From tulip mania to today’s AI boom, markets still chase hype. Investors weigh opportunity against looming bubble warnings, as Poland accelerates AI adoption despite uncertainty over valuation, readiness, and sustainability. by Sean Reynaud
Since at least 1637, during the Dutch tulip mania, stock market bubbles have been a thing. If not tulips, then speculation on stocks, as in 1929, or the dot-com craze of the late 90s, or subprime mortgages in 2008. Whatever the root cause of bubbles, they remain part of the risk, because stock trades involve human beings and human beings are, to put it kindly, driven by emotion. As John Maynard Keynes put it, “The markets are moved by animal spirits, and not by reason.” Investors invest when the vibe is right, or as some are saying about AI, because of FOMO. The promise of AI stocks has fed their need for green in a way no other stock has, and also raised fears of another bubble.
But suppose the AI bubble doom sayers are correct, and we hope they are not. In that case, the Nasdaq may be looking at a $40 trillion wipe-out of market capitalization. The catalyst for this disaster, according to analysts at Goldman Sachs, is a weakening of profit margins, increased volatility in stock markets, increased tensions in financing large AI infrastructure projects and/or rapid growth of debt-financed investments. Or it could be a minor correction.
Investors are already spending significant sums of money on AI and AI data centers. In Poland, as of 7 November this year, DL Invest Group and America’s Boosteroid formed a joint venture to build a network of data centers in Europe. DL Invest Group, led by Dominik Leszczyński, chairman of the management board and founder, currently manages a portfolio of more than a billion euros, comprising over 400 active tenants from the industrial, office and commercial sectors. This one group is reasonably big, and an example of how real estate is invested in Artificial Intelligence (AI) as well.
Investments today are flowing primarily into Large Language Models (LLMs), algorithms that simulate human intelligence. LLM’s AI label has triggered choruses of “Shut up and take my money!” from investors. But LLMs are not AI in the way we think of Data from Star Trek, or HAL 9000 from 2001: A Space Odyssey. No, our current generative AI is a different sort of algorithm than the image classification and object detection tech, Cognitive Neural Networks (CNNs), that have been around for a while, and much different than artificial general intelligence, which is capable of human-level cognitive function. LLMs and deep learning are new, they are not self-aware, they do not have goals, and they cannot make decisions in the way a human does. They are good at averaging and drawing from extensive (huge even) databases. Its mathematical calculations are sometimes so mysterious that we refer to the mystery of their decision-making as the “black box problem”.
Are LLMs the way to GenAI? According to Richard Sutton, the 2024 Turing Award winner, LLMs will not lead to artificial general intelligence (human-level intelligence) and are a “dead end”. Or, as Neil deGrasse Tyson put it, Generative AI is “overrated.”
As reported in WBJ in our last issue, in our interview with Arek Hajduk, the climate of startup investment since AI has been, “If there’s no AI in a company, then something is wrong.” Many startups have since promised AI in one form or another. This “cart before the horse” mentality is such that executives demand engineers, “Find a problem that AI can solve, and do it fast!” Engineers, not wanting to tell the boss there is no use case for AI, do whatever it takes to show results, sometimes even fabricating them. The hype is real, dangerous, and may have created a house of cards.
The giddy reporting of 2022 has given way in some corners to stress and doubt, and to an overinflation of the value of a misunderstood technology. The euphoria over OpenAI’s ChatGPT not only affects the demand for AI applications but also for chips, and has created a noticeable decline in human employment.
Type “AI Bubble” in any search engine, or ask your favorite LLM, and you will likely see a flood of stories about the incoming crisis. USA Today stated that the last time the price-to-earnings ratio (CAPE) spiked – standing at 39.65 in October of 2025 for the S&P 500 – was during the dot-com bubble of the 90s and the 1929 crash. Fed Chair Jerome Powell seems to think stock prices “are fairly highly valued.” The Guardian reported in November of this year that global stock markets were falling, with the “magnificent seven” (Nvidia, Amazon, Apple, Microsoft, Tesla, Alphabet, and Meta) AI-related stocks declining by 2% in early November. Palantir, an earlier AI adopter, has become the target of Michael Burry, a speculator who saw the decline in the U.S. real estate market in 2008.
