The large language models that power most of the AI industry are very good at words and surprisingly unreliable at numbers. Google's latest move is an admission that, for science, a different kind of model is needed.
The company said it will start offering specialist AI models from SandboxAQ through Google Cloud, adding what SandboxAQ calls large quantitative models to the cloud marketplace. The aim is to widen enterprise and research access to AI built for drug discovery, materials science, and semiconductor manufacturing, the announcement said.
Why quantitative models matter
The distinction is the whole point. Large language models are trained on text and excel at generating it. Large quantitative models, by SandboxAQ's description, are trained on numerical data and scientific equations rather than prose, which is meant to make them better suited to problems in chemistry, biology, and physics, fields where the right answer is a number or a structure, not a fluent paragraph.
SandboxAQ was spun out of Alphabet in 2022 and focuses on applying AI and quantum technology to solve complex scientific and industrial problems. Its large quantitative models (LQMs) are built on neural network architectures that can handle mathematical relationships, molecular structures, and physical simulations with higher accuracy than generic language models.
On Google Cloud, researchers will be able to combine these with Gemini, using the language model for reasoning and interface and the quantitative model for the underlying science. Google paired the marketplace move with Gemini for Science, a bundle of tools and experiments aimed at the research workflow itself. It draws on projects the company has been building for a while, including its AI co-scientist, the AlphaEvolve coding agent, an empirical research assistant, and NotebookLM, and is pitched as a way to speed up the routine, laborious steps of the scientific method rather than to replace the scientist.
DeepMind's legacy in scientific AI
That framing is consistent with where Google has put its scientific weight. DeepMind's protein-structure work has already reshaped parts of drug development, and a separate effort produced an AI that found more new materials in a year than science had catalogued in its entire history. The common thread is that the highest-value AI in the sciences tends to be narrow and trained on real measurements, not general and trained on the internet.
DeepMind's AlphaFold, for instance, solved the protein folding problem that had eluded scientists for decades. It predicted the three-dimensional structures of hundreds of millions of proteins, accelerating drug discovery, enzyme design, and understanding of diseases. Similarly, the Graph Networks for Materials Exploration (GNoME) discovered 380,000 stable materials, including many with properties that could lead to new batteries, superconductors, or catalysts.
These successes demonstrated that AI trained on scientific data—crystallographic databases, molecular simulations, and experimental results—can generate predictions that are both accurate and novel. However, such models are expensive to build and require domain expertise. By offering SandboxAQ's LQMs through its marketplace, Google is lowering the barrier for organizations that lack the resources to develop their own.
Marketplace strategy and commercial logic
The commercial logic is straightforward. Google is competing with the other hyperscalers to be the default place enterprises run AI, and scientific and industrial R&D is a high-value segment that general chatbots do not serve well. Selling specialist models through the marketplace, the same channel through which it already offers a wide catalogue of third-party systems, lets Google capture that demand without having to build every domain model itself.
It also fits a broader scramble to turn AI into actual laboratory results. DeepMind's own drug-discovery spinoff Isomorphic Labs is moving toward trials, and rivals across the industry are racing to convert algorithmic promise into treatments and materials that work outside a benchmark. Putting quantitative models in front of enterprise researchers is Google's bid to be the infrastructure underneath that race.
SandboxAQ's models are not the only ones of their kind. Microsoft offers AI for science through Azure Quantum, and Amazon Web Services provides tools for computational chemistry and materials design. But Google's advantage lies in the integration with Gemini and its existing AI ecosystem, including Vertex AI and the newly announced Gemini for Science suite.
How Gemini for Science enhances the workflow
Gemini for Science is a collection of tools that target different stages of the scientific method. The AI co-scientist helps researchers formulate hypotheses and design experiments. AlphaEvolve is an AI coding agent that can write and optimize simulation code. The empirical research assistant uses natural language processing to summarize papers and extract data from experimental reports. NotebookLM allows users to upload documents and interact with them using AI-driven questioning.
Together, these tools are meant to reduce the time scientists spend on literature reviews, data processing, and writing code. Instead of spending weeks cleaning data, a researcher can ask the AI to identify patterns. Instead of manually combing through thousands of papers, they can query NotebookLM for specific results. The quantitative models from SandboxAQ then run the actual simulations, returning numerical predictions that the researcher can validate in the lab.
Applications in drug discovery and materials science
In drug discovery, LQMs can predict how a small molecule will bind to a target protein, which is traditionally done through expensive computational chemistry or physical experiments. They can also suggest modifications to improve potency or reduce toxicity. In materials science, they can propose new compounds with desired properties—like higher electrical conductivity or greater strength—without having to synthesize every candidate.
Semiconductor manufacturing is another frontier. As transistors shrink to atomic scales, classical physics breaks down and quantum effects dominate. LQMs trained on quantum mechanical equations can simulate electron behavior more efficiently than traditional methods, potentially reducing the time and cost of designing new chip architectures.
Google said the capabilities are already in use by partners in private preview for real-world R&D, though it has been sparing with specifics on which organisations and what results. One likely early adopter is the pharmaceutical industry, where speed-to-clinic is critical. Another is the energy sector, where better battery materials could accelerate the transition to renewable energy.
The challenges of scientific AI
Despite the promise, there are significant hurdles. Scientific AI models require high-quality, well-annotated data. Many experimental results are not shared publicly, and proprietary datasets are often small. Furthermore, AI predictions must be validated experimentally, which is time-consuming and expensive. A model that generates a thousand potential drug candidates is useless if each one costs $100,000 to test.
There is also the problem of uncertainty quantification. A quantitative model that confidently returns a wrong answer can send researchers down a dead end. Google and SandboxAQ claim their models include measures of confidence, but in practice, these are still being refined. The scientific community is cautious, and trust in AI is built incrementally through reproducible results.
Regulatory hurdles also apply. In drug discovery, any prediction that leads to a clinical candidate must be backed by sufficient evidence to satisfy the FDA or other agencies. AI-generated insights are not yet accepted as primary evidence; they are used to prioritize targets and design experiments.
What this means for the future of AI in science
The availability of specialist models on a cloud marketplace marks a shift from bespoke, in-house AI to accessible, rentable tools. Over time, this could democratize access to state-of-the-art scientific AI, allowing smaller universities, startups, and even individual researchers to use models that were once the preserve of large labs.
It also encourages standardization. If many researchers use the same underlying models, it becomes easier to compare results and build on each other's work. However, there is a risk of over-reliance on black-box models whose training data and limitations are opaque. The scientific method demands transparency, and it remains to be seen how Google and SandboxAQ will address reproducibility.
Google's investment in scientific AI is not just about cloud revenue. It is a long bet that the hardest problems in science—understanding diseases, designing new materials, fighting climate change—are solvable with the right computational tools. By providing both the quantitative models and the reasoning engine, Google positions itself as an essential partner in the next era of discovery. Whether the marketplace model will yield breakthroughs or simply faster spreadsheets is the question the private previews are meant to answer.
Source: TNW | Google News