Scale AI Names Google Cloud Veteran deSouza CEO for Enterprise Push

Scale AI named former Google Cloud executive Francis deