Billions invested. Billions billed back.
More and more companies are funding AI startups that quickly become their customers.
The firms providing chips, cloud services, and infrastructure are often the same ones investing in the businesses that use those services. This setup accelerates AI growth, but it also creates a tax issue that many companies may not yet fully realize.
The main issue is not just whether AI companies are overvalued. It is whether these close relationships start to look less like independent business deals and more like coordinated actions under U.S. transfer pricing rules.
For example, a big tech company invests in an AI startup, and that startup then buys major infrastructure services from its investor. Microsoft’s investment in OpenAI is one example. Amazon and Alphabet have similar deals with Anthropic. NVIDIA has invested in several AI companies that depend heavily on its chips.
At first glance, these deals seem normal.
Companies often invest in promising businesses, and suppliers want customers who are growing quickly. The problem arises when both relationships occur simultaneously and reinforce each other.
Under §482, transfer pricing rules apply to controlled transactions. Most people assume “control” means ownership. The actual regulations are broader. Control can include “any kind of control, direct or indirect,” including parties “acting in concert or with a common goal or purpose.”
This wording is important because these AI deals form ecosystems. One company might provide funding, infrastructure, computing power, and even strategic help all at once. The startup then relies on that support to run and grow.
The real driver here is the dependency, not control.
Tax authorities focus more on how companies actually behave than on official labels. If one company can influence another’s business decisions through funding, reliance on its infrastructure, or shared incentives, regulators might say there is sufficient control to require a transfer pricing review.
This doesn't mean these deals are incorrect; rather, it just means the responsibility now shifts.
When deals are seen as related-party transactions, companies may need to demonstrate that their prices are fair and comparable to what independent parties would agree on. This means keeping records, comparing to market data, and doing economic analysis. If they cannot back up their positions, they could face audits and long disputes.
In large companies, transfer pricing teams usually look at international deals first. Many of these AI deals are within the same country, so they might not appear right away in routine tax risk reviews.
That operational gap matters more than the technical rules. This gap in how companies operate can be more important than the technical rules themselves.
Commercial teams negotiate infrastructure contracts. Tax departments often see the full picture later, after the structure already exists.
AI makes this issue bigger because the industry is moving very quickly. Many of these deals happened close together in time. This concentration affects how regulators view them.
A single investment and supply deal might seem unimportant. But when this pattern repeats across the industry, it starts to look like a system.
As any transfer pricing expert will tell you, enforcement is heavily shaped by pattern recognition. Regulators look for recurring structures that create opportunities to shift income, manipulate deductions, or obscure economic relationships. The regulations themselves even state that a “presumption of control arises if income or deductions have been arbitrarily shifted.”
What matters most now is the real economic relationship, not just formal separation.
In the last ten years, regulators around the world have started to look more closely at how companies actually depend on each other, rather than just checking legal ownership. This trend shows up in transfer pricing, anti-avoidance rules, and related-party laws. Companies can be legally separate but still seen as working together economically.
The AI industry is especially open to this kind of analysis because companies rely heavily on infrastructure. Training advanced AI models requires significant computing power and specialized chips, which gives suppliers considerable leverage.
Older industries also faced tax questions about dependency. But in AI, these relationships are forming much faster and on a larger scale, in an industry already under political and regulatory scrutiny.
For tax professionals, the main takeaway is not to try to predict enforcement, but to understand how regulators think. Tax authorities will not look at deals one by one forever. Once an industry shows clear patterns, enforcement rules will develop around those patterns.
The AI industry has grown so much that its funding deals are no longer seen as just startup experiments; rather, they are considered standard business practices.
This shift changes the risk landscape.
When rules use broad terms like “control,” the facts and stories behind the deals matter most. Regulators and companies often argue more about real influence than about exact legal wording.
Business teams focus on growth, access, and strategy. Tax departments often end up with structures that were built quickly to stay competitive. By the time questions about “control” come up, these business dependencies may already be firmly in place.
This overall pattern is common. New industries usually start with flexible ideas about independence because growth is more important than structure. As the industry grows, regulators begin to check whether the relationships work as the documents claim.
AI might be a new technology, but the way institutions respond is not new.
The tax system always shifts its focus to the real substance of deals once enough money is involved in a repeated setup.