Google Launches AlphaEvolve for Everyone on Google Cloud, Bringing AI-Powered Algorithm Discovery to Enterprise Optimization

Google Makes AlphaEvolve Generally Available on Google Cloud

Artificial intelligence is entering a new phase where it is no longer limited to helping developers write code. Instead, it is beginning to discover entirely new algorithms and optimize software in ways that would be nearly impossible through traditional engineering alone.

Google has officially announced the general availability (GA) of AlphaEvolve, making its advanced AI-powered optimization platform accessible to organizations worldwide through the Gemini Enterprise Agent Platform on Google Cloud.

Originally introduced as a private preview, AlphaEvolve has evolved from a research initiative into an enterprise-ready AI system capable of discovering better algorithms, improving existing code, and solving optimization challenges across multiple industries.

From supply chain optimization and semiconductor design to genomics, scientific computing, financial modeling, and AI infrastructure, AlphaEvolve is already demonstrating measurable business impact.

The launch represents another major milestone in Google's vision of transforming AI from a productivity assistant into a discovery engine capable of expanding human innovation.

What Is AlphaEvolve?

AlphaEvolve is an AI-powered code optimization and algorithm discovery agent built on top of Google's Gemini models.

Unlike traditional coding assistants that generate code based on prompts, AlphaEvolve continuously searches through enormous numbers of possible algorithmic variations to discover solutions that outperform existing implementations.

Rather than suggesting small improvements, the system can redesign significant portions of software while respecting performance requirements and operational constraints.

Its primary goal is simple:

Find better algorithms than humans might reasonably discover through manual experimentation.

This makes AlphaEvolve particularly valuable for problems where millions of possible implementations exist.

Examples include:

• Supply chain optimization

• Route planning

• Financial forecasting

• Semiconductor design

• GPU optimization

• Machine learning training

• Scientific simulations

• Quantum computing

• Genomics

• High Performance Computing (HPC)

The Four-Step Optimization Process

Google designed AlphaEvolve around a structured optimization workflow.

1. Define

Developers provide:

• Existing code

• Baseline algorithm

• Business objectives

• Technical constraints

• Additional domain knowledge

This establishes the starting point for AI exploration.

2. Measure

Organizations define an evaluation function that scores candidate algorithms.

Typical evaluation metrics include:

• Accuracy

• Runtime

• Latency

• Cost

• Memory usage

• Reliability

• Operational constraints

This objective scoring system allows AlphaEvolve to determine which generated solutions are superior.

3. Optimize

The AI agent repeatedly generates new algorithmic candidates.

Each candidate is automatically evaluated.

Poor solutions are discarded.

Successful solutions become the foundation for further optimization.

This evolutionary process enables AlphaEvolve to explore software designs far beyond what manual development could realistically achieve.

4. Apply

Once optimization completes, organizations can directly deploy the improved algorithm into production environments.

The optimized code becomes part of live business systems while engineers continue reviewing and validating results.

Real-World Enterprise Success Stories

Google revealed that AlphaEvolve has already been tested across numerous enterprise environments.

The diversity of industries demonstrates that optimization challenges exist almost everywhere.

Below are some of the most notable deployments.

BASF

Building a Digital Twin for Global Supply Chains

Chemical giant BASF used AlphaEvolve to build a sophisticated digital twin capable of modeling its highly complex supply network.

Previous deterministic models had failed.

AlphaEvolve successfully captured both system behavior and human operational decisions.

The result was a highly accurate, data-driven supply chain model.

Reported improvement:

• More than 80% improvement in planning and forecasting models.

Coolblue

Better Demand Forecasting

E-commerce retailer Coolblue used AlphaEvolve to optimize its 28-day demand forecasting pipeline.

Within roughly 200 optimization iterations, AlphaEvolve produced:

• Better feature engineering

• Improved preprocessing

• Enhanced regression model ensembles

Result:

• More than 5% reduction in forecasting error (WMAPE)

Improved forecasting translates directly into better inventory management and reduced stock shortages.

FM Logistic

Warehouse routing is already heavily optimized.

Yet AlphaEvolve still managed to produce:

• 10.4% better routing efficiency

• More than 15,000 km reduction in employee travel

Benefits include:

• Faster fulfillment

• Lower operational costs

• Reduced fleet wear

• Better employee working conditions

Infineon

Semiconductor manufacturer Infineon is exploring AlphaEvolve for multiple chip development stages.

Initial experiments indicate strong potential across:

• Surrogate modeling

• Silicon optimization

• Chip design workflows

JetBrains

Software development company JetBrains applied AlphaEvolve to IDE performance optimization.

Results showed:

• 15 to 20% performance improvements

Engineers remained responsible for:

• Validation

• Benchmarking

• Release decisions

AI handled the exploration of massive optimization possibilities.

Kinaxis

Supply chain software provider Kinaxis reported:

• More than 22% improvement in forecasting metrics

• Over 90% runtime reduction

These gains improve enterprise planning speed while maintaining forecasting quality.

