An Investment in AI Is Not Free From Risk
Artificial intelligence (AI) is in the headlines every day, and for good reason. AI promises to revolutionize the economy and supercharge productivity. It may very well drive breakthroughs in healthcare, finance, and education in ways that are hard to predict but appear to be hugely beneficial. AI powered automation may reshape the workforce and could streamline supply chains and resource allocations. All of that is fantastic, but how can an investor benefit, and what risks might come with an allocation to this transformative new technology? With this blog, we’ll review 10 risks you may not have considered before.
Market Hype and Overvaluation
Many people reading this will remember the great bull market of the late 90s when anything associated with the internet was hot. We’ve seen this phenomenon many times before, with the automobile, the airplane, biotech, personal computers, smartphones, and many other groundbreaking technologies.
The dot-com bubble provides a cautionary tale. CMGI was an internet incubator and venture capital firm that invested in internet startups, including some hot names like Lycos and AltaVista. In 1999, its stock price skyrocketed over 900%, reaching a market capitalization of $40 billion despite little to no profitability. When the bubble burst in 2000, CMGI’s stock collapsed over 99%, falling from $163 to less than $1 by 2002.
In these early days, many AI companies, particularly startups, may be highly overvalued due to market enthusiasm, similar to the dot-com bubble. Companies like OpenAI, Anthropic, and Hugging Face are raising billions of dollars of capital without clear profitability models.
Warren Buffett wisely said, “In the short run, the market is a voting machine, but in the long run it is a weighing machine.” What the market weighs, above all, is profits. AI stocks may experience sharp corrections if earnings fail to meet expectations.
AI Regulation and Government Oversight
AI is in its “wild west” days, but they could end in a sea of government interventions. Already, governments are moving toward regulating AI, especially in areas like privacy, bias, and intellectual property. The EU AI Act and U.S. executive orders are early examples of government regulation over AI.
This is a huge topic that we are just beginning to comprehend. We won’t even try here to put a price on these issues, but any one of them could create significant friction for AI companies in the form of significant costs.
For example, courts and legislators are debating whether AI-generated content infringes copyrights. Artists, writers, and software developers have filed lawsuits against OpenAI and Stability AI for using copyrighted data without permission. It’s not hard to imagine stricter data protection laws that aim to limit AI companies from collecting and using data without consent.
Governments are pushing for AI fairness, transparency, and accountability to prevent discrimination. The EU AI Act classifies high-risk AI systems and imposes stricter oversight on AI used in employment, banking, and public services. All of these regulations are likely to affect end-user costs, perhaps significantly, which will dampen demand.
Governments are proposing clear liability rules for AI developers, users, and businesses that deploy AI. The EU AI Act introduces strict liability for AI-related damages, shifting responsibility to AI developers, and that will have to be priced into applications and subscriptions.
Governments are requiring AI developers to disclose how models make so-called “high-stakes decisions” in areas like finance, healthcare, and criminal justice. When the stakes go up, so does the cost of contingent liability, which will make the ultimate products more expensive, and potentially significantly so.
There are more concerns, including content verification, watermarking, and criminal penalties for AI-enabled fraud. There is also the possibility of export control, as we’ve seen with banning sales of NVIDIA chips to China, and the question of AI taxes, collected to provide UBI (universal basic income) to workers displaced by AI. All of these things fall under the category of “no free lunch.”
High Infrastructure Costs
AI models require expensive graphics processing units, or GPUs, cloud computing, and energy to train and operate. OpenAI spent over $500 million in 2022 developing ChatGPT. Training GPT-4 costs over $100 million, and AI server costs are straining cloud providers like Amazon, Microsoft, and Google.
Data centers for AI require massive power and cooling infrastructure, adding to costs.
Meta, Microsoft, and Google are building AI-specific data centers costing $10 billion or more each. Obviously, these high costs put pressure either on the developer, the user, or both. There are very real concerns regarding the electrical grid, and AI will have to compete with electrified vehicles for power.
AI Model Hallucinations and Reliability
“Garbage in, garbage out” is an old term in computer technology, dating back to the 50s, but it’s still relevant to AI today. There is a real risk that AI systems could generate false or misleading outputs, leading to legal and reputational risks. Google’s Gemini AI recently faced backlash for inaccurate and biased outputs.
Job Displacement and Public Backlash
AI automation could replace entire categories of jobs, leading to political and social resistance. We’ve seen this before: when machines do the work of skilled workers, real problems, including social unrest, can arise.
The Luddites were a group of British textile workers who protested the introduction of mechanized looms and knitting frames in the early 19th century. Companies used machines to replace skilled workers with cheaper, less skilled workers. Luddites broke into factories and destroyed weaving frames, and they attacked employers, magistrates, and food merchants. Who can tell where all that sort of thing will end?
Dependence on a Few Hardware Suppliers
AI infrastructure depends heavily on a handful of chipmakers, NVIDIA, AMD, Intel, which can create supply chain vulnerabilities. We’ve seen this already, as international AI development has been hampered by the U.S. ban on AI chip exports to China.
China seems to have stockpiled enough NVIDIA chips to create DeepSeek, but at this writing, we don’t know very much about any of that, and, of course, transparency is an issue with the Chinese Communist Party. What we can say is that geopolitics can lead to hardware shortages, price volatility, and heightened tensions. The world is in a high-stakes game.
Competition and Technological Disruption
AI innovation is evolving so fast that today’s leaders may become obsolete within months. New releases and better models can make older models obsolete almost overnight. This can be a great thing for consumers but can be the kiss of death for shareholders. Startups may encounter scale barriers, while competitors can deliver a knockout blow to leaders with little warning. And, of course, deep-pocket firms can acquire hot new companies to keep themselves on top. Alphabet, Microsoft and Cisco Systems together have acquired almost 500 AI startups in the past several years.
Ethical and Bias Issues
One would expect AI to add value to credit scoring, hiring, and law enforcement, but that could lead to claims of bias and big fights over proprietary algorithms. If an AI model or firm is found, or even suspected, to have bias, companies may face legal risks and brand challenges. Given the litigious nature of our society, this seems almost inevitable.
The Environment and Climate Change
An AI data center can consume as much energy as a small city. This can lead to questions about environmental, social, and governance issues (ESG), which in turn can lead to higher costs. This will be an advantage for an AI platform that is energy efficient, but AI as an industry could find itself with branding issues similar to those faced by nuclear power.
Will AI Ever Make Money
Many AI companies offer free or low-cost AI services without clear revenue models. ChatGPT, Bard, and Claude offer “free tiers,” which obviously raises concerns about profitability. Some AI businesses may burn cash without a viable revenue stream, and weighing machines tend not to like them very much.
AI is one of the most expensive technologies ever invented. AI infrastructure runs to billions in chips, cloud computing, energy, and data center buildouts. AI is an energy hog; running AI models is just as costly as training them. Because of these tremendous barriers to entry, the largest companies have a massive advantage, but even they face very real challenges.
Like the internet and other industrial innovations before it, AI will likely create many more losers than winners, even as it revolutionizes everyday life. Investors should think clearly about whether they are investing in hype or reality because, eventually, the weighing machine will punish firms that don’t produce earnings and value ones that do.
