A Policy Blueprint for US Investment in AI Talent and Infrastructure | Andreessen Horowitz

This post was originally published on this site.

In Brief
America’s ability to compete globally in artificial intelligence depends on strategic investments in talent and infrastructure, particularly to ensure that Little Tech can innovate alongside bigger players. Building on emerging bipartisan consensus in Congress, and responding to China’s significant investments in AI capabilities, the United States should focus its investment in three core areas:

  1. Lowering barriers by creating a National AI Competitiveness Institute, expanding access to computational resources, establishing an “AI-Ready Data Initiative” for federally funded research, and ensuring affordable energy infrastructure through power grid expansion to support AI development;
  2. Training the workforce through modernized education pathways, specialized training programs, and updated apprenticeship models designed for AI careers; and
  3. Investing in AI innovation via strategic R&D funding, grand challenges across federal agencies, and procurement reform that enables startups to more easily sell AI solutions to the government.

America’s ability to compete in artificial intelligence is rooted in talent and infrastructure. A workforce must have the skills to build innovative AI products and to adapt to an AI-driven economy. To build competitive AI products, developers must have access to reliable AI infrastructure that can power their innovations. The stakes are high: the release of DeepSeek, an open-source model developed by a startup in China, underscores the importance of ensuring that Little Tech can compete to offer the most innovative AI products in the world. In a recent report, China stated that it would “strive to create an enabling environment for innovation that encourages exploration and tolerates failure.” US public policy must keep pace.

AI startups have unique talent and infrastructure needs because of the unique challenges of building AI products. Training and developing AI systems requires computational resources, specialized talent, and significant capital—requirements that can overwhelm early-stage companies before they even reach proof of concept, let alone become household names. Because of these barriers, startups face formidable challenges in competing with larger AI platforms with deeper pockets.

A public policy agenda rooted in American competitiveness can play a role in ensuring that Little Tech can compete in the United States and globally. As we have emphasized, AI competitiveness requires that policymakers regulate harmful uses of AI, rather than imposing burdens on model development itself, and that states and the federal government both have a role to play in AI policy, with Congress taking the lead in governing a national market in AI and states policing harmful conduct within their borders.

As we recently submitted to the White House Office of Science and Technology Policy, we propose an additional pillar of an American competitiveness agenda: the government should invest in AI talent and infrastructure to ensure that Little Tech ability to compete and thrive.

This investment includes three components:

  1. Lowering barriers
  2. Training the workforce
  3. Investing in AI innovation

This three-pillar approach aligns with the bipartisan consensus emerging in Congress. As highlighted in both the Bipartisan House Task Force Report on Artificial Intelligence and the Senate AI Working Group’s “Driving U.S. Innovation in Artificial Intelligence” roadmap, federal investment in AI infrastructure and talent development is critical for maintaining US leadership. Both bipartisan congressional reports recognize that America’s ability to compete in AI rests on lowering barriers through accessible computational resources, training a specialized workforce, and investing strategically in AI innovation—precisely the pillars that form the foundation of our policy recommendations.

Building on established tech policy tools, a16z recommends targeted interventions to address the unique computing, capital, and talent requirements of AI development for startups and established players alike. The goal isn’t simply to establish a level playing field—it’s to invest in America’s AI future to ensure that we realize the full spectrum of benefits that only a diverse, competitive AI ecosystem can provide.

The goal isn’t simply to establish a level playing field — it’s to invest in America’s AI future to ensure that we realize the full spectrum of benefits that only a diverse, competitive AI ecosystem can provide.



Lowering barriers

Recommendations at-a-glance

  • National AI Competitiveness Institute to expand compute and other resources
  • Unlock data access via AI-Ready Data Initiative and Open Data Commons
  • AI Hubs to provide regional access to high-performance computing
  • Compute vouchers & grants to support infrastructure needs
  • Power grid expansion to ensure affordable energy for AI development

Training and deploying AI models requires extraordinary computational resources and extensive datasets—requirements that can overwhelm even well-funded emerging companies. This fundamental challenge threatens to concentrate AI innovation in the hands of those who can afford the infrastructure, rather than those with the best ideas.

The federal government can directly expand access to essential AI resources by providing infrastructure and data that certain public and private entities could utilize. For instance, establishing a National AI Competitiveness Institute (NAICI), housed within the National Institute of Standards & Technology (NIST), could provide cloud-based computational resources to startups, researchers, and government agencies. It could also connect model developers to model benchmarking and evaluation resources, since early-stage model developers may be interested in performance feedback but may not have easy access to it. Creating affordable access to reliable evaluation providers could enable more startups to use this performance feedback to improve their products.

The NAICI could also play a role in improving access to public data. Despite government investment in research and data collection, much of this valuable content remains inaccessible to AI developers and researchers. Federal agencies from the National Aeronautics and Space Administration (NASA) to the National Institutes of Health (NIH) generate extensive datasets that could drive AI innovation, yet these resources often sit behind paywalls or in formats that aren’t easy to use and hinder effective utilization.