In the five stages of an economic bubble (Displacement, Boom, Euphoria, Profit-taking and Panic), we are heading towards the panic phase in some very specific areas of AI development. However, according to analysts at Goldman Sachs, investments in AI currently account for a smaller proportion than telecommunications investments did at the peak of the dot-com bubble. Growth has been shorter and narrower in the AI tech sector than in the dot-com bubble. AI data centers are increasingly funded through debt, requiring external financing.
ChatGPT’s Sam Altman has grown tired of all the AI bubble talk, but when your biggest supporter, Microsoft, is accused of obscuring “OpenAI’s $11.5 billion loss last quarter,” as Windows Central reports, perhaps Mr. Altman can understand investor concern. OpenAI’s transition to a hybrid profit/non-profit, with Microsoft holding a 27% stake in the company, may help. Vast data centers, specialized chips, and elite global engineering teams, not to mention the power to run the data centers and water to cool them, have cost OpenAI a lot.
A 2024 report by the Startup Poland Foundation found that one-third of Polish startups stated that their product or service was based on or supported by AI. According to Statista, the AI market in Poland is projected to reach USD 1.83 billion in 2025. The annual growth rate (CAGR 2025-2031) is 37.60%, resulting in a market volume of USD 12.43 bln by 2031, compared to the United States’ USD 46.99 billion in 2025. Certainly, there is a lot invested in Poland in AI.
According to trade.gov.pl, Poland is a leader in the implementation of Gen AI, and a June 2025 Capgemini report found that 63% of respondents actively use Gen AI tools. Sixty-two percent of Polish firms have increased AI spending, and 18% have reached full implementation – another 32% report partial AI use. Polish organizations achieve strong returns, averaging 1.7 times ROI, with many expecting results within 1-3 years. At the same time, challenges remain, particularly employee readiness and training.
Contrarily, a recent study by Webcon found that only eight percent of Polish managers believed their companies were ready to use AI. Webcon explained that half of the Polish respondents were not provided with training and guidelines for using AI. This survey suggests that Polish managers are uninterested in AI, though some employees use it on their own.
The healthcare industry in Poland utilizes LLMs for disease diagnosis, drug discovery, personalized treatment plans, AI-powered customer service tools, chatbots and virtual assistants, according to Statista. Polish medtech CancerCenter uses AI in diagnostic laboratories through digitalization and AI-driven tools, with an integrated Laboratory Information System (LIS) platform that streamlines pathology workflows and automates repetitive tasks. This reduces the administrative burden and is used in 20 laboratories across Poland, Switzerland, and Georgia.
Outside the medical industry, Poland’s Eleven Labs reached a valuation of USD one billion, achieving unicorn status, by using AI to produce voices with accents for audiobooks. AI Clearing used AI to enhance quality control across construction projects, including solar farms and roads. AI has also proven useful in creating ESG reports to fulfill the EU’s Corporate Sustainability Reporting Directive (CSRD), which obliges around 50,000 companies to disclose detailed ESG data.
GenAI tech may have sound applications, but there is clear evidence of speculation and flaky overpromising that will need to be pruned from the promise of AI. Just as the dot-com bubble required a readjustment, so too does the tech industry’s GenAI need a financial readjustment to address its overpromises. Valuations, hiring and infrastructure investment are out of scale with revenue reality. AI startups without revenue, the “we’ll monetize later” groups, are at very high risk. AI chip/GPU suppliers are experiencing speculative pricing amidst insane demand. AI consultancies and service integrators are at high risk of selling the “AI transformation” faster than organizations can adopt. Text/image/video AI apps are at risk because they have low barriers to entry, many clones, and few real revenue models. According to Forbes, OpenAI is blowing through $15 million a day on Sora 2, a video-generation AI. It’s only a matter of time before the correction comes, but it’s anyone’s guess as to when.
Europe and Poland have no homegrown AI companies. If and when the shock comes, Poland faces a double whammy: equity contraction followed by further reliance on foreign service providers. As one of the top 10 countries targeted by hackers, Poland needs to develop its own homegrown models, which is why it is leading the Baltic AI Gigafactory, a €3 billion infrastructure project aimed at developing and deploying advanced AI models across the region.
