The Loop from the Chicago River, Wells Street bridge.
The AI boom and World War Two share similar responses in the sense of an immediate request of great quantities of advanced goods as well as a general trend of pivoting towards serving the new markets.
The second however ended in a crisis the US managed to escape, whereas today the signs of a crack are starting to appear, the question is if the States could be capable of escaping once more.
This article starts from the premises of the American industrial might of the Second World War and uses the results of a century of studies to explain why the AI uncertainty of the moment is nothing unforeseen and what issues are plaguing its development.
At the beginning of the 20th century the United States were amongst the world’s most powerful countries, however their defense industry paled, at least in theory, when confronted with the knowledge and tradition of its European Counterparts, boasting household names such as Krupp, Ansaldo or Vickers-Armstrongs.
The US’s greatest assets were, however, their modernity and conversion possibilities, as their modern and efficient plants, paired with the innovation brought by pioneers such as Sikorsky or the Lockheed’s legend Kelly Johnson, had enormous potential if successfully converted for the war effort.
During the First World War American Pilots flew on French planes, when the Second started their air force still fielded models inferior to Bf-109 and Spitfires, by the end of the war the United States had the most powerful air force in the world; the same can be said for their Navy, particularly in the Pacific.
The union between the two would bring forth the era of power projection through aircraft carriers, but, for that to happen, a man had to witness the greatness of military aviation first.
Returning from Europe at the end of the First World War, General Mitchell, commander of the Army Air Service in Pershing’s Army, became a strong advocate of the unification of the Navy’s and Army’s air forces to emulate the British model, which he deemed superior in functioning and effectiveness.
Even after both denied it initially, the navy did show particular interest in the application for anti-ship duties, which led to a series of Congress-required tests from 1920, which culminated between July 20/21st 1921, when his 1st Provisional Air Brigade sank the captured German battleship Ostfriesland.
Although the Navy accused General Mitchell of exceeding test parameters, and yet they had already started development of the first aircraft carrier in 1920, which resulted in the retrofit and recommissioning of USS Langley as a carrier on March 20th 1922.
Yet, the greatest demonstration for the US on just how devastating carriers could be, would come from a man from the Rising Sun, who visited the States, and used what he learned to create a doctrine centered around them, and, later on, use them against the very country he was sent to in order to learn more about modern superpowers.
Truly it was under the crystal clear skies of Oʻahu, when the mighty US Pacific Fleet bled under the bombs from the Kidō Butai in the bay of Pearl Harbor, that the USA, and the world, would grasp just how devastating these new weapons could be, as that day would be a date which will live in infamy.
Isoroku Yamamoto’s view on America was different from the one of other members of Japan’s high military command: the latter held onto a traditional military approach rooted in the longstanding military practices of Japanese imperial doctrine.
On the other end, Yamamoto studied at Harvard and had a different idea on how a potential confrontation with the USA was to be approached; as he witnessed and understood the industrial might of the States and firmly opposed a prolonged conflict, supporting instead the usage of a carrier fleet to execute rapid attacks to hopefully, overwhelm the country and force it into a peace deal before the “sleeping giant” could awake.
One of the main reasons the effort failed to reach its full potential was that the American carriers were not at port, and therefore, got spared and would come back to bite the IJN, and Yamamoto knew it.
Albeit the most celebrated, the admiral was not the only decision maker, and unluckily for him, the fierce competition between navy and army would force his forces in less than optimal positions, even then, Japan had one last chance to stop the Americans in time before the momentum could swing.
The American superiority in technology however would prove to be too much even before factories started putting out numbers: naval codebreakers were able to provide the American fleet with enough information to identify the next Japanese target in the Midway Atoll, and the ability of technicians to repair USS Yorktown in record time, would lead to the upset victory in the battle of Midway, changing the tide of the naval conflict in the Pacific.
If in 1942 the Japanese forces could count on superior machinery and experience, the same would not continue to be the case in the following years: Corsairs and Hellcats outclassed the Zeros and Hayabusas and the same was true for the great majority of military assets.
At the same time, sheer numbers alone became too much for the Japanese to oppose, as the American factories converted to full military production and outclassed the Nipponic counterparts, for instance the Willow Run Ford Motor Company plant produced one B-24 Liberator Bomber every 63 minutes in 1944, a benchmark impossible to even remotely match for any Axis power.
