, the father of the triple-helix model. During World War II, Franklin D. Roosevelt understood that scientific advancements were crucial to national defense. He recruited Bush, an esteemed inventor and professor of electrical engineering at MIT, to lead cooperation between the government, military, and private entities at the Office of Scientific Research and Development (OSRD). At OSRD, Bush oversaw the physicists working on the Manhattan Project who created the first atomic bomb, as well as the development of radar, instant coffee, and penicillin. 

Near the end of the war, in 1944, President Roosevelt directed Bush to determine how science could promote national health, prosperity, and welfare. Bush’s report, entitled Science, the Endless Frontier, laid the groundwork for post-war U.S. science policy and established the National Science Foundation (NSF). Under Bush’s direction, NSF would conduct research on military problems, agriculture, housing, public health, and medicine. Bush decreed that the government should collaborate with private entities, colleges, universities, and research institutions to “foster the opening of new frontiers.” This collaboration, later dubbed the Triple-Helix Model, combines universities, government, and industries to shepherd an innovation from research to development and deployment. In the years after the war, federal spending on R&D outpaced the private sector, and collaboration thrived among the three parties of the Triple-Helix Model.

Cold War fears about Soviet competition fueled NSF funding. After the Soviet Union launched its Sputnik satellite in 1957, NSF’s total budget funding more than doubled. From the 1950s to the mid-1980s, the agency supported projects on satellite transmissions, deep-sea life, 3D printing, and weather prediction. In 1985, NSF linked a nationwide network of computer facilities to the NSFNET, the first “backbone” of the Internet. 

The increase in federal scientific R&D, both during and after the war, led to the emergence of patents and job opportunities in concentrated geographic areas, known as local research ecosystems. After the war ended, researchers continued to work in those same ecosystems, utilizing the expertise they had gained from their work on OSRD programs to develop innovations, patents, and companies. Government funding fostered precursor innovations and talent pools that set the stage for private entities to assume the burden of funding R&D in the late 1980s and 1990s. In fact, federal spending on R&D as a share of total R&D spending peaked at 67% in 1964 but decreased to 40% in 1990 before falling to 19.5% in 2020, limited by post-Cold War retrenchment and restrictions on federal commercialization of technology. One example of this shift is the NSFNET project, which laid fiber optic cable to expand the Internet to more research universities. The federal component of the project ended in 1995 because NSF could not lawfully finance and facilitate a commercial entity. NSF then allowed companies to build their own “backbones” for providing Internet service. 

How it Fell Apart

American artificial intelligence development mirrored this broader trend in scientific research. In the 1950s, researchers first developed computational methods and learning algorithms that laid the foundation for AI. Over the following decade, federal R&D funding grew 14% annually. Frank Rosenblatt, a researcher in the U.S. Office of Naval Research, invented a machine that could distinguish between marks on the left and right of cards, the first example of independent machine learning. Throughout the 1970s and 1980s, the failures of Vietnam and a stumbling economy created backlash against scientific R&D. Although a lack of prominent inventions during that time has led some scholars to title the period an ‘AI winter,’ membership at AI research institutes and references to AI in publications rose. That scholarship, coupled with NSF funding projects at universities and private entities, produced advancements in neural networks and reinforcement learning in the 1980s and 1990s. 

Following the end of the Cold War in 1991, the U.S. significantly reduced its defense budget and encouraged private firms to consolidate. Private firms replaced the government’s long-term R&D projects with a focus on short-term profits. When defense spending spiked after the terrorist attacks of September 11, 2001, the government offered federal contracts to major companies. These contracts increased revenue but solidified the financialization of the industry, with firms prioritizing profit and shareholder payouts over reinvestment in research. Meanwhile, venture capital (VC) firms and corporate research labs filled the funding vacuum left by the federal government, contributing to the rise of Silicon Valley startups. The Triple-Helix model, where universities, government, and industry collaborated to produce innovations, eroded in favor of a private enterprise ecosystem. 

Private ecosystems funded and developed emerging technologies with minimal government support and involvement from universities. For example, in 2017, Google Brain, Alphabet’s AI research division, released a paper entitled “Attention is All You Need,” which proposed a new form of computational modeling to replace the models of the 1990s. OpenAI and the Google Brain Project employed this model to build the underlying LLM technology. 

Another Sputnik?

The U.S. government tried to keep pace with new developments in artificial intelligence. A year after OpenAI’s founding, in 2016, the Obama administration formalized the U.S.’s earliest AI strategy, offering five broad goals: (1) collaborate with industry,  (2) incorporate AI into federal agencies, (3) investigate the effects of AI, (4) educate the public, and (5) monitor international developments. In his first term, President Trump echoed these priorities in his AI strategy plan. While President Biden advocated for increased precautions and oversight, he did not enact any substantive policy to that effect. 

The release of DeepSeek, China’s LLM model on January 27, 2025, prompted panic in Washington. Policymakers expressed concern that Chinese scientists, educated within China’s own scientific ecosystem, had developed an efficient model that required less advanced chips to run at relative parity with U.S. models. This ‘Sputnik moment’ for the U.S. triggered a renewed fervor for technological prowess. In July 2025, President Trump released America’s AI Action Plan, which addressed all five of President Obama’s policy priorities. In light of state regulations creating onerous reporting requirements or rules for AI technologies, President Trump stated that the federal government “should not allow AI-related Federal funding to be directed towards states with burdensome AI regulations.” 

🌎Why It Matters🌎

Regardless of the political party in power, the U.S. government is intent on accelerating its AI technology. However, it is accelerating a technology that it did not create and does not own. When the U.S. Triple-Helix Model disintegrated in the late 1980s, private companies swooped in, taking over the lion’s share of AI R&D funding. As private companies centralized control and proliferated their AI technologies, the government has refrained from regulating the AI. To continue accelerating, the government integrates these proprietary AI systems into its operations. However, the government cannot control the AI’s development, nor can it use the proprietary algorithms to invent its own AI system. Instead, it can only accelerate the technology through regulatory incentives and integrate it into its own operations.

The government is at the mercy of the emerging private AI technologies. While it reaps the rewards of the flexibility and resources of the private marketplace without expending its own capital and manpower, it risks integrating AI’s algorithmic flaws into its operations. For example, AI systems are being integrated into drone warfare. The Department of Defense’s (DOD) program, JADC2, employs AI to expedite the decision to select a target. The AI improves efficiency and speed in warfare, but the U.S. risks that the AI relies on flawed algorithms and strikes the wrong target. 

These risks are enormous, yet inevitable. Private companies control the underlying technology and can adjust it independently. They will reckon with balancing profit and navigating the unknown and potentially dangerous effects of AI. Meanwhile, the U.S. government and the American public will cope with the consequences.