Klarna

Financial services company Klarna used AlphaEvolve on one of its largest machine learning training pipelines.

During three weeks:

• Nearly 6,000 candidate programs explored

Results:

• Doubled throughput

• Improved model quality

• Maintained strict financial compliance requirements

Kuro Games

Game developer Kuro Games applied AlphaEvolve to backend infrastructure optimization.

The AI successfully improved server-side workloads, allowing engineers to focus more on gameplay innovation.

Oak Ridge National Laboratory

One of the most technically impressive deployments occurred at Oak Ridge National Laboratory.

Running on the Frontier exascale supercomputer, AlphaEvolve optimized GPU kernels through automated generation, compilation, execution, and validation.

Researchers described it as an important step toward AI-assisted scientific software development.

Old Dominion University

Researchers studying biological aging used AlphaEvolve to discover mathematical models explaining mortality rates.

Remarkably, the system independently rediscovered established scientific models without prior knowledge while simultaneously producing new improvements.

PacBio

Genomics company PacBio used AlphaEvolve to enhance DNA sequencing.

Results included:

• 30% reduction in variant detection errors

Higher sequencing accuracy can improve medical research and disease discovery.

Pebble

Pebble applied AlphaEvolve to GPU inference optimization.

Results:

• 56% reduction in performance modeling error

This enables more accurate hardware configuration recommendations.

qBraid

Quantum computing startup qBraid used AlphaEvolve to improve quantum error correction.

The AI searched design spaces too large for human researchers to explore manually.

Schrödinger

Drug discovery company Schrödinger reported:

• Four times faster molecular simulations

This accelerates pharmaceutical research while reducing development cycles.

Substrate

Semiconductor simulation company Substrate achieved multi-fold runtime improvements in computational lithography.

Larger chip simulations can now be completed significantly faster.

WPP

Global advertising company WPP used AlphaEvolve to optimize campaign prediction models.

Results included:

• Up to 10% higher prediction accuracy

• Up to 7% better recommendation performance

This demonstrates AlphaEvolve's value beyond traditional engineering applications.

Google Is Using AlphaEvolve Internally

Google emphasized that AlphaEvolve is not merely an experimental research project.

It has become an important internal optimization engine.

According to Google DeepMind, AlphaEvolve has already improved:

TPU Design

Optimized silicon layouts for future Tensor Processing Units.

Google Spanner

Reduced write amplification by:

• 20%

Storage Optimization

Lowered software storage requirements by:

• Nearly 9%

Natural Disaster Prediction

Improved prediction accuracy across:

• 20 disaster risk categories

• Approximately 5% higher predictive accuracy

Quantum Computing

Discovered quantum circuits delivering:

• 10 times lower error rates

These internal deployments demonstrate that AlphaEvolve is already contributing to Google's own infrastructure before broader customer adoption.

From Productivity Assistant to Discovery Engine

Pushmeet Kohli, Chief Scientist at Google Cloud and Vice President of Science at Google DeepMind, described AlphaEvolve as a major evolution in artificial intelligence.

Rather than simply accelerating software development, AI is now becoming capable of discovering entirely new computational approaches.

This shift could fundamentally change research and engineering across industries.

Instead of replacing engineers, AlphaEvolve expands the search space humans can realistically explore.

The final decisions remain with developers.

Getting Started with AlphaEvolve

Google has designed AlphaEvolve to require only two primary inputs.

1. Seed Program

Developers provide the existing implementation.

Specific sections of code can be marked for optimization.

2. Evaluation Script

Organizations create a deterministic evaluation program that:

• Compiles candidate solutions

• Executes tests

• Scores results

• Returns measurable performance metrics

The optimization cycle operates continuously:

1. AlphaEvolve proposes new code.

2. Client infrastructure evaluates it.

3. Scores return to AlphaEvolve.

4. Better algorithms continue evolving.

This iterative feedback loop allows organizations to discover increasingly optimized solutions over time.

Why This Matters

Optimization has traditionally required years of specialized expertise, manual experimentation, and extensive trial and error.

AlphaEvolve introduces a fundamentally different approach.

Rather than asking engineers to manually search for better implementations, it allows AI to autonomously explore vast computational possibilities while humans define objectives and validate outcomes.

The breadth of early customer deployments suggests that this technology is not limited to one industry. Whether improving warehouse logistics, accelerating AI model training, designing semiconductors, optimizing quantum circuits, enhancing scientific simulations, or refining marketing models, AlphaEvolve has demonstrated measurable improvements across a remarkably diverse set of domains.

Its availability through the Gemini Enterprise Agent Platform also lowers the barrier for organizations looking to incorporate AI-driven algorithm optimization into existing workflows.

As enterprise AI continues to evolve, tools like AlphaEvolve signal a shift from AI as a coding assistant toward AI as an active partner in scientific discovery, engineering innovation, and large-scale optimization. Organizations willing to embrace this new paradigm may unlock performance gains that were previously beyond the reach of conventional software development methods.