On June 2, 2026, Microsoft officially general-launched Microsoft Discovery, its highly anticipated agentic AI platform for scientific and industrial research, and simultaneously disclosed that mining titan BHP has already harnessed the service to screen over half a million copper-leaching reagent candidates—underscoring the platform’s immediate enterprise viability.

The announcement, made at a virtual event, positions Microsoft Discovery as a cornerstone of Azure’s expanding AI portfolio, one that moves beyond conversational assistants into autonomous, multi-step reasoning for complex problem domains. For BHP, the world’s largest miner by market capitalization, the collaboration signals a step-change in how the resources industry approaches R&D for sustainable metal extraction.

Inside Microsoft Discovery

Microsoft Discovery is not a single model but an orchestration layer that coordinates multiple specialized AI agents, each responsible for a segment of the scientific workflow. The platform builds on Azure’s infrastructure, combining large language models (LLMs), graph-based reasoning engines, quantum chemistry simulations, and high-throughput computing.

At its core, Discovery allows researchers to define a goal—such as “find a reagent that maximizes copper recovery while minimizing environmental impact”—and then autonomously generates, tests, and refines hypotheses. The system iterates through cycles of virtual experimentation, learning from each result to narrow down the search space.

Key capabilities include:
- Hypothesis generation: Agents mine existing literature, patents, and internal databases to propose new molecular candidates or experimental conditions.
- Virtual screening: Integration with Azure Quantum and GPU clusters enables rapid simulation of molecular properties, reaction kinetics, and toxicity profiles.
- Multi-objective optimization: The AI balances conflicting goals, such as cost versus efficacy, using Pareto frontier analysis.
- Uncertainty quantification: Every prediction comes with a confidence score, allowing researchers to prioritize candidates for physical validation.

The platform supports a broad range of scientific domains, from materials science and chemistry to biology and physics, with domain-specific modules that can be customized via APIs or a low-code interface.

The BHP Case Study: Tackling Copper Leaching

Copper is indispensable for the global energy transition, used in everything from electric vehicles to wind turbines. Yet extracting it from low-grade ores—an increasingly common situation as high-grade deposits dwindle—relies on leaching, where chemical reagents dissolve the copper for recovery. Finding optimal leaching agents is both economically critical and environmentally sensitive.

BHP’s collaboration with Microsoft began through a joint innovation program focused on accelerating digital transformation in mining. Even before the general availability date, BHP gained early access to Discovery under a private preview, allowing its scientists to define the problem, set constraints, and let the AI loose on a chemical landscape of staggering breadth.

According to BHP, the agentic AI screened more than 500,000 potential reagents in a matter of weeks. By comparison, a state-of-the-art physical high-throughput screening lab might manage 5,000–10,000 experiments per day, but at significant cost in materials, labor, and time. Virtual screening entirely eliminates the physical bottleneck, and agentic AI adds the ability to intelligently navigate the search space rather than brute-force it.

The AI considered factors such as:
- Copper dissolution rate and yield
- Reagent stability and reusability
- Toxicity and biodegradability
- Compatibility with existing heap leaching or in-situ processes
- Cost of synthesis at scale

Initial results identified several novel chemical families that had not been previously explored for copper leaching. BHP’s R&D team is now in the process of synthesizing and testing the top candidates in the lab—a focused task made possible only by the AI’s ability to filter 500,000 possibilities down to a manageable shortlist.

Why Agentic AI Changes the Game

The term “agentic AI” describes systems that don’t just respond to prompts but proactively plan and execute multi-step tasks. In the context of scientific discovery, this means the AI acts like an ever-vigilant researcher: constantly reading, hypothesizing, designing experiments, analyzing results, and adjusting its strategy.

Traditional AI-assisted drug discovery or materials screening often requires human scientists to define each step, run simulations, and interpret output before deciding the next move. Agentic AI removes much of that human-in-the-loop friction, enabling continuous, 24/7 research cycles.

For BHP, this autonomy meant that once the initial parameters were set, the Discovery agents worked nonstop, exploring chemical space while human scientists focused on higher-level strategy. The system not only scanned known reagent classes but also ventured into unexplored molecular territory, something a manual process might never attempt due to bias or resource constraints.

Azure as the AI Backbone

The computational demands of screening half a million reagents are immense. Each candidate typically requires quantum mechanical calculations, solvation models, and possibly molecular dynamics simulations. Microsoft Discovery orchestrates these tasks on Azure’s global infrastructure, automatically scaling GPU and CPU resources as needed.

Microsoft has tightly integrated its quantum computing initiatives with Discovery. While full-scale quantum computers remain years away, quantum-inspired optimization algorithms and classical quantum chemistry solvers run on conventional hardware, accelerating the virtual screening. This hybrid approach gave BHP the throughput required to tackle the combinatorial explosion.

