The private sector drives progress in artificial intelligence. National governments were once the prime movers behind strategic technologies, from networked systems to nuclear energy, and supported foundational work on AI techniques. But today, governments mostly rely on private companies to build their AI software, furnish their AI talent, and produce the AI advances that underpin economic and military competitiveness.

This shift brings risks and opportunities for the United States. America could reap massive security benefits from private sector AI innovation in the coming decades. Policymakers may be able to extend these benefits even further by developing policies that boost American AI companies’ economic prospects and guide them toward work supporting national security and public interests. Yet at the same time, other countries could harness their own companies to similar ends—or even exploit American private-sector strength by co-opting, subverting, or stealing from U.S. firms leading in AI innovation today.

Policymakers have many tools to mobilize American AI companies and protect their long-term edge in a competitive global marketplace, from R&D subsidies and public-private partnerships to defensive measures such as investment screening, sanctions, and export controls. To achieve the intended outcomes and avoid unwanted distortions and side effects in the market, policymakers should understand where commercial AI activity takes place, who funds it and carries it out, which real-world problems AI companies are trying to solve, and how these facets are changing over time.

This paper explores these issues by analyzing equity investment into privately held AI companies, defined here as for-profit businesses (including state-owned or affiliated enterprises) focused on AI and not traded on a stock exchange. (We sometimes use the term “private-market” to describe this investment.) Using general purpose investment data from leading sources along with our own analytic tools, we find:

1. As of the end of 2019, the United States had the world’s largest investment market in privately held AI companies. (moderate to high confidence)

• In 2019, privately held AI companies attracted nearly $40 billion in disclosed equity investment—defined as venture capital, private equity, and mergers and acquisitions—across more than 3,100 discrete transactions.
• U.S. companies attracted most of this investment: $25.2 billion in disclosed value (64 percent of the global total) across 1,412 transactions.
• Based on estimates of transactions without publicly disclosed values, the U.S. market and the overall global AI market could be twice as large as public data indicates.

2. China’s market faded in the last two years, while investment elsewhere grew. (moderate to high confidence)

• Consistent with broader market trends and data from other sources, we assess that China’s AI market roughly quintupled between 2015 and 2017 (as measured by disclosed transaction value), then fell back to near-2015 levels.
• U.S.-based AI companies account for a steadily shrinking percentage of global transactions, but remain ahead in transaction value.
• AI investment in Western Europe, Israel, India, Japan, and Singapore is growing quickly by all metrics.

3. While active in AI both at home and abroad, Chinese investors are minor players in markets outside China. (moderate confidence)

• Seven percent of the transactions in our dataset involved at least one disclosed Chinese investor (whether alone or together with additional Chinese or non-Chinese investors). These transactions typically involved Chinese targets.
• In 2019, disclosed Chinese investors participated in only 2 percent of investments into U.S. AI companies, down from a peak of 5 percent in 2016.

4. Mergers and acquisitions (M&A) activity accounts for a significant share of AI investment outside China. (moderate confidence)

• Outside China, total M&A value may have exceeded venture capital value from 2015 to 2019, based on estimates of M&A transactions with undisclosed values.
• Including M&A transactions significantly reduces Chinese companies’ share of the market. Chinese restrictions on foreign investment may play a role.

5. Most privately held AI companies focus on transportation, business services, or general purpose applications. However, Chinese AI companies may be more likely to focus on certain applications. (moderate confidence)

• Compared to the United States and the rest of the world, investment into Chinese AI companies is concentrated in transportation, security and biometrics, and arts and leisure.
• China’s active industrial policy, and the United States’ greater reliance on the private sector, may help explain these differences.

6. National security applications attract little direct private-market investment. (high confidence)

• While many AI technologies might be adapted for government use, just a tiny percentage of all AI companies receiving investment make products designed specifically for government and military use.
• However, security and biometrics applications such as facial recognition, which have obvious governmental uses, account for a larger share of private-market investment in China than elsewhere.

7. In the aggregate, when they invest outside China, Chinese investors do not seem to disproportionately invest in different AI applications from non-Chinese investors. (low to moderate confidence)

• Our data do not indicate that Chinese equity investors disproportionately seek out defense-relevant AI companies when investing outside China.
• Although some China-based investors clearly invest abroad to extract security-sensitive information or technology, our data suggest they are probably a relatively small piece of a larger and more diverse AI investment market.

For U.S. policymakers, these findings are cause for some optimism. America’s AI startups and other privately held AI companies lead the world in attracting equity investment. Little evidence suggests that America’s closest AI competitor, China, is narrowing the overall gap according to this metric, and Chinese investors don’t seem to be co-opting privately held U.S. AI companies in large numbers through the equity investment marketplace.

At the same time, the findings point to significant challenges for the United States. America’s technological leadership is often taken for granted, but the United States has no monopoly on commercial AI activity. Other countries collectively account for a large and, by some metrics, growing share of the investment measured in this paper. And although the U.S. AI sector is booming, few of American AI companies examined focus on national security or other governmental priorities. In some priority areas, such as transportation and security, America’s lead in investment over China, its closest competitor, shrinks or even disappears. Finally, with respect to technology transfer, the aggregate trends explored in this paper can mask troubling transactions and developments at the level of individual companies and technologies. Finding these needles in the haystack of the broader market, and addressing them without unduly disrupting that market, will present a regulatory problem in the years to come.

This report explains our findings in detail, presenting the methodological choices, the assumptions shaping them, and the numbers supporting them. Our findings are subject to two basic caveats. First, they are not comprehensive. We measure only one aspect of commercial AI activity: equity investment flows into AI companies that are not publicly traded. While our approach provides meaningful insight into AI innovation and growth in the commercial sector, the numbers in this paper are not meant to measure all such activity. In addition, our dataset ends in late 2019 and does not cover more recent shifts in AI investment—including shifts related to the COVID-19 pandemic, certain to reshape AI investment in coming years.

Second, our numerical calculations are estimates. Defining AI investments and AI companies is inherently subjective, and the “AI hype” phenomenon increases uncertainty. Moreover, implementing these definitions always entails some error, and even the best available investment datasets have gaps. Our analysis of Chinese investment patterns also involves simplifying assumptions. Most importantly, we only count publicly disclosed Chinese investors and generally assume organizational investors have the nationality of the countries where they are headquartered, which could lead us to underestimate Chinese investors’ activity to some degree. Despite these unavoidable uncertainties, our basic findings (particularly those described as “high confidence”) would be unlikely to change under a range of alternative approaches or assumptions.