The last piece of the industrial masterclass performed by the US in the second world conflict was the ability to turn factories from civilian to military production in an extremely time and resource effective manner as, without this passage, all the innovation and design efforts would be vain, as proven by the Italian industrial complex during the same period of time. In fact, even without the many strategical mistakes of Italian high command, such as the decision to keep using biplanes when it was clear that aerial combat shifted from agility to raw power, the country was unable to produce numbers on par with expectations, rendering a once pioneering nation dependent on German blueprints to field somewhat competitive fighter aircraft.
The successful Italian hydroplane racers of the thirties boosted powerful engines and aerodynamic lines that would not be seen until the concluding stages of the country’s involvement in the war, whereas the British took theirs and turned it into the Supermarine Spitfire. At the same time Italy’s record-breaking Breda 88 twin motor plane failed to be successfully converted to military use while the Americans built the P-38 into a dominant heavy fighter and interceptor from scratch.
The comparison underlines the stark difference in conversion capacity: a country’s ability to retain power and control lies in its ability to adapt to the demands of developing situations.
Not long after the end of the conflict, another set of ideas and theories, already partially pioneered during the war, would become increasingly interesting to the military first, and corporations later, as in the 1950’s the progress made on mechanical automation and computing would receive increasing funding and be renamed under the broader concept of Artificial Intelligence.
The reason why the US so heavily invested in aeronautical production and carrier doctrines is that strategists individuated the possibility to project power beyond visual range. Artificial intelligence answers a similar demand: it extends a system’s ability to act beyond direct human interaction, allowing operations to be monitored, coordinated and executed on a new level in terms of scale and speed.
AI also shares similar limitations to big carrier groups: albeit formidable tools, their development, construction, deployment, maintenance and operational costs renders them inefficient for smaller scale operations and budgets; a fact many companies that bought into the initial hype are discovering, still, capital follows the trend of expected profitability.
In 2026 Amazon, Microsoft, Alphabet and Meta are expected to record a combined capital expenditure of about $700-725 billion, a projected increase of around 75% from 2025, a year when investment was already high, drawing an ideological similarity with the steep increase in planes production in the US from 1943 to 1944.
It is important to denote how adoption of AI systems is more common than conversion, as around 9 in 10 organizations use AI systems, with only 44% scaling it across the enterprise and around 37% reporting a contribution to EBIT.
In 1942 the conversion of American industry was imposed by the state as on January 1st the Office of Production Management froze all car sales, which effectively halted on the 22nd of February, with Ford going from more than 600,000 cars in 1941 to zero in 1943, while continuing truck production and pivoting the rest for military production of planes, parts, etc.
Today a similar conversion is being recorded in the tech industry, although in this case the reason is not dictated by a government’s order but rather by market demand; two case studies are famously Nvidia and Micron.
The first is a prime example on how AI can impact the earnings of a company to the point of changing its priorities completely: during fiscal 2023 the company recorded $27 billion in revenue, of which $15b coming from data center products, by fiscal 2026 the revenue figure reached $215.9 billion, of which $193.7 billion, almost 90% of which came from the latter category of products.
The other main provider of revenue for the company is gaming, which increased from $9.1 billion to $16.0 billion a mere 76% increase when compared to an almost thirteenfold multiplication of data center products.
On December 3, 2025, Micron announced the decision to exit their 29 years long commitment to produce memory and storage production as the Crucial consumer business, effectively terminating consumer shipments by February 2026.
As explained by the company itself, the decision to exit was dictated by the AI-driven growth in data center demand, which led to excluding retail customers in order to: “improve supply and support for our larger, strategic customers in faster-growing segments” in a move similar to the closure of Ford’s car lines in 1942 when a profitable civilian product was closed so that the same factories could produce for a different buyer, albeit in 1942 the driver was a government order on national security grounds, whereas in Micron’s case it is profitability.
These are not isolated cases as, according to IDC, other famous companies such as Samsung and SK Hynix have all shifted production, similarly, towards products such as high-bandwidth memory and high-capacity server modules, distancing themselves from more conventional memory used in phones and computers.
An unpleasant consequence of this shift for the general consumer is a supply growth of 16% for DRAM and 17% for NAND, both below historical norms, which in moderate scenarios translates to a fall in PC shipments of 4.9%, with prices rising by an estimated 4% to 6% for 2026.
A similar situation unravels for smartphone shipments, which fall by 2.9% and registers a price increase of an average between 3% to 5%, this situation is particularly tedious for political, social and moral reasons as the brunt of the hit is absorbed by the retail market (individuals rather than the corporations themselves), which is the first of two major issues with this trend.
