Post
The United States of AI Dominance
Two things became abundantly clear this week:
1) The United States economy is finding another gear.
2) Generative AI and domestic technology companies are the strongest growth engine in the world, and stand to further supercharge economic decoupling.
Several economic indicators came in that further bolstered the case for a stronger US economy than many had anticipated at the start of this year: labor market remains tight (Feb unemployment below 4% for 25th month in a row), headline and core inflation rates fell to 2.4% and 2.8% (first “2 handles” since March 2021), and regional business surveys rebounded from January towards 50. The MSCI ACWI index is up 0.59% YTD vs. -1% for the MSCI ACWI ex USA index, and the S&P 500 is up 7.5%. Additionally, the latest data shows that US GDP has grown 7.4% since Q4’19 vs. 4.7% for the G7, while China is in the midst of a prolonged economic slowdown. This is not meant as a market forecast as much as the latest example of the ability of the US’ dynamic economy to surprise to the upside. Consensus about Fed policy has shifted dramatically as a result, with odds implied by futures pricing showing a 5% probability of a rate cut at the March Fed meeting (it was 38% at the start of February). So far it seems there is no immediate need for monetary easing. The economy continues to pull ahead of others, with or without rate cuts, and we expect AI to accelerate this. Like many others, we have our concerns regarding near-term hype and product capabilities. But the more founders we meet building at the edge of frontier technology, the stronger our conviction grows that the US is clearly best positioned to benefit from its tailwinds. An interesting bit of color on adoption – Klarna recently announced that one month after the launch of its AI assistant, the chatbot is doing the work of 700 support agents, has reduced repeat inquiries by 25%, and can resolve certain requests in less than 2 minutes vs. 11 minutes by a human agent. Additionally, Dell is up ~30% after reporting earnings on Thursday night, and highlighted that its backlog of AI servers (powered by H100 Nvidia chips), has reached $2.9B vs. $1.6B in Q3. This bull market is showing we are in the early days of AI adoption.
IPO activity demonstrates the resilience of the US economy as well: last year the number of IPOs on the LSE fell to 23 from 74 in the year prior, while total IPO proceeds from Europe, Middle East, and EMEIA regions have fallen by nearly 40% as well. In the Americas, IPO proceeds grew 155% last year and there were 132 IPOs on US exchanges (total IPOs in the Americas grew 15% YoY). The NYSE’s vice chair, John Tuttle, recently commented on this trend:
“No matter how you look at the data, the United States is the deepest pool of liquidity and capital in the world, which has the broadest investor base,” he said at the World Economic Forum (WEF) in Davos. “It has a lot of analysts and investors that are focused on growth, not just dividends and value.”
US equities represent 70% of the MSCI world index by weight vs. ~50% in 2000. EMEA represents 19% vs. 34% in 2000, and APAC represents 10% vs. 13% in 2000. And if we go back to the 1980s, many will remember when Japan was over 40% of the index (it now accounts for ~6%). The US of course has its share of problems (debt-to-GDP is 123%, the labor force participation rate is still below pre-covid levels, etc.) and ~40% S&P 500 revenue comes from abroad. But when we think about which economy is best positioned to absorb a potentially enormous innovation cycle driven by emergent AI capabilities, there is little doubt in our mind that the US stands above the rest. It was only a few years ago (”B.CGPT” – before Chat GPT times) that many were declaring China well ahead in the AI race. Obviously perceptions changed in 2022, and China is now on its back foot:
Even as the country races to build generative A.I., Chinese companies are relying almost entirely on underlying systems from the United States. China now lags the United States in generative A.I. by at least a year and may be falling further behind, according to more than a dozen tech industry insiders and leading engineers, setting the stage for a new phase in the cutthroat technological competition between the two nations that some have likened to a cold war.
