Closing the Data Loop: How AI Is Rewriting the Economics of Drug Discovery

The pharmaceutical industry's most stubborn problem — escalating costs and timelines — is being tackled by AI systems that learn from their own experimental results, and the implications reach far beyond pharma.

---

AI Summary

Drug discovery has long been governed by Eroom's Law — the pharmaceutical industry's inverse of Moore's Law — where the cost of developing a new drug has roughly doubled every nine years since the 1950s. Today, bringing a single new drug to market takes an average of 10 to 15 years and costs billions. AI-driven drug discovery is now attempting to break this cycle by "closing the data loop": creating continuous feedback systems where AI predictions are tested experimentally, and the results flow back into the models to improve future predictions. For Malaysian businesses, this matters not only for the healthcare and biotech sectors but also as a masterclass in how any industry can use AI to compress research cycles, reduce waste, and gain first-mover advantage in increasingly competitive markets.

---

Key Takeaways

• Eroom's Law is the pharmaceutical industry's core problem — costs have doubled every nine years since the 1950s, making drug development economically unsustainable without intervention.
• "Closing the data loop" means AI systems learn from experimental outcomes — predictions are validated in the lab, and results feed back into the model, creating a self-improving cycle.
• First-mover advantage is becoming decisive in pharmaceuticals, meaning companies that adopt AI-driven discovery early capture disproportionate market value.
• The 10 to 15-year drug development timeline is the target AI is attempting to compress — not by skipping steps, but by dramatically reducing failed experiments.
• The data loop concept applies beyond pharma — any Malaysian industry that runs iterative research, testing, or quality control can adopt the same architectural pattern.

---

What Happened

The pharmaceutical industry has been locked in a decades-long spiral of rising costs and diminishing returns. Since the 1950s, the cost of developing a single new pharmaceutical has roughly doubled every nine years — a trend known as Eroom's Law, which is Moore's Law spelled backwards. While computing power gets cheaper and faster over time, drug development has moved in the opposite direction: more expensive, slower, and less efficient.

Today, bringing a new drug to market takes an average of 10 to 15 years. The vast majority of compounds that enter preclinical testing never make it to human trials. Of those that do, only a small fraction receive regulatory approval. This means that for every successful drug, pharmaceutical companies have absorbed the cost of dozens or hundreds of failed candidates — and those costs are passed on to healthcare systems, insurers, and patients.

AI-driven drug discovery is emerging as a structural response to this problem. Rather than relying solely on traditional wet-lab experiments — where scientists physically synthesise and test thousands of compounds — AI models can predict which molecules are most likely to bind to specific disease targets, simulate their behaviour, and rank candidates before a single physical experiment is conducted. But the key innovation is not prediction alone. It is what the industry calls "closing the data loop."

Closing the data loop means that when AI predictions are tested in actual laboratory experiments, the results — whether positive or negative — are fed back into the AI model. A prediction that fails is just as valuable as one that succeeds, because the model learns from the discrepancy. Over time, the AI system becomes progressively more accurate, reducing the number of failed experiments, compressing timelines, and ultimately lowering the cost of drug development.

This approach is gaining traction because the pharmaceutical market is increasingly defined by first-mover advantage. The company that reaches the market first with a new therapy captures the lion's share of value, particularly in areas like oncology, rare diseases, and infectious diseases where unmet need is high and competition is intense.

---

Why It Matters

The significance of closing the data loop extends well beyond the pharmaceutical industry. It represents a fundamental shift in how AI is being applied to complex, high-cost, high-uncertainty problems — and it offers a blueprint that applies to any sector where experimentation is expensive and failure rates are high.

Consider the economics. If Eroom's Law continues unchecked, the cost of developing new drugs will eventually exceed what healthcare systems — including Malaysia's — can afford. This is not an abstract concern. Malaysia's pharmaceutical market, valued at several billion ringgit, depends on a global pipeline of new drugs. If that pipeline slows or becomes more expensive, the consequences flow directly to Malaysian patients, hospitals, and the Ministry of Health's procurement budget. Drug pricing negotiations, access to innovative therapies, and the sustainability of public healthcare are all connected to what happens upstream in drug discovery.