If the Japanese navy was constrained by production numbers, what hit the US was the complete opposite: with an economy fully converted to wartime production, once the conflict ended and the main purchasing power as government spending plummeted from 44.0% of GNP in 1944 to 8.9 percent by 1948, which prompted two recessions, a first one in 1945 and a second one in the 1948 to 49, the reason that the country managed to quickly bounce back and enter a thriving economic state was because people had money to spend and the reconversion of factories for civilian production was successful, however today the same might not be possible.
Many companies in tech today face a similar concentration weak spot even if AI is not their main source of revenue; it is a common misconception in fact that an AI crisis would significantly impact the finances of AI specialized companies alone.
For instance, corporations such as Meta and Alphabet, record the majority of revenue from other activities, namely advertisement and cloud services, and yet they spend more on AI infrastructure than their operations generate.
Analyzing Meta’s 2026 Q3, expenses for capital expenditure amounted to around $34.91 billion, which compared to $32.79 billion of cash from operations, results in a free cash flow of negative $2.12 billion, with long-term debt reaching $80.92 billion, a figure close to their $82.45 billion of cash and marketable securities.
Regarding Alphabet’s capital expenditure that same quarter: the total amounts to $51.4 billion, more than double the previous year’s total, however the company’s free cash flow was negative $9.1 billion, with a confirmed spending of around $200 billion for 2026 and an expected increase to $300 billion in 2027.
Due to these circumstances, markets have revalued risk regarding these companies: on October 29th 2026, the day after reporting results and confirming spending plans, Alphabet saw its shares drop nearly 13% as the entire sector deals with doubt and dissatisfaction, which is their worst day on record and losing around $500b in valuation.
What is also interesting is that, at the same time, Cloud services recorded an increase to almost $30.7 billion in the quarter, more than doubling, which indicates that the market’s approach to AI risk is strong enough to upset positive results even from the core business of a corporation, as investors do not price results of sales, but rather the uncertainty on the scale of the commitment and returns on the AI bet.
The exposure is not limited to the companies’ shareholders as the 10 largest companies, in the S&P 500 accounted for around 39% of the index’s market value as of October 2026, which is well above the 27 percent reached at the peak of the .com bubble, what is most relevant for this case is that all of them have significant investments in AI.
Most of those companies are the same ones leading AI spending. Anyone holding a broad index fund, including most pension savers, therefore carries part of the same concentrated position. As at Midway, the risk is not that the asset fails to work, but that too much depends on it at once.
After Pearl Harbor Yamamoto was not fully satisfied as he knew that, albeit a great result, a greater liability had survived and continued to pose a possible threat; today corporations look at their books and, albeit great successes, they cannot ignore the lingering liability posed by their AI investments, the final question is: what can they do to avoid a financial Midway?
If the AI investment circle ends in significant losses, companies could be open to legal action being carried forward by shareholders or government entities, history however proves that corporations would be, most likely, shielded under the business judgment rule.
An instance of this concept and the way it could unravel can be seen in the lawsuit brought forward by Citigroup shareholders against the bank’s directors for failing to monitor the risks behind its subprime-related losses.
The Court of Chancery of the state of Delaware, where most US corporations are incorporated, dismissed most of the claims as, according to the interpretation, the directors did not act in bad faith but rather “failing to recognize the extent of a Company’s business risk” which in itself does not create a liability.
At the same time the structure of most of these companies renders smaller investors almost irrelevant in the decision-making process, as for example, Alphabet’s Larry Page and Sergey Brin together held 52.7% as of April 2026, and a similar situation is rather recurrent and even stronger considering that most funds hold Alphabet’s class C shares or equivalents, which carry no voting power at all.
Regarding the moral question, the theme is much wider, and it mainly centers around the imbalance of power, as a decision made by the few that hold control of these corporations has effects that retail investors and the markets in general bear, which leads to pension savers and index funds holders carrying a significant risk in the AI bet without necessarily having a say on it, while other consumers that may not be directly involved in these financial instruments still suffer from excessive prices of everyday technological items.
The situation is also worsened by the public reputation of AI companies and AI in general, which has taken a turn for the worse as many people started to see them more negatively, oftentimes to an excessive degree.
This has led to the topics being brought under heavy political scrutiny, and oftentimes weaponized as well. Which, on one hand, can help bring attention and increase awareness on the topic and its different weaknesses, such as cost, security, etc. The side effect of being weaponized for political and social purposes is that it has led to misinformation and sensationalization of the topic, and the resulting general approach will be less rational and more focused on satisfying ideas and expectations, rather than achieving the best financial outcome.
It is exceedingly clear, at least up to now, that AI is a true asset that holds great potential, both already explored and untapped, it is, however, equally plainly evident that the initial hype around this has led to a less than optimal allocation of capital and resources, this is not, however, the first time that this has happened, and it will certainly not be the last.