These barriers put American Little Tech at a disadvantage to AI model developers in China, which appears to train its models on a broader corpus of data. The government could play a critical role in addressing this imbalance, such as by creating an “Open Data Commons” of data pools that are managed in the public’s interest. By providing access to this open data for Little Tech, this approach could help to ensure startups can readily access the resources they need to compete.

Another path to expanding data access is to create an “AI-Ready Data Initiative” that could permit any NAICI participant to access any data or research produced by federally funded research. It could publish recommendations for standardized data formats for AI training, and require any recipient of federal funds to publish data in these formats. A technical, standard-setting organization like the NIST would be well positioned to play a lead role in managing this initiative. These open data sets could be housed within the NAICI.

These resources might not be exclusively cloud-based or headquartered in Washington, but could also include regional, on-premise options as well. Regional AI hubs could provide startups with access to high-performance computing without requiring massive capital investment. These facilities can be distributed strategically to support AI development across different regions of the country, ensuring innovation isn’t confined to traditional tech centers.

Beyond direct government provision, policy can expand access to commercial computing resources through targeted support programs. Direct financial support, including cloud compute vouchers and infrastructure acquisition grants, can help startups access commercial cloud services and build their own computing capabilities. These programs can be tailored to startup needs, with sliding scales based on company size and development stage.

Public-private partnerships offer another powerful tool for expanding access. Rather than directly mandating support for startups, policy could create compelling incentives for cooperation. This could include preferential procurement terms for cloud providers offering startup discounts, tax benefits tied to startup support programs, and making startup access a condition for participation in major government AI initiatives.

Finally, an AI-driven economy will require that Little Tech has affordable access to the energy it needs to build and run AI models. To ensure that America’s energy infrastructure can provide startups with the energy they need to be competitive with larger developers in the United States and with developers in China, the Council on Environmental Quality should issue regulations that allow for a substantial buildout of the US power grid. With countries like China, Japan, and Saudi Arabia racing to build out their own AI infrastructure, it is important that America’s energy investment–and public policy enabling that investment–keeps pace.

Training the workforce

Recommendations at-a-glance

  • Modernize AI education, including technical bootcamps & certifications
  • Reform the National Apprenticeship Act
  • Train workers for new jobs created in an AI-driven economy
  • Public AI awareness and literacy

Beyond infrastructure and procurement reform, America’s AI future depends on developing a robust talent pipeline while fostering broad public understanding of AI technology. Investing in workforce development and public education are critical to meeting the talent needs of emerging AI companies and to preparing society for a world where AI is critical to our economic and social lives.

The foundation of America’s AI capability rests on systematic development of technical talent across the full educational spectrum. Current educational pathways require significant modernization to meet evolving industry needs, particularly those of emerging AI companies. Federal initiatives, such as the bipartisan AI and Critical Technology Workforce Framework Act, illustrate one approach to structuring AI workforce development by aligning education and training programs with industry needs. Professional training programs, including technical bootcamps and certifications, need similar support to create multiple pathways into the AI workforce. In addition, federal investments should focus on STEM foundations in early education while expanding advanced AI specializations at the university level.

Beyond foundational education, advanced technical training represents a critical component of workforce development. Worker retraining programs, industry residency programs, and specialized government research positions must expand to address the growing gap between supply and demand for specialized AI talent. Lawmakers should modernize the 80-year-old National Apprenticeship Act, as the current system wasn’t designed for emerging fields like AI and computing. With 94% of apprenticeship participants finding employment upon completion, this model is an encouraging one, but it needs updating to create structured pathways into AI careers. Supporting infrastructure, including access to computing resources and digital learning platforms, must be integrated throughout all these programs.

New AI tools could also create new AI jobs. The government should announce a public-private “AI-Empowered Workforce” initiative to help Americans secure jobs in AI labeling. The government could commit to support workforce training programs to train the workforce in AI labeling, while partnering companies could commit to hire labelers who demonstrate proficiency.

Broader societal use of AI requires investment in public education and awareness. Basic AI literacy programs should extend beyond traditional educational settings to reach adult learners and working professionals. Consumer awareness initiatives need to focus on practical understanding of AI capabilities, benefits, and limitations, moving beyond both hype and unfounded fears.

Workforce digital literacy demands particular attention as AI tools become ubiquitous across industries. Training in basic AI tools and human-AI collaboration should become standard components of workforce development programs. Local community engagement programs can help ensure these initiatives reach beyond traditional tech-focused audiences.

Investing in AI innovation

Recommendations at-a-glance

  • Grand Challenges to fund AI breakthroughs
  • Support for academic AI research
  • Procurement reform to streamline government buying and contracting

The United States has a proven history of catalyzing innovation through strategic research and development (R&D) investments. From the Manhattan Project’s $2 billion effort ($30 billion in today’s dollars) that established our national lab system, to NASA’s Apollo Program that advanced computing and materials science, to DARPA’s creation of the internet, federal funding has consistently driven technological breakthroughs. These investments followed a common pattern: substantial public funding targeted ambitious goals that not only solved immediate problems but built knowledge infrastructure for private sector growth.