The data loop concept also matters because it represents a maturation of AI from a tool that generates predictions to a system that learns from its own mistakes. This is a critical distinction. Many businesses currently use AI in a static way — they train a model, deploy it, and leave it. Closing the data loop requires a dynamic architecture: the model must continuously ingest new experimental data, update its understanding, and refine its output. This is harder to build but dramatically more valuable.

For investors and business strategists, the first-mover dynamic is worth noting. In pharmaceuticals, as in technology, the returns to being early are outsized. Companies that build AI-driven discovery platforms with closed data loops today will compound their advantage over time, because each cycle of prediction and experimentation makes their models better than competitors who start later. This creates a widening moat — a pattern familiar to anyone who has watched platform dynamics in software or e-commerce.

---

What This Means for Malaysia

Malaysia's connection to this trend operates on several levels. First, Malaysia has a growing pharmaceutical and biotechnology sector, with companies like Duopharma Biotech, Kotra Pharma, and CCM Pharmaceuticals operating in the generic and branded-generics space. While these companies are not currently at the frontier of AI-driven drug discovery, the downward pressure that AI exerts on drug development costs globally will eventually reshape the competitive landscape. If multinational pharmaceutical companies can develop drugs faster and cheaper using AI, Malaysian generics manufacturers may face shorter exclusivity windows and tighter margins.

Second, Malaysia is positioning itself as a player in the broader AI and digital economy through initiatives like MyDIGITAL, the National AI Roadmap, and MDEC's digital economy programmes. The data loop architecture — continuous learning from experimental feedback — is directly relevant to Malaysia's ambitions in smart manufacturing, agricultural biotechnology, and halal pharmaceuticals. A Malaysian company developing halal-certified nutraceuticals, for example, could apply the same closed-loop AI approach to optimise formulations faster than competitors using traditional trial-and-error methods.

Third, Malaysia's clinical research infrastructure — including institutions like Clinical Research Malaysia (CRM) and the network of Clinical Research Centres under the Ministry of Health — positions the country as a site for clinical trials. If AI-driven discovery compresses the preclinical phase and pushes more compounds into human trials faster, Malaysia could see increased demand for clinical trial sites. This represents an economic opportunity, but it also requires regulatory readiness from the National Pharmaceutical Regulatory Agency (NPRA) to evaluate AI-assisted drug applications.

Finally, there is a talent and capability dimension. Closing the data loop requires interdisciplinary teams — computational biologists, data engineers, wet-lab scientists, and regulatory specialists. Malaysian universities like Universiti Malaya, Universiti Sains Malaysia, and Universiti Teknologi Malaysia are producing graduates in these fields, but the integration of AI into life sciences curricula remains uneven. Strengthening this pipeline is essential if Malaysia wants to participate in the AI-driven drug discovery economy rather than merely consume its outputs.

---

How Your Business Can Use This

You do not need to be a pharmaceutical company to learn from the data loop architecture. The core principle — use AI to make predictions, test them in the real world, feed results back, and improve — applies to any business that runs iterative processes.

For manufacturing and quality control teams: If you operate a factory in Penang, Selangor, or Johor, you can apply closed-loop AI to process optimisation. Use AI to predict which production parameters will yield the best output, run the batch, measure the results, and feed the data back into the model. Over time, your AI system learns which combinations of temperature, pressure, timing, and material inputs produce optimal results — reducing waste and improving yield.

For R&D-intensive SMEs: If your company develops products — whether food formulations, cosmetics, specialty chemicals, or materials — you can structure your testing programme around AI predictions. Rather than testing every variant, use AI to prioritise the most promising candidates, test those, and use the results to refine the next round of predictions. This compresses development cycles and reduces the cost of failed experiments.

For healthcare and life sciences organisations: If you are a hospital, research institute, or health-tech startup in Malaysia, consider how patient outcome data could feed back into predictive models. A diagnostic AI that predicts patient risk, for example, becomes more accurate when actual outcomes are systematically fed back into the model. Ensure your data architecture supports this loop — many organisations collect outcome data but never connect it back to the prediction system.