Enterprise-grade features, including role-based access control, data residency, and compliance with industry standards, made it feasible for BHP to use a cloud-based platform for sensitive research. The consumption-based model also means BHP paid only for the compute and storage used, avoiding the capital expenditure of building an in-house supercomputing cluster.

The Road to General Availability

Microsoft Discovery was first teased at Build 2024 as a research prototype under the name “Project Alexandria.” A private preview followed in 2025, limited to select partners in pharma and energy. During that period, Microsoft refined the agentic reasoning engine, expanded the simulation backends, and developed industry templates.

Feedback from early adopters shaped the GA release. Key improvements include a more intuitive “goal-first” user interface, where researchers describe desired outcomes in natural language, and the platform automatically decomposes the problem into agent tasks. A library of pre-built agents for common tasks (e.g., retrosynthesis, molecular docking, alloy design) accelerates onboarding.

The June 2 launch also introduced integration with Microsoft Copilot in Azure, allowing researchers to converse with the system and visualize results in Power BI dashboards. A new “Experiment Designer” module lets users create and share multi-step virtual lab protocols.

Broader Impact on Industrial R&D

BHP’s success is not an isolated case. The same agentic AI approach can be applied to:
- Drug discovery: Screening billions of molecules for target binding.
- Battery materials: Identifying new electrolytes or cathode materials.
- Carbon capture: Designing sorbents optimized for CO₂ selectivity and stability.
- Polymers and composites: Finding materials with specific mechanical or thermal properties.

The common thread is the dramatic compression of the R&D timeline. What once took a decade and billions of dollars can now potentially be accomplished in months and for a fraction of the cost. For industries facing existential shifts—like mining, which must decarbonize and minimize environmental footprint—the ability to rapidly innovate becomes a competitive imperative.

Moreover, the platform enables a form of “instant expertise” in niche chemical areas. A mining company might not have in-house specialists for exotic reagent chemistry, but Discovery can synthesize knowledge from global research and apply it to the problem at hand.

Challenges and Realities

Despite the reported success, virtual screening is not a silver bullet. Computational predictions can fail due to inaccurate force fields, missing solvent effects, or unforeseen kinetic barriers. BHP acknowledges that only a fraction of the AI-identified reagents will prove successful in real leaching columns, where complex mineralogy and microbial activity come into play.

Data availability is another constraint. The AI’s reasoning is only as good as the data it learns from. In mining, much of the chemical data is proprietary or buried in historical reports. Microsoft and BHP addressed this by curating a vast internal dataset, but smaller players may struggle to achieve similar results without high-quality training data.

There are also ethical and workforce considerations. The automation of R&D could reduce the demand for certain lab roles, while creating new opportunities for data-savvy scientists. Microsoft emphasizes that Discovery is a tool to augment, not replace, human ingenuity, but the long-term impact on employment in research-intensive fields remains a topic of debate.

Pricing and Availability

Microsoft Discovery is available now on the Azure Marketplace, with pricing based on compute consumption, simulations executed, and agent cycles used. A free tier allows exploration with limited throughput, while enterprise agreements include dedicated capacity and advanced support. BHP’s deployment likely falls under a custom quantum computing and AI engagement, though financial terms were not disclosed.

What’s Next for Microsoft Discovery

Microsoft’s roadmap includes tighter integration with robotic labs, enabling AI to not only suggest experiments but also order their execution in fully automated wet labs. Ambitions also extend to “federated Discovery,” where multiple organizations can pool data securely while preserving privacy, accelerating research across industries.

In the near term, the company plans to release domain-specific modules for metals and mining, battery technology, and climate science, building on learnings from the BHP engagement.

The Windows and Azure Developer Connection

For the Windows enthusiast audience, the Microsoft Discovery launch reinforces Azure’s role as the place where cutting-edge AI workloads run. The same Azure tools used for web apps or databases now support autonomous scientific agents. Developers can build on Azure AI Studio and Copilot Stack to create intelligent applications that leverage agentic frameworks.

While Discovery is primarily for researchers, its underlying orchestration engine will eventually trickle down to more mainstream services, enabling smarter automation in everything from business intelligence to IT operations. The message is clear: Microsoft’s AI bet is on agents that can think, plan, and act—and that future is arriving faster than many expected.

Conclusion

The general availability of Microsoft Discovery, paired with BHP’s striking early results, marks a seminal moment in the journey toward AI-driven R&D. Screening over 500,000 copper-leaching reagents in weeks—a task that would otherwise stretch over years—demonstrates tangible value, not just theoretical potential.

As industries grapple with sustainability pressures and intensifying competition, platforms like Discovery could become as essential as cloud computing itself. For Windows enthusiasts, it’s a glimpse of how Microsoft’s ecosystem is evolving, turning Azure into the world’s largest automated laboratory. The experiments may be virtual, but the impact on mining, medicine, and materials will be very real.