In addition to lowering infrastructure barriers and training American talent, public policy should maintain this tradition of strategic government investment through direct funding in AI-related research and development. With China demonstrating its commitment through substantial investments—including a five-year, $137 billion commitment by the Bank of China to safeguard China’s AI supply chain—the United States must respond both quickly and strategically. American R&D investment has historically catalyzed innovation across sectors. That investment is now critical.

For instance, federal grand challenges have historically proven effective at catalyzing breakthrough innovation in complex technical domains. The success of programs like DARPA’s autonomous vehicle challenges demonstrates how well-structured competitions can accelerate progress while expanding participation beyond traditional players. An AI Grand Challenges Initiative, integrated within NAICI, could apply this model to pressing AI development priorities such as developing more energy-efficient AI systems, creating more powerful foundation models, or advancing AI capabilities for scientific discovery.

This initiative would operate across key federal agencies including NSF, DARPA, DOE, and NIH, establishing specific technical benchmarks with funding rewards to stimulate competition across academic, startup, and established industry participants. Critical to its success would be coupling prize funding with NAICI compute resources and government datasets, ensuring that smaller teams can compete effectively without massive upfront capital investment. The program should extend beyond pure research competitions to include clear pathways to deployment, incorporating follow-on procurement opportunities that help winning teams bridge the valley between prototype and practical implementation.

The government could also support research at public universities that focuses on American competitiveness in AI. This approach aligns with the report of the Bipartisan House Task Force on Artificial Intelligence, which emphasized that “federal investments in fundamental research have enabled the current AI opportunity.” Potential topics for this type of academic research might include developing and evaluating worker retraining programs that are aimed at creating an AI-ready workforce, the role of open-source tools in fostering competition in AI markets and enhancing product security, and options for protecting against potential national security risks without undermining American competitiveness in AI, including how AI might be used as a tool to combat cybersecurity threats.

Investing in innovation also requires making it easier for the government to acquire innovative AI tools. The federal government’s potential as a catalyst for emerging AI companies remains largely untapped, constrained by procurement systems that may inadvertently favor large incumbents over innovative startups. Programs like DoD’s Tradewinds initiative—which streamlines AI procurement—demonstrate the possibility of more flexible approaches, and systematic reform is needed to transform government from a barrier to an enabler of startup-driven AI innovation.

Current federal procurement practices create multiple overlapping obstacles that disproportionately impact AI startups and small companies. Procedural requirements, designed with large defense contractors in mind, impose administrative burdens on lean startup teams that make it harder for them to offer competing bids. Security clearance processes can take months or years—timeframes that burn through startup runway—while past performance requirements create catch-22 situations where new companies cannot win contracts without experience but cannot gain experience without contracts. The classification of AI systems as non-commercial items triggers additional compliance obligations, creating particular challenges for startups developing novel applications without commercial analogues.

Technical barriers amplify these procedural challenges for small companies. Many government solicitations include requirements that effectively mandate the use of proprietary systems from large vendors, either explicitly or through legacy integration requirements. Data sharing restrictions, while often necessary for security, can prevent startups from using their existing development workflows and force costly process changes. Security certification processes, designed around traditional software development models, impose particularly heavy burdens on small teams trying to maintain their rapid development and deployment cycles.

Addressing these challenges requires both expanding flexible authorities and streamlining traditional procurement channels with a specific focus on reducing barriers for startups. The Other Transaction Authority (OTA) model, successfully employed by DoD for AI acquisition, should be expanded across civilian federal agencies with explicit provisions for small company participation and simplified technical evaluation processes. Commercial Solutions Openings (CSOs) offer another promising pathway, allowing agencies to solicit novel solutions from startups through dramatically simplified procedures. Both mechanisms should be paired with reforms to the Small Business Innovation Research (SBIR) and Small Business Technology Transfer (STTR) programs that better align with startup development timelines and capital requirements. In addition, a default commercial preference for software system procurement might help ensure that the federal government is able to integrate leading-edge AI tools into its operations.

Traditional procurement channels also need startup-friendly modernization. The Federal Acquisition Regulation’s commercial item determinations should be updated to better accommodate AI systems from new companies, recognizing that many startup AI applications represent novel approaches to existing commercial functions. Micro-purchase thresholds should be raised specifically for AI solutions from small businesses, enabling faster adoption of low-risk applications. Simplified acquisition procedures should be expanded for AI procurement from startups, with clear guidance on when and how agencies can leverage these streamlined processes.

These investments in American infrastructure and talent will help to ensure that our AI future is enriched by the contributions of Little Tech and will help to unleash the next wave of American AI innovation.