Getting started this quarter: Map one iterative process in your business. Identify where predictions are made (even manually), where they are tested, and where the results are or are not fed back into decision-making. The gap between your prediction and your outcome data is where a closed-loop AI system would create value.

---

The Agentic AI Angle

Closing the data loop in drug discovery is fundamentally an agentic AI problem — it requires AI systems that do not merely answer questions but actively plan, execute, and learn across multiple steps.

An agentic AI system for drug discovery would operate as follows. First, the agent receives a target — for example, a protein implicated in a disease. It searches molecular databases, generates candidate compounds, predicts their binding affinity using trained models, and ranks them. It then designs the experimental protocol — which assays to run, in what order, with what controls. After the lab returns results, the agent ingests the data, identifies where its predictions were wrong, updates its model parameters, and generates a refined set of candidates for the next round.

This is not science fiction. The architecture is being built now in leading pharmaceutical AI companies. The agent operates across multiple tools — molecular databases, simulation software, lab information management systems, and statistical analysis packages — coordinating them without human intervention at each step.

For Malaysian businesses, the agentic AI lesson is about workflow orchestration. An agent that manages a closed-loop process — predict, test, learn, repeat — can be applied to supply chain optimisation, customer behaviour modelling, predictive maintenance, and financial forecasting. The pharmaceutical industry is simply the most visible testing ground for a pattern that will diffuse across sectors.

---

Risks and Limitations

Several risks warrant attention. First, AI predictions are only as good as the data they are trained on. If historical drug discovery data contains biases — for example, over-representation of certain molecular families or disease areas — the AI system will inherit and potentially amplify those biases. This is particularly relevant for diseases that affect Southeast Asian populations, which have historically been under-represented in global pharmaceutical research databases.

Second, biological systems are enormously complex. A molecule that performs well in computational simulations may fail in living organisms due to factors the model did not account for — toxicity, metabolic instability, or unexpected immune responses. AI can reduce the number of failures, but it cannot eliminate them. Overconfidence in AI predictions could lead companies to cut corners in safety testing, which would be dangerous and regulatorily unacceptable.

Third, there are data governance considerations. Closing the data loop requires sharing experimental data across systems and potentially across organisations. In a Malaysian context, this raises questions under the Personal Data Protection Act (PDPA), particularly if patient-derived data is involved in the loop. Companies must ensure their data architecture is compliant before building closed-loop systems.

---

The Bottom Line

Eroom's Law has made drug discovery progressively more expensive for seventy years. AI-driven closed-loop systems — where predictions are tested, results are fed back, and models continuously improve — represent the most credible structural attempt to reverse that trend. The companies that build these systems early will compound their advantage, and the industries that adopt the same architecture will see similar gains in efficiency and cost.

For Malaysian business leaders, the actionable insight is this: find the iterative process in your business where predictions are made and tested, and build a feedback loop around it. You do not need a pharmaceutical budget to benefit from the data loop architecture. You need the discipline to connect your outcomes back to your predictions — and the AI infrastructure to learn from the gap.

---

FAQ

What is Eroom's Law and why should a Malaysian business care? Eroom's Law is the observation that drug development costs double roughly every nine years. Malaysian businesses should care because rising global drug costs affect local healthcare budgets, and because the AI methods being developed to break this pattern are applicable to other industries.

Can Malaysian companies participate in AI-driven drug discovery? Yes, but it requires investment in interdisciplinary talent, computational infrastructure, and partnerships with global pharma or research institutions. Malaysian clinical research infrastructure and the NPRA regulatory framework are relevant enablers.

What does "closing the data loop" mean in simple terms? It means using AI to make predictions, testing those predictions in the real world, and feeding the results — successes and failures — back into the AI model so it improves over time. The same principle can be applied to manufacturing, R&D, and operations in any industry.

---

Sources / References

• MIT Technology Review — "Closing the data loop in AI-driven drug discovery" (https://www.technologyreview.com/2026/07/27/1139667/closing-the-data-loop-in-ai-driven-drug-discovery/) — Provided the core facts on Eroom's Law, drug development timelines and costs, the concept of closing the data loop, and the first-mover advantage dynamic in pharmaceuticals. All factual claims in this article derive from this source.