During the 19th century, the Industrial Revolution was facilitated by a new form of technology: railroads. By 1890, an estimated $10 billion ($347 billion in today’s value) had been raised for the railroad industry in public stock offerings and private financings. We are living through a similar speculative bubble in the development of the latest transformative technology: artificial intelligence (AI).
Railways made it possible to ship high volumes of goods overland across previously treacherous routes and long distances at a high rate of speed. Railroad companies became the single-largest target industry for speculative investment in the mid-century, attracting capital from individual investors, banks, other investment companies, and governments.
Some of these startup companies were fraudulent and never laid a single mile of track. Others were legitimate but overexpanded and took on too much debt. Without adequate demand and too many competitors, they struggled to make a return on their investment. By 1890, approximately half of all the companies that funded more than 50 years of railroad mania had gone bankrupt.
The Panic of 1893 would, in one year alone, take out many more, representing 25 percent of U.S. railroad companies. Bankruptcies included major rail lines such as Philadelphia and Reading, Erie, Northern Pacific, and Union Pacific. The industry was saved by a banking consortium led by financier J.P. Morgan, which forced consolidation and created higher-priced monopolies over the regions.
We are currently experiencing a speculative bubble in the development of AI. While specific numbers are difficult to pinpoint, the cumulative investment in AI already easily totals hundreds of billions of dollars. Companies involved in AI represented the vast majority of earnings and market value gains in 2024. Investors have piled into anything related to the novel technology.
The “Magnificent Seven” tech companies grew their earnings by 33 percent in 2024. Excluding these seven largest AI-related tech companies, the remaining 493 companies in the S&P 500 grew their earnings by only 4 percent. In the fourth quarter alone, the market value gains of the Magnificent Seven represented more than 100 percent of the S&P 500’s returns, implying the rest lost market value.
President Donald Trump recently announced plans for a private sector initiative called Stargate, which represents an investment commitment of $500 billion in AI infrastructure. Nvidia, the leading provider of AI-specific semiconductors, an essential component of AI technology, has become the third-largest company in the world by market capitalization, with approximately $3.3 trillion in equity value at the end of 2024.
On Jan. 27, Nvidia lost $600 billion in its equity value in a single day—the largest stock market value loss ever recorded by one company. Other tech companies associated with AI also experienced severe market value losses. The Philadelphia Semiconductor Index, a proxy for U.S.-based AI investment, dropped by more than 9 percent, the largest single-day decline since the COVID-19 panic of March 2020. Including Nvidia, total market value losses topped $1 trillion.
The rout was the result of news of a Chinese startup called DeepSeek. This company introduced an advanced AI tool that claims to operate at a small fraction of the cost and data processing power of U.S.-based AI competitors. DeepSeek quickly vaulted to the top of AI tool downloads from online app stores, beating out incumbents such as U.S.-based ChatGPT.
U.S. investors fear that DeepSeek will completely undermine the investment thesis for AI and challenge the dominance of U.S. companies such as Nvidia. Like the miles of redundant rail lines that crisscrossed each other in a competitive land grab in the 19th century, tens if not hundreds of billions of dollars of investment in data centers and other infrastructure may be stranded by lower-cost and more efficient competitors.
Only time will tell whether the threat is real. DeepSeek may prove little more than a well-orchestrated “deepfake.”
DeepSeek’s capabilities are only beginning to be tested and compared with competing U.S. models. Observers noted that the timing of the news—coming on the heels of the U.S. announcement of Stargate—seemed suspiciously convenient and intended to take the air out of the market. Similar to controversies over Chinese telecom company Huawei and social media company TikTok—both ultimately controlled by elements of the Chinese Communist Party—concerns have been raised about Beijing’s involvement in DeepSeek and whether it might be a Trojan horse to data gathering, surveillance, and government censorship.
DeepSeek’s AI tool refuses to answer questions about the CCP, Chinese leader Xi Jinping, or Taiwan, providing an early indication of its informational bias. These and other issues imply that DeepSeek may ultimately run into substantial regulatory hurdles to its deployment and utilization in the United States and Europe.
It is without question that the advent of AI has the potential to create enormous value and dramatically enhance productivity. The question is whether that power will be concentrated in the hands of a few oligopolistic monoliths, as has often been the case with other technological revolutions, or whether this time might be different.
DeepSeek is probably not the answer, but it is worthwhile to find a lower-cost, less data-intensive, more distributed, and decentralized model that could potentially provide a more democratic technological future. Blockchain technologies, such as those that developers in the Ethereum community are seeking to deploy, may provide part of the solution.
Like all tech-enabled speculative bubbles—railroads, automobiles, radio, television, computers, or the Internet—there will be a few winners and many losers. Picking the unicorns and rejecting the mules is a difficult task. The blind speculation on the Magnificent Seven and the knee-jerk fear of DeepSeek annihilating its U.S. competitors are likely to be equal overreactions.