It is also foreseeable that the investments and general stance will not be, at least in the short term, vastly modified to better align with the economically optimal approach, and the reason for this traces back to human behavior applied to finance.
In the Psychology of Sunk Cost by Arkes and Blumer, 1985, there is a scenario in which an airline spends $10 million in funding on a radar-blank plane, while a competitor launches a better one before the funding and development is over. In this experiment, 85% of the subjects chose to keep funding the project, whereas in case the company did not invest yet, only 17% would continue down the same path, this correlates both to the carriers’ example at the beginning of this analysis and AI.
In fact, countries such as Italy remained on projects because they had already started to fund them, even after they revealed themselves to be obsolete, and fleets preferred not to pivot toward more advanced carrier-based strategies earlier because of the long-standing tradition and unwillingness to break it.
This also explains the current situation with AI funding: it is clear that there is a mismatch that leads to many instances of excessive capital allocation, however, it is unlikely that companies would cut back on these positions because the most foreseeable result would be a loss in valuation and trust from the markets, which makes admitting the mistake even more costly than continued spending hoping for a big breakthrough that would justify the expense.
Furthermore, as explained by John Maynard Keynes in the General Theory in 1936, professional investors are usually judged against their peers, therefore, they are inclined to follow the crowd rather than risk with a different approach, as he puts it, “worldly wisdom teaches that it is better for reputation to fail conventionally than to succeed unconventionally”. This explains why it is hard for one big corporation to be the first one to give up on AI, if not completely, at least to redimension their investment, as this would possibly be seen by the market as admitting defeat in the race if in the future there will be a great advancement that leads to increased profitability.
This is reinforced by Scharfstein and Stein’s “Herd Behavior and Investment”, from the American Economic Review in 1990, which expands the argument to managers as well, where they found that the latter tend to follow the decisions taken by others in the same field rather than propose a different approach, as its failure would be imputed to them single-handedly, whereas a broader sector failure tends not to have single culprits.
There is no better way to sum up what has previously been said than to use a quote from Chuck Prince, CEO of Citigroup in 2007, when he told, just before the financial crisis, to the Financial Times: “When the music stops, in terms of liquidity, things will be complicated. But as long as the music is playing, you’ve got to get up and dance. We’re still dancing”, as this is the perfect example of an executive seeing the risk, knowing the consequences for a full meltdown, and still deciding to dance.
In the 1990s, a great interest in the telecommunication sector created a situation in many ways comparable to the one of AI today.
What is interesting to note, more than the timeline and more than the precise numbers, is that at the end of the crisis, when the companies failed or got heavily hit, their technology and infrastructure still stood and are still used, to this day.
The connection between a possible AI bubble and a communication bubble was made by Jeff Bezos at Italian Tech Week in Turin, on 3 October 2025, where he pointed out that the nature of said bubbles would be industrial and not financial, as even if the companies suffer, society still gets great benefits, so far, this has not entirely held true, as the repercussions of a crisis in the market have fallen on retail investors and the general public.
But what is undeniable is that whereas telecommunication companies suffered and many closed, the infrastructure is still widely used and benefits society, the same way, most likely, that a crisis in the AI sector wouldn’t result in the complete failure of AI as a field, but rather would terminate in a financial hit for companies, and still huge benefits for society in general.
The analysis of US production before, during, and after the war shows how an economy can convert to accommodate the concrete and ideological needs of a new market, history also shows us how dangerous it is when the socioeconomic factors do not allow for a quick bounce back if a crisis hits a significantly changed market; and AI today is as disruptive as any other major shift since the Industrial Revolution.
What has been stressed, especially in the last part of this article, is perhaps the most important and ignored fact: in a properly Schrödingerian twist, the very observation by outside investors and their opinion ends up to be the defining factor in internal decisions taken by companies investing in AI.
Decision makers, in fact, know that after an initial excessive excitement for this new, very real field, investors and the public will punish them regardless of the choice they make, because every one of those carries an issue people tend not to ignore: doubling down on investment is dangerous for capital allocation valuation and market risk and risk pricing.
However, pulling out of the commitment will cause valuations to go down regardless, because of failed expectations, therefore, lower and upper management in big companies who decided to invest in AI are stuck in a loop, and the result is that market efficiency and proper analysis are inevitably stained, which creates additional complications to an already tedious situation.
Ultimately, this research aims to show how the situation is much more complicated than many make it out to be, and the ultimate solution should not be to excessively punish companies for their positions, but rather to allow them to recalibrate them without adding further distorting factors to an already very complicated trajectory.