Google Cloud will add quantitative AI models from SandboxAQ (an Alphabet spinout focused on quantum-informed AI for scientific computing) to its cloud marketplace, the company announced June 30. The deal gives enterprise and research customers a way to run AI built for chemistry, biology, and semiconductor manufacturing through the same cloud console they already use for Gemini.
The distinction between what SandboxAQ calls large quantitative models and the large language models that dominate the industry is not cosmetic. LLMs are trained on human-generated text and are very good at generating more of it. Quantitative models are trained on numerical data, scientific equations, and laboratory measurements. The right output for a drug-discovery problem is not a well-phrased paragraph; it is a molecular binding prediction, a simulation result, or a structural formula. SandboxAQ argues its models are built for that output class, where LLMs consistently underperform.
The practical configuration on Google Cloud pairs the two types. Gemini handles natural language reasoning and the user-facing interface. The quantitative model handles the underlying computation in drug discovery, materials science, or semiconductor process design. Neither replaces the other; the architecture treats them as complementary layers.
The commercial logic deserves direct analysis. Google is reselling models from a company it already owns through Alphabet. SandboxAQ was spun out from Google in 2022 as a standalone entity, though Alphabet retains an interest. Putting SandboxAQ models on Google Cloud is not a neutral third-party distribution deal; it extends one Alphabet asset’s reach through another. The enterprise research segment is high-value and poorly served by general chatbots, which makes it a sensible bet. Whether the arrangement gives SandboxAQ’s models a privileged position on the marketplace, ahead of genuinely independent scientific AI providers, is a question the announcement does not address.
The Next Web reported June 30 that Google paired the marketplace listing with a broader product called Gemini for Science, a bundle drawing on AlphaEvolve, an AI co-scientist, an empirical research assistant, and NotebookLM, aimed at accelerating the routine steps of the scientific method. The bundle is positioned as a productivity layer for researchers, not a replacement for them.
Google’s scientific AI bets have produced credible results before this announcement. DeepMind’s protein-structure work reshaped parts of drug development; a separate GNOMEe materials effort reportedly found more new materials in a single year than science had catalogued previously. The common thread in those successes is narrow, measurement-trained AI rather than general-purpose systems, which validates the premise SandboxAQ is selling.
Google said unnamed partners are already using the models in private preview for real-world R&D. No results or organizations were disclosed.
The substantive change here is distribution. SandboxAQ’s quantitative models were previously accessible mainly to specialized research groups with the expertise to run them. A Google Cloud listing converts that access into something any enterprise R&D team can provision with a billing account. Research teams currently evaluating AI infrastructure contracts for 2027 budgets should benchmark quantitative model performance against their domain’s specific numerical tasks before assuming Gemini alone covers the workload.
Reporting by Ana Maria Constantin for The Next Web, published June 29, 2026.
