What's New
SandboxAQ, a company spun out of Google in 2022, has raised over $950 million in funding and is backed by NVIDIA. The company is focused on developing Large Quantitative Models (LQMs) — AI systems that simulate physics and chemistry to discover new drugs and materials. This was highlighted in a recent podcast episode featuring Nadia Harhen, GM of AI Simulation at SandboxAQ, who argues that LQMs may have a larger real-world impact than LLMs.
Technical Details
LQMs differ from LLMs in a fundamental way: instead of predicting the next token in a sequence of text, they predict the behavior of physical systems. SandboxAQ's models are trained on quantum mechanical simulations, molecular dynamics, and other physics-based data. The models can:
- Predict molecular properties (e.g., binding affinity, toxicity) without costly lab experiments
- Simulate chemical reactions at a quantum level
- Optimize material structures for specific properties (strength, conductivity, etc.)
The company has not published specific benchmark results or model architectures publicly, but the core technology leverages neural networks trained on large-scale physics simulation data. This is distinct from the transformer-based LLMs that dominate text and image generation.
How It Compares
Data type Text, code, images Physics simulations, molecular data Output Text, code, images Molecular properties, material structures Target domain Language, reasoning, creativity Drug discovery, materials science, chemistry Funding Billions from major tech $950M+ from strategic investors Backers Microsoft, Google, etc. NVIDIA, Google, othersWhile LLMs have captured public attention, LQMs operate in a domain where even small improvements can have massive economic and societal impact. A 10% better drug molecule can save billions in R&D and save lives.
What to Watch
SandboxAQ's approach is promising but faces significant challenges:
- Validation: The company has not yet published peer-reviewed results or demonstrated a drug or material that entered clinical trials or commercial use.
- Competition: Other players like Schrödinger, D.E. Shaw Research, and Microsoft's Azure Quantum are also working on AI-driven molecular simulation.
- Scaling: LQMs require enormous compute for training — likely why NVIDIA is a backer. The cost of simulating quantum systems at scale remains high.
- Integration: For LQMs to be useful, they must integrate into existing pharmaceutical and materials R&D workflows, which are slow and risk-averse.
gentic.news Analysis
SandboxAQ's $950M+ raise is a strong signal that investors see real potential in AI beyond language. The company's focus on LQMs — rather than yet another chatbot — positions it in a space where AI can directly impact physical industries like pharmaceuticals, energy, and manufacturing. This aligns with a broader trend we've covered: the shift from pure language models to domain-specific AI systems that solve concrete engineering problems.
Notably, the backing from NVIDIA is strategic. NVIDIA's GPUs are the workhorses for training both LLMs and LQMs, but LQMs require specialized hardware for quantum chemistry simulations. This partnership could give SandboxAQ preferential access to next-gen hardware and software optimizations.
However, the lack of published benchmarks is a concern. In the AI community, claims without evidence are treated with skepticism. SandboxAQ should release open benchmarks or peer-reviewed papers to build credibility. Until then, the $950M+ is a bet on the team and vision, not on proven results.
Frequently Asked Questions
What are Large Quantitative Models (LQMs)?
LQMs are AI models trained on physics and chemistry simulation data to predict molecular properties, chemical reactions, and material behaviors. Unlike LLMs that process text, LQMs work with numerical data from quantum mechanics and molecular dynamics.
How does SandboxAQ differ from other AI drug discovery companies?
SandboxAQ focuses on "quantitative" models — meaning they simulate physics directly rather than learning from experimental data alone. This allows them to explore novel molecules that haven't been synthesized yet, which is a key advantage over models that only learn from existing data.
Why is NVIDIA backing SandboxAQ?
NVIDIA's GPUs are essential for training large-scale physics simulations. Backing SandboxAQ gives NVIDIA a foothold in the emerging LQM market, which could drive demand for their hardware in scientific computing — a growing segment beyond AI.
When will we see real-world products from SandboxAQ?
SandboxAQ has not announced specific timelines. The company will likely need several years to validate its models in real drug or materials discovery pipelines. Watch for partnerships with pharmaceutical companies or material science firms as leading indicators.
