By Penelope Henderson

It is no secret that AI is evolving on an exponential scale, taking over not only the business world, but our social spheres. However, there is an immense amount of public overconfidence and expectation invested in AI, which will be dissected, explained, and debunked in this piece through a business and geopolitical lens. 

Fundamentally, the life cycle of AI development can be split into six key areas: 

1. Design: The initial phase of defining specific production goals.
2. Production: The technical creation of the AI model, gathering data, and ensuring reliable future performance.
3. Supply: The acquisition of hardware components (specifically semiconductors and chips).
4. Infrastructure: The physical power plants and data centers needed to facilitate combining the functions of the hardware and software. 
5. Large Language Models (LLMs): The most commonly used AIs, trained off of large data sets to produce efficient generative AI models.
6. Applications: The most publicly competitive layer. This is where AI models are integrated into services a user can access and build off of. 

Globally, China and The United States have the most LLMs, making them the most competitive and also the fastest growing. The primary differentiator between the two tech giants’ capabilities is where the state bases production of these six elements. Due to U.S. imposed sanctions, China does not have the same access to chips and infrastructure as the United States. The United States government has made it difficult for major chip suppliers, like Nvidia, to supply chips to China, forcing China to rely on domestic production. These restrictions produce an unfortunate, unintended consequence for the United States; a China that is training LLMs more efficiently and effectively due to their limited access to supply and infrastructure. 

Similarly to how China is looking to shift away from buying chips from Nvidia due to its risks and restrictions, the United States is looking to make a domestic production shift. Currently, The United States relies primarily on Taiwan Semiconductor Manufacturing Company (TSMC) to produce their chips for tech giants like Apple and Nvidia. In 2022, Congress passed the CHIPS and Science Act, helping to incentivize domestic production of semiconductors. With $52 billion subsidies and tax credits to boost domestic semiconductor research, this bill aimed to strengthen US supply chains, create jobs, and counter China’s tech dominance. The Peterson Institute for International Economics details a massive increase in private investment, incentivizing investment at an estimated investment in the private sector between $235 billion and $500 billion from tech giants like Intel and Samsung. As of January 2026, CNBC reports a $250 Billion investment in U.S. chip technology, with the U.S. lowering its base tariff on Taiwanese goods from 20% to 15%. This reflects a projected structural shift in the global semiconductor economy. The United States aims to produce 20% of semiconductors by the end of the decade, as their global share of chip production dropped from 37% in 1990 to 10% in 2024 (Semiconductor Industry Association).

This shift away from Taiwan is slow but necessary. Remaining reliant on Taiwan puts the United States in a dangerous position should Chinese and Taiwanese tensions escalate.  If the U.S. can no longer support the physical infrastructure, they can no longer meet market expectations, something that they are already struggling to meet. Companies are investing billions into AI to maximize company capability and productivity, in turn supporting this domestic shift. Many of these companies however are not seeing high returns; this is the root of the term, “AI bubble.” Companies invest in AI expecting that the program’s future abilities will produce enough demand to make their money back. AI’s biggest investors are their own consumers. Nvidia is investing in Microsoft and Microsoft in Nvidia, producing an incestuous and dangerous closed feedback loop, masking the fact that companies are not seeing returns, therefore boosting this over confidence in the market. Apple reports that AI reasoning capability is not as powerful as commonly assumed. The “thinking” feature is perceived as a fact check, serving as an example of a factor in consumer over confidence as improvements are getting smaller and more expensive. 

It was reported in August of 2025, 95% of the 52 companies that have invested in over 300 generative AI initiatives have not seen their returns. A sizable portion of these investments are not going immediately to large language models as many assume, but a bulk of them are going to physical capital like data centers. But these data centers will all be used, so these companies will surely see their returns, right? Wrong. As developers reach the physical ceiling of processing power in regards to how many transmitters can fit on a chip. To put this simply, we need something more powerful than a chip. We need quantum. 

Quantum computing sounds foreign, but it is something real and rapidly approaching. Rather than using traditional binary, quantum computers use qubits, these quantum bits can be 0, 1, or both at once. Qubits can also be linked so that the state of one instantly affects the other, no matter the distance. This allows for massive parallel processing and simultaneous analysis. Quantum Machine Learning (QML) can find complex patterns in tiny amounts of data that classical AI would simply see as “noise” as they have been trained on almost the entire internet. 

As we see a shift from physical chips to quantum computing, the risk of wasted investment emerges. If these data centers are not fully utilized and the significant capital deployed that does not deliver returns, the debt cannot be paid back. Wasted investments paired with slow returns runs the risk of the “bubble burst.” If this should happen, all companies hinging on the application layer of LLM’s will go bankrupt, companies like Nvidia and Microsoft’s stock prices will drop significantly, causing a ripple effect.

It is not that these companies will never see their returns, it is the fear of what will happen if not seen fast enough at the rate that obstacles and competition emerge.

Sources: Semiconductor Industry Association, 2025 State of the U.S. Semiconductor Industry Report; Lauren Feiner, CNBC, January 15, 2026; Nick Lachance, Fortune, August 18, 2025; Mirzadeh et al., Apple Machine Learning Research, 2024.