Bern, 1905. A young patent clerk named Albert Einstein published four papers that transformed physics. More than a century later, scientific breakthroughs still tend to be associated with exceptional people and years of painstaking work.

Artificial intelligence (AI) may change the pace of that work. Models are beginning to help prove mathematical results, spot patterns in biology that scientists had not noticed, and choose which experiments to run next. The prospect raises an understandable question: are we about to witness a new era of discovery?

We believe the answer depends on more than the models themselves. Turning an AI-generated idea into a useful medicine or material requires laboratories, reliable data, and evidence that the result works in the real world. For investors, the companies building that infrastructure may offer opportunities well before a headline announces the next historic breakthrough.

AI is moving beyond answering questions

Some recent results show why the excitement is growing. OpenAI reported that its AI system had produced a proof addressing the Navier–Stokes Millennium Prize problem, a famous challenge involving the mathematics of fluid motion. The proof concerns a fluid model with an applied force and was checked in Lean, software that verifies mathematical reasoning step by step.¹

Biology offers a different kind of example. Anthropic reported that roughly 950 Claude agents searched DNA databases over 21 hours and identified a previously unrecognized biological system called ART. The system includes repeating DNA sequences reminiscent of CRISPR, the bacterial defense system that became a gene-editing tool.²

That comparison needs care. Researchers have not demonstrated that ART can edit genes or function like CRISPR. The interesting result is that AI helped uncover something worth investigating, then pointed scientists toward experiments that could establish what it does.

These examples suggest AI can contribute to original research. They also reveal a crucial difference between fields: checking a mathematical proof can be a software task. Understanding a biological discovery requires experiments.

Discovery still needs a laboratory

An AI model can suggest thousands of promising molecules. Scientists still need to make them, measure their behavior, and determine whether they are useful. A convincing explanation on a screen cannot establish that a drug works in a patient.

This is why we are watching laboratories that connect AI directly to experiments. The model proposes a test, laboratory equipment runs it, and the results guide the next test. Each cycle gives the model evidence it could not obtain from existing publications alone.

OpenAI and Ginkgo Bioworks demonstrated this approach by connecting GPT-5 to an automated laboratory. Across six rounds, they tested more than 36,000 recipes for producing proteins without living cells, using the results to improve subsequent proposals. The study involved one protein and one production system, so broader applications still need testing.³

Lila Sciences has applied a similar process to materials research. The company reported screening nearly 3,000 catalysts, materials that help chemical reactions occur, and identifying six promising groups for producing hydrogen from water. The work remains research rather than a commercial product, and people still handled parts of the experimental process.⁴

The appeal is straightforward. A model that learns from measurements can improve its recommendations as it works. But the speed of progress depends on the experiments: a test that takes weeks cannot become instantaneous simply because the AI thinks faster.

Three places investment value may emerge

We see three connected areas where better scientific infrastructure could create value.

Models of cells and tissues. Scientists are developing "virtual cells," computer models designed to predict how cells respond to changes such as a new drug. Better predictions could help researchers decide which treatments deserve expensive testing.

The challenge is obtaining data that reflects human biology. Cells grown in a laboratory offer useful information, but they do not capture everything happening inside a patient's tumor. San Francisco-based private biotechnology company Noetik, for example, trains models on human tumor samples and studies how cancer cells interact with surrounding cells. Its goal includes helping drugmakers identify promising targets and patients who may respond to treatment.⁵

The investment question is who can keep producing useful, hard-to-replace data and turn it into a product customers will pay for. A static collection may lose its advantage as competitors catch up. A business that continually connects new measurements to better predictions may be more durable. Noetik's licensing collaboration with GSK, a British multinational pharmaceutical and biotechnology company, provides an early example of how biological models can become a commercial service.⁶

Laboratories that learn from their results. Traditional automation follows instructions set by people. A laboratory that also uses results to choose its next experiments takes automation further, helping researchers decide what to investigate as well as how to test it.

Washington D.C.-based healthcare company Danaher plans to bring an AI-powered laboratory into operation at scale in early 2027. The lab is designed to create and test antibodies and related research tools, then use those results to improve subsequent designs.⁷

The commercial question remains open. Some companies may sell access to laboratories or AI models; others may use the technology to improve their own products. Equipment and consumables suppliers could also benefit if researchers run more experiments. Scientific progress alone does not tell us which approach will generate attractive returns.

