Research consumes scarce resources in exchange for uncertain knowledge. You commit time and effort to a promising scientific question, often with no upfront clarity on the potential discovery's utility and impact. What motivates such a choice? For an individual, curiosity and determination may be enough. However, here I'm more concerned with the institutional incentives that drive the research behavior of corporate and academic labs. Because whether the laboratory belongs to a university or a corporation, somebody must decide which questions deserve attention, somebody must pay, and somebody must judge the results. The recent AI boom, mixing both euphoric hype and fear of the technology, has been instrumental in amplifying the different mechanisms of corporate and academic research, enough so that the differences become evident.

Private funding. Corporate labs are predominantly financed through company revenues and private investment. When a company invests in R&D, the funds come from current or future savings. This implies that the firm gives up alternative uses such as improving operations, hiring more workers, or distributing those funds to shareholders. Naturally, the research effort aims to increase the company's earnings and profits in the long term. A project need not produce an immediate return, but its funding must ultimately be justified through some expected commercial advantage: a better product, lower costs, or a capability that competitors lack. These constraints make the research entrepreneurial and guided by calculated risks. And since in capitalism the way to make money is by providing the consumer with ever more useful and beneficial goods and services, the profit-motive of a corporate lab correlates well with maximizing consumer welfare.

State funding. In contrast, academic labs commonly depend on public funding allocated through university budgets and research grants. The funds are not saved from voluntary transactions, but instead come from involuntary taxes. Specifically, the state designates a budget for research and public institutions compete for it. Evidently, this creates institutions which depend on the state to survive and gradually become increasingly focused on satisfying the state, rather than the average consumer. There are no shareholders expecting dividends, and no hard requirement for the research to make an accounting profit. Instead, we are usually told this spending is in the country's best interest, or for strategic positioning, or for national prestige.

AI excitement. The world is currently in an AI frenzy. The hype is enormous and has flooded into both research and entrepreneurship. Every time I go to a conference I feel it more strongly. As an example, in 2022 CVPR received ~8K submissions. In 2026 that number was >16K. NeurIPS 2026 received >40K submissions. ICLR 2027 received a staggering >60K paper registrations. This incredible number is due to a surge in PhD students hungry to make a name for themselves, and a surge in the need to write papers as a main form of validation for their output. The peer-review system is overwhelmed by this monumental volume. It is estimated there are from 18.5K to 33K AI startups currently with VC funding having increased 31 times in the span of ten years.

Peer review. The decrease in quality of the peer-review system has been evident for some time. What happens is that when you submit a paper it is increasingly likely that despite the paper-reviewer matching, reviewers do not sufficiently understand the details of your work and cannot provide a good review. Why? Because as the field becomes more complex, the line between a good and a bad contribution is blurring. AI research is becoming more open-ended. It is hard to scope a contribution to be compact and without too much spillover into other topics and areas, which makes it hard to evaluate. One ends up wondering whether it is the training, architecture, data size, task domain, execution, or something else that will move the needle in the end.

Here's another way to understand what is happening with peer review. We can imagine contributed papers and reviewers as vectors in a high-dimensional space. For a paper, the vector represents the latent semantics of the contribution. For a reviewer, it describes the person's expertise - what the reviewer is familiar with. Most contributions that have no major errors are epsilon-small seemingly random modifications around the main ideas that work. In a high-dimensional space two random vectors are with high probability nearly orthogonal, so reviewers cannot properly assess the contribution and score it with "borderline", "weak accept" or "weak reject". The final outcome becomes more sensitive to any one reviewer's rating and more random in terms of the papers accepted. It is still unclear how to solve this: with AI reviews which scale in principle, or with a market-based mechanism.

Feedback and correction. AI is not a bubble. But we are very likely reacting to it as if it were: we are overshooting in terms of the quantity of capital, time, and effort that we allocate to it. Currently, expectations lead to funding, which leads to results, which create bigger expectations, which lead to more funding and so on. Soon it becomes a feedback loop that grows and grows. Eventually reality will kick in, triggering a correction. There will be a crunch in which projects will be abandoned, budgets will be revised, people will be laid off. All of this is a normal and healthy process after which capital will be reallocated to better uses. Unless a structural factor gets in the way...

State-funded AI research. The big problem of state-funded research is that there is no direct market mechanism to evaluate your contribution. The goal of a PhD student is to publish papers. After sufficient toil, every paper eventually gets accepted at a conference. The authors celebrate, post on X and LinkedIn, create a project website, advertise, and present their work at the conference. Yet, it's still unclear whether this project will survive in a free market where people have to spend actual scarce money on it. Even if you top the leaderboards, how do you know whether to put the project in the profit or loss column at the end of the day? How do we know that the best use of a year's worth of a PhD student's salary, funded through our taxes, is yet another Gaussian splatting, or VLA, or transformer architecture paper? Approximately 90% of startups fail. What is the percentage for papers?

Academia has no market mechanism to guide research. There is simply no profit-and-loss pricing for your output. In such an environment, state funding very likely diverts scarce resources from one use-case to another without demonstrating that the benefits justify the opportunity cost. Promotions depend on publication count and citations, so what do you expect other than more papers? A university's strongest claim is that it can sustain a "valuable" kind of work. But that claim deserves to be evaluated separately from whether taxpayers should be compelled to finance it.

Free market competition. On the other hand, consider what is going on in the tech world. Our lives have changed in the span of a few years. The focus on engineering and productization has turned research into tools people use to write code, search for information, and automate tedious work. Competition gives companies a reason to make these tools cheaper, faster, and more reliable: when users can switch, every shortcoming creates an opportunity for a rival. The current race between big US tech companies, and broadly between countries, is a blessing for consumers. It has decreased AI prices faster than any other transformative technology in history. Likewise, it has pushed quality higher, with expert-level intelligence coming from the device in your hand. Even with a few big companies like Google, OpenAI, and Anthropic, the free market does what it does best - it coordinates resources with needs, driving an optimization process beyond that which any central planner could ever fathom. We are witnessing something extraordinary.

Do you see what the free market can achieve? It is private companies like Waymo that are deploying autonomous vehicles with accident rates below comparable human-driver benchmarks. It is also private companies that are building the agents to solve the crowning mathematical problems of human thought. And it will be private companies to bring forth the next wave of advancements in robotics, space sciences, biochemistry, energy, or anything else. A state-funded academic institution does not have the scale and incentives for these things. Its existence distorts the reward landscape. It almost seems like these institutions underestimate the benefits of an efficient economy.

Freedom to invest. Some research projects demand more capital, time, and risk than a small firm can bear. That alone does not establish a need for state funding. Venture capital and other forms of private investment allow investors to pool their savings, share the risk, and finance work whose returns may lie years ahead. Investors can be wrong, but the prospect of losing their capital gives them a concrete reason to scrutinize costs and expected returns. A country seeking to succeed in this technological race should not introduce barriers to entry under the pretense of safety. Instead, it should protect competition, completely eliminate regulatory burdens, and let those who create value retain the rewards. Let him who has earned the palm bear it.