The EU, as usual, seems determined to lay down as many regulatory speedbumps as possible to ensure its private sector remains at a technological disadvantage. On February 13, several Parliament Committees voted to adopt the EU’s proposed AI act, which means the final vote by the full European Parliament will occur in April this year. We believe the legislation’s protective stance will inadvertently hamstring the very innovation it seeks to nurture. The Act’s risk-based categorization of AI systems imposes a heavy yoke on developers and enterprises. And the stringent criteria set for high-risk applications threatens to curb the developmental trajectory of these technologies as it imposes a disproportionate burden on the shoulders of innovators. At its core, the Act employs a risk-based approach, categorizing AI systems into different levels of risk from minimal to high, and tailoring regulatory requirements accordingly. It sets out stringent compliance measures for high-risk AI applications, including transparency obligations, data governance, and accuracy standards. It also outlines specific prohibitions on certain uses of AI that it deems to pose unacceptable risks. To oversee and enforce these regulations, the Act proposes the establishment of the European Artificial Intelligence Board, tasked with ensuring a unified regulatory landscape across the EU. One of the most concerning aspects of this are the requirements imposed on open source systems:
The one-size-fits-all approach of the EU AI Act, which demands comprehensive control over the development process, creates significant and impractical barriers for providers of open-source foundation models. A key question arises: who is responsible for maintaining ten years of documentation after a foundation model is deployed… especially when the model is a product of decentralized, open-source collaboration? Consider BLOOM, an open multilingual language model developed by a consortium of 1,000 AI researchers, primarily academics. Under the current stipulations, these researchers would be obligated to maintain extensive documentation for their model.
While the EU and China hamstring their own AI ecosystems, the US continues to build up a robust early and growth-stage one as the government tests out its own regulatory framework. There are reasons to be concerned about this as well, but it’s still very early days. At the end of January, the Biden administrated announced new actions stemming from its AI executive order across areas such as reporting, federal risk assessments, and new research initiatives. For example, they are going to “compel developers of the most powerful AI systems to report vital information, especially AI safety tests, to the Department of Commerce.” We will be watching this closely but are optimistic that the regulatory environment will remain considerably more founder friendly than any other country.
A related topic worth touching on here is decoupling. US decoupling from China (and ROW) and reshoring to Mexico has continued. The US trade deficit last year fell to its lowest in a decade (-$280B). There are some misleading aspects of this data, for example inputs from China are flowing through trade from Vietnam and Mexico, but at the very least we know that there are concerted efforts to diversify critical supply chains. We expect decoupling to continue at a slow pace, in part due to efforts by China to evade tariffs, in line with how supply chains have evolved historically:
Historically, supply chains move gradually. Often, just one step or component goes offshore before an ecosystem of suppliers develops. Over time, the Chinese component in U.S. imports from third countries seems destined to drop… “Greenfield foreign direct investment into developing countries has remained constant, but the share that’s going to countries that are not China and not Russia has gone way, way up,” said Olivia White, one of the authors of the McKinsey report. “That’s consistent with that investment helping those countries’ capacity to do more and more.”
We mention this because taken together, US economic strength, a (so far) more careful approach to AI regulation, and continued supply chain diversification all makes for a setup that places the US at the center of AI innovation. Still, the tailwind risks are serious. Ken Griffin summarized this well in the MFA Network Miami Conference this past January: “If we lost access to Taiwanese semiconductors, how many weeks until Tesla stops making cars, or GM or Ford…those chips are used in every part of our economy. Estimates range from a GDP hit of 8-10% if we lost access to Taiwanese semiconductors.”
This risk is not going away anytime soon. But in the meantime, the US economic engine continues to defy expectations, AI capabilities are improving each week, and US founders are rapidly iterating on new products and infrastructure to usher in the next technology super cycle. One of our core theses is that increasing adoption and integration of AI into products and services will necessitate robust, scalable infrastructure to support the entire lifecycle of AI and ML initiatives — from conception and development to deployment and ongoing management. The corporate announcements mentioned above (Klarna and Dell) are just a preview of the initiatives that many others are currently exploring. If the data continues to show clear productivity gains from these initial experiments, and related demand growth for the infrastructure required to conduct them, we anticipate a significant acceleration across sectors. Our conviction here has only grown as we continue meeting with founders building this next generation of critical tooling.