AI innovation

This issue brief is part of the GeoTech Center’s “Atlantic Council Commission on Artificial Intelligence: US leadership in the age of AI” report, which offers an action-oriented roadmap for strengthening US domestic AI capacity, aligning with allies, and sustaining global leadership.

US leadership in AI innovation set the technology on its current trajectory. While the recent acceleration of capabilities has occurred over increasingly short time frames, the foundation on which it rests was constructed over the course of decades—fueled by long-horizon federal investments in basic research, a world-leading academic enterprise that attracted the best and brightest from across the globe to pursue their ideas in the United States, and a vibrant private-sector environment that provided funding and pathways for new ideas to reach the market.

The rest of the world took note. Competitors and allies alike have modeled national AI strategies on US strengths. China, in particular, has pursued a series of structured national plans aimed at achieving global leadership in AI innovation and adoption. The results have been substantial. Chinese universities now rank among the top institutions in the world, academic publishing on AI out of China has surged, and a vibrant AI start-up ecosystem has emerged. In 2025, China’s approach evolved to make bets on open-source AI and prioritize AI integration. Based on usage data from the OpenRouter platform, over the course of 2025, Chinese open-source models grew to account for 30 percent of global AI use, up from 1.2 percent at the end of 2024. According to analysis published by Hugging Face, the most popular platform for open-source AI development, over the course of 2025 open-source models from China surpassed those from the United States in both monthly and overall downloads, swiftly accounting for the plurality of downloads on the platform.

While US frontier labs continue to release the most advanced, capable models that exist, recent reports illustrate how once-authoritative gaps continue to narrow. This state of affairs demands the United States assess its current approach and take decisive steps to address the flashpoints and cracks that are emerging across the nation’s innovation landscape.

Weakening academic base

US innovation has long been sustained by the wellspring of ideas and talent developed through the nation’s academic enterprise, the strength of which has been enabled through sustained federal funding for research and education initiatives. The Information Age would not have existed without the creation of the internet in a narrow science community funded by the US Advanced Research Projects Agency (now the Defense Advanced Research Projects Agency) and the US National Science Foundation (NSF) before the technology became commercialized for the broader public. The techniques of deep learning and reinforcement learning that underpin the impressive capabilities of today’s AI models came out of federally funded academic labs. However, academia’s role in the future of AI has been overshadowed by that of the private sector in recent years. And while the importance of AI has accelerated, federal funding for AI research and development (AI R&D) over the course of the past four years has leveled out at around $3.3 billion annually. This plateau comes in spite of efforts, such as those authorized in the CHIPS and Science Act of 2022, to boost federal support for research, development, and education in technical areas of strategic importance such as AI. Since the passage of the act, appropriated spending for agencies such as NSF, the US Department of Energy (DOE), and the US National Institute for Standards and Technology (NIST) have fallen well short of the levels authorized in the legislation.

Limited funding for AI research in academia means a shortage of technical resources that are critical for AI innovation and academic research applications—most importantly, access to advanced computing resources. This has set up a widening gap between private-sector R&D initiatives that have access to the computational resources needed to innovate on the cutting edge of AI and academic endeavors that do not. Per a Stanford University report: “Among academics with access to compute, it is typical for them to have access to between 1 and 8 AI chips, whereas industry researchers may have access to thousands.” A federal initiative, the National AI Research Resource (NAIRR), was launched as a public-private partnership in 2024 with the goal to connect researchers with the necessary computational, software, and data resources (including both federal and commercially available options) to sustain AI innovation in academia and small businesses. However, the research community continues to face challenges in accessing computational resources. This creates barriers to sustaining essential R&D that might not have direct commercial incentive but sets the groundwork for future breakthroughs, including novel approaches to more generalizable capabilities and improvements related to AI assurance, interpretability, and security. In addition, this lack of access slows the pipeline of talent development on which the private sector relies for growth.

Availability and accessibility of data 

High-quality data enable AI systems to generate analysis and predictions accurately and efficiently, which becomes particularly salient when developing AI models for specialized tasks in high-impact domains such as healthcare, manufacturing, and science. The Trump administration’s AI Action Plan sets a goal for the United States to “lead the creation of the world’s largest and highest quality AI-ready scientific datasets.” In November 2025, Trump launched the Genesis Mission, which aims to build an integrated AI platform to leverage scientific datasets to drive scientific discovery through DOE’s national laboratories. However, many factors still hinder access to data for public benefit, including the restriction of access to corporate-owned data, regulatory uncertainty around data governance, and the downsizing of statistical agencies. Within organizations, data readiness remains a prerequisite for AI innovation and adoption. However, challenges related to data siloing, governance, quality, and interoperability often limit the ability of organizations across sectors to harness AI for their benefit. Increasing the availability of data is instrumental in optimizing task- and sector-specific AI applications and fueling innovation in both academia and the private sector. Yet questions remain about how to responsibly expand use of federal datasets, incentivize open access, and better leverage technical options such as synthetic data and other privacy-enhancing technologies.

Uneven AI adoption across sectors  

China recently issued its “AI Plus” initiative, aimed at promoting integration of AI across various sectors including science and technology, industrial development, consumption, social welfare, governance, and global cooperation, with the overarching goal of reaching AI penetration of 70 percent across its economy by 2027. This stands in contrast to the US model, which to this point has centered more around pursuit of AGI and a race for the next generation of model rather than the integration of current AI capabilities into economic sectors. Recent Census data show a picture of highly uneven AI usage rates by sector in the United States: while the information sector has a 38 percent adoption rate, the manufacturing sector has 13 percent usage, and the transportation sector’s rate is even lower at 8 percent. Uneven application of AI across the economy prevents innovative experiments and use cases from being developed in diverse sectors, undermining US competitiveness in AI and weakening the ability to build public trust through the demonstration of promising AI capabilities and innovations.

Infrastructure readiness

Compounding infrastructure deficits across the US economy present challenges for national efforts to push forward on AI adoption. As detailed in Chapter 7 of this report, the nation’s aging energy grid requires modernization to meet the demands of data centers. Simultaneously, the country’s connectivity infrastructure must evolve to support the high-bandwidth, low-latency networks demanded by AI workloads, a challenge that will become more pressing as AI increasingly moves into edge devices and into the physical world through robotics. In addition, existing digital divides between rural and urban areas across the country will deepen if not addressed. Across US industries, accumulated technology debt, legacy hardware, and flawed or immature data infrastructure leave many organizations across critical sectors without the foundational capacity required to deploy and scale AI effectively.

Risks of frontier models and AGI

As AI capabilities rapidly progress, so does the need to address risks related to safety, security, and transparency. The increasing rate of change from the relatively narrow use of large language models to more complex, autonomous deployment of agentic AI and multimodal AI adds urgency to the need for technically grounded approaches to responsible deployment and governance. Organizations such as the Frontier Model Forum represent industry-led efforts to collaborate on understanding and managing risks to public safety, such as from the intersection of AI with chemical, biological, radiological, nuclear weapons and advanced cyber threats. Governments around the world have established national institutes focused on various elements of the AI safety and security paradigm. In the United States, the Center for AI Standards and Innovation, housed within NIST, collaborates with the private sector on evaluations and assessments of potential AI security vulnerabilities and has carried out more than forty evaluations of frontier AI models. It recently launched an AI Agent Standards Initiative focused on “ensuring a trusted, interoperable, and secure agentic frontier.” Globally, the International Network for Advanced AI Measurement, Evaluation, and Science, previously led by the United States and now led by the United Kingdom’ AI Security Institute, promotes the exchange of knowledge and best practices among national institutes. Such efforts are paramount as AI capabilities progress rapidly.

Finding: The US government’s role as a promoter of AI innovation remains essential. US industry relies on the ideas and talent that come out of the US university ecosystem—the strength of which is predicated on sustained federal support. Over-reliance on the private sector to drive AI innovation will constrain the opportunity space for new approaches and applications, weakening the nation’s ability to innovate at the pace of global competition. Action is needed to ensure that the foundations of US innovation can continue to power progress.

• Recommendation: Strengthen funding for AI research and development. Federal support is the linchpin of the US national research enterprise. Federal funding should focus on strengthening associated components of the broader AI ecosystem and supply chain, in addition to advancing AI capabilities. This includes fostering innovation in AI hardware design and manufacturing, cloud infrastructure, next-generation networking, and advancing novel AI applications in critical areas such as healthcare, energy, manufacturing, and transportation. The commonly cited recommendation from the National Security Commission on AI to double annual federal funding for non-defense AI R&D year over year until it reaches $32 billion remains a worthy goal.
• Recommendation: Explore novel funding structures. The pace of technological advancement and the nature of AI ecosystems mean that the standard approaches to federal support might no longer suffice. The government should explore and expand funding mechanisms such as large-scale funding for top investigators in the model of the Howard Hughes Medical Investigator program (which has led to more than thirty Nobel Prizes), support for focused-research organizations, rapid grants to allow for quick reaction to technological developments, and joint research programs that can link students directly with private-sector AI projects to enable graduates to contribute immediately to frontier innovation. It is also worth sustaining NSF efforts to pioneer promising approaches to spur regional innovation and build nimble research teams that can accelerate innovations to market.

Finding: Investing in the infrastructure for innovation and adoption will power progress. Funding for AI and AI-related R&D is necessary but insufficient on its own to fully harness the potential of AI at the national level. Recent federal initiatives, such as the Genesis Mission and the NAIRR, represent promising steps toward establishment of the required infrastructure to sustain innovation, but additional action is needed to achieve the envisioned impact and to support broader integration into the economy.

• Recommendation: Resource promising federal initiatives like the Genesis Mission and the NAIRR. If adequately resourced with federal funding that transcends political cycles, the Genesis Mission and the NAIRR hold the potential to supercharge AI innovation: the Genesis Mission from the perspective of harnessing AI for advances in scientific discovery, and the NAIRR for ensuring that the US research community can pursue big, bold ideas in AI that set the groundwork for the next generation of applications and capabilities. The task force set up by Congress to develop a roadmap for the NAIRR recommended that it be resourced at a level of $2.6 billion over six years. As of fiscal year 2026, it has received annual funding of about $30 million per year.
• Recommendation: Launch public-private partnerships to support research in critical, hardware-intensive domains such as robotics and manufacturing. In addition to the computational and data infrastructure enabled through initiatives such as the Genesis Mission and NAIRR, there is potential to accelerate progress in areas such as AI-enabled robotics and manufacturing, in which the United States is falling behind China. China has five times as many robots working in its factories as the United States, and more than the rest of the world combined. The United States should pursue public-private partnerships in which the private sector can provide researchers access to high-cost hardware infrastructure and robotics platforms, including networking and other testbeds, and the federal government can fund the research initiatives—working toward the shared end goal of innovation and advancement within critical sectors. Models such as the Platform for Advanced Wireless Research program, which is supported by NSF and an industry consortium of thirty companies and associations, provide a blueprint for impactful collaboration to accelerate breakthroughs in critical technology areas such as advanced wireless networking.
• Recommendation: Enhance efforts to expand broadband access. The federal government should accelerate efforts to connect all Americans to high-speed internet—such as through the Broadband Equity, Access, and Deployment (BEAD) Program—ensuring that connectivity requirements reflect the demands of AI workloads, particularly in terms of uplink capacity.

Finding: Signals and incentives from the government can accelerate AI adoption in key sectors. Integration of AI in key sectors of the economy holds the potential to strengthen the overall US economy and build trust among the population as people begin to see the benefits from the technology in their lives. The history of general-purpose technologies shows that the rate of diffusion is a key determinant of their capabilities, and the United States should fight to win the race to diffuse AI throughout society. The government can play an outsized role, particularly in certain sectors that rely on federal funding or involve significant federal regulatory oversight, but concerted actions will be needed.

• Recommendation: Implement federal innovation incentives in key sectors to encourage AI adoption. In sectors such as education, healthcare, or transportation, federal initiatives that encourage AI integration and modernization of legacy technical infrastructure through existing funding mechanisms hold the potential to drive the development of high-impact AI applications and accelerate tangible, positive impacts of AI on Americans’ everyday lives. Such efforts could be realized by leveraging tax credits, low-interest financing, or long-standing grant programs by building in opportunities and requirements related to AI innovation and infrastructure modernization. Federal funding can also support pilot programs to test AI applications in high-impact settings, developing best practices and guidance for integration that can facilitate scaling across industries.
• Recommendation: Advance consistent regulatory policy to facilitate responsible AI adoption. As discussed in more detail in Chapter 5, clear and consistent policies at the federal level around the use of AI in specific sectors can both provide clarity to companies in terms of compliance and accelerate responsible adoption of AI.

Finding: Safety, security, and assurance are all necessary ingredients for continued innovation and leadership. Without trust in AI systems—built through rigorous safety testing, robust security measures, and independent assurance mechanisms—public and institutional adoption could stall, effectively slowing the pace of innovation regardless of technical capability. Significant safety or security failures coming from AI systems risk setting the industry back by years. Steps should be taken to reduce this risk.

• Recommendation: Double down on international cooperation. International cooperation on AI safety and security can help develop and enforce shared safety and evaluation best practices and standards while building common awareness of emerging risks. The United States should lead in developing these AI safety and evaluation standards through NIST, engaging through international standards development organizations. Such cooperation becomes more urgent as AI capabilities rapidly evolve and transform the landscape in areas such as cybersecurity.
• Recommendation: Fund continued innovation in development of AI evaluation infrastructure, new benchmarks, and assurance methods. The US government should allocate AI R&D support to advancing the science of AI evaluation and developing new approaches to assuring AI model performance, which is particularly essential for high impact uses such as national security.

Tess deBlanc-Knowles is the senior director of the Atlantic Council’s Technology Programs.

Ryan Pan is a program assistant at the Atlantic Council’s GeoTech Center.

The GeoTech Center champions positive paths forward that societies can pursue to ensure new technologies and data empower people, prosperity, and peace.

Learn more

Image: Credit: Manuel via Unsplash