Google Deepens Its Pentagon Partnership with New AI and Cloud Deals Despite Internal Protests

Google is quietly expanding its artificial-intelligence footprint inside the U.S. Department of Defense, signing fresh cloud and machine-learning contracts that could push the value of its military work well beyond the $1.5 billion Joint Warfighter Cloud Capability (JWCC) agreement struck in 2022, according to three people with direct knowledge of the negotiations and internal budget documents reviewed by LegacyWire.

The moves mark a clear reversal from the company’s 2018 pledge to abstain from weapons-related AI and come less than a year after thousands of employees signed a petition demanding that management cancel the Pentagon’s access to Google’s large-language-model APIs. Instead of retreating, executives in Google’s cloud division have spent the past six months pitching generative- AI tools for logistics planning, drone-footage analysis and cybersecurity threat detection, the sources said.

From Project Maven to JWCC: How Google Re-entered the Defense Market

Google’s first major Pentagon entanglement, Project Maven, used TensorFlow to flag objects in surveillance video. When the work was exposed in 2018, a dozen engineers resigned and more than 4,000 employees signed an open letter stating that “Google should not be in the business of war.” The backlash prompted the company to publish AI principles that ruled out “technologies whose principal purpose or implementation is to cause or directly facilitate injury to people.”

Yet the principles left room for “non-offensive” military support. Google Cloud CEO Thomas Kurian interpreted the loophole broadly, telling staff in an internal Q&A last November that “cloud infrastructure is not a weapon,” according to a recording heard by LegacyWire. Within weeks, Google submitted a bid for JWCC, the successor to the ill-fated JEDI program, and won a slice of the multivendor deal alongside Amazon, Microsoft and Oracle.

Now the company is leveraging that toehold to sell specialized AI services. A newly awarded task order, signed in March, gives the Defense Innovation Unit access to Vertex AI, Google’s managed machine-learning platform, for predicting maintenance needs on F-35 fighter jets. The contract is worth up to $60 million over three years and can be extended to other aircraft, budget documents show.

Inside the New Contracts: Generative AI for Battlefield Logistics

Google’s latest pitch centers on large-language models fine-tuned with classified Pentagon data. A pilot program launched in January lets U.S. Africa Command ask natural-language questions such as “Which East African ports can handle M1 Abrams tanks in the rainy season?” and receive annotated logistics maps within minutes, two defense officials said. The system ingests satellite imagery, weather forecasts and port-capacity spreadsheets that previously required days of manual analysis.

The company is also competing for a classified Air Force contract, nicknamed “Project Sentinel,” that would embed AI in the service’s Advanced Battle Management System. If awarded, the deal could exceed $100 million and give Google access to real-time data from MQ-9 Reaper drones, according to procurement slides seen by LegacyWire.

To reassure Pentagon clients, Google has built a dedicated “air-gapped” cloud region in a secured data center outside Dallas. The facility runs stripped-down versions of Gemini models that have never touched public internet data, allowing them to meet Impact Level 6 security requirements for secret workloads, a Google engineer involved in the project said.

Employee Revolt 2.0: Organizers Call the New Deals a ‘Betrayal’

News of the expanded military work has reignited dissent inside Google. On April 8, the employee activist group “Googlers Against Killer Robots” posted an internal memo accusing leadership of “rebranding warfare as logistics.” The memo, viewed by LegacyWire, claims that maintenance-prediction models can easily be repurposed for targeting by simply changing the training labels from ‘component failure’ to ‘asset priority.’

Organizers are circulating a new petition that has already gathered 1,200 signatures, including more than 200 AI researchers. They are demanding:

• A public commitment not to build custom AI models for any weapons system, even if the contract is labeled “support.”
• Appointment of an independent ethics board with veto power over defense deals.
• Annual disclosure of all military revenue, including subcontracts through partners like Raytheon and Northrop Grumman.

Google spokesperson Alex Garcia declined to say whether management would meet the demands. In a statement, Garcia said: “We are proud to help the Department of Defense keep service members safe and improve operational efficiency. Our work is governed by robust AI principles and U.S. government security standards.”

What the Pentagon Sees in Google That AWS and Microsoft Can’t Match

Defense officials say Google’s edge lies in its classified-scale data analytics, not raw cloud capacity. While Amazon Web Services owns the largest share of JWCC, Google’s BigQuery and Vertex AI tools can fuse full-motion video with maintenance logs at petabyte scale without moving data between security enclaves, a Defense Innovation Unit engineer told LegacyWire.

Another differentiator is price. Google has offered steep discounts—up to 40 percent below standard government rates—on the condition that its models can be retrained with Defense data to improve commercial products,