Data from real patients. Eventually, research must explain what happens in people. That requires connecting a patient's condition, treatment, and test results with outcomes over time, including side effects and whether the disease returns.

Chicago-based health technology company Tempus AI, an AGIX holding as of October 31, 2026, illustrates this opportunity.⁸ The company combines clinical and molecular data with analytical tools. GSK is a known customer of Tempus AI, and Danaher subsidiary Leica Biosystems has partnered with the company.^9,10 In the second quarter of 2026, Tempus AI's Data and Applications business generated $93.2 million in revenue, up 28% from a year earlier. Tempus also reported delivering its first AI model for cancer research to AstraZeneca.¹¹

We believe the potential advantage lies in continually adding useful patient information and applying it to decisions that customers value. Access to patient data, the right to use it for research, and the ability to turn it into reliable products all matter. Simply holding a large dataset does not ensure a successful business.

These areas can reinforce one another. Patient samples can improve models of cells; models can suggest experiments; experimental findings can improve predictions about patients. Building those connections is an important part of the opportunity, even as the business models continue to develop.

The opportunity may unfold over years

There are good reasons to be optimistic about AI in science. There are also limits that more computing power cannot immediately remove. A promising laboratory result still needs to survive testing in more complex biological systems and, ultimately, in people. A patient's response to treatment may take months or years to measure.

That makes predictions of when AI will "cure cancer" premature. Progress is more likely to emerge through specific improvements: a better research tool, a more useful prediction, or a treatment that succeeds in clinical trials.

Genome sequencing offers a useful precedent. The estimated cost of producing a high-quality draft sequence of a human genome fell from about $14 million in 2006 to below $1,500 in late 2015.¹² The lasting opportunity developed as cheaper sequencing supported research, diagnostics, and new approaches to treatment.

For investors, recurring needs may matter as much as a breakthrough. A laboratory needs equipment and supplies each time it runs an experiment, while researchers need fresh measurements to improve their models. That could support businesses with repeat customers and ongoing demand. The commercial test is whether these tools save time, reduce costs, or improve decisions enough to justify what they charge.

For the KraneShares Public-Private AI & Technology ETF (AGIX), this supports looking across the AI ecosystem, including the infrastructure that turns model capabilities into useful applications. In science, we believe that means watching how companies acquire better data, improve experiments, and earn revenue from the resulting tools.

Investors worried about missing a historic discovery may find opportunities while these systems are being built. The companies helping researchers establish which ideas actually work deserve attention long before the biggest medical breakthroughs become clear.

---

Holdings are subject to change.

For AGIX standard performance, top 10 holdings, risks, and other Fund information, please click here.

Citation:

1. OpenAI. On the Navier–Stokes Millennium Prize Problem. September 8, 2026.
2. Anthropic. Claude discovers a novel enzyme system with CRISPR-like repeats. September 23, 2026.
3. OpenAI. GPT-5 lowers the cost of cell-free protein synthesis. February 5, 2026.
4. Lila Sciences. How an AI-run lab cracked open green hydrogen's catalyst problem. September 25, 2026.
5. Noetik. How do you use a virtual cell to do something actually useful? (2/3). September 12, 2025.
6. Noetik via Business Wire. Noetik Achieves First Milestone in Landmark GSK Collaboration.
7. Danaher via PR Newswire. Danaher Accelerates Life Sciences Research with New AI-powered Autonomous Lab. October 7, 2026.
8. KraneShares. AGIX holdings. Data as of October 31, 2026.
9. GSK, "GSK Expands Precision Medicine Collaboration with Tempus to Accelerate R&D," retrieved 10/31/2026.
10. Leica Biosystems, "Leica Biosystems brings Tempus pathology products to Aperio AI Store, accelerating AI-powered capabilities for oncology research," as of 9/25/2026.
11. Tempus AI. Tempus Reports Second Quarter 2026 Results. July 30, 2026.
12. National Human Genome Research Institute. The Cost of Sequencing a Human Genome. Retrieved 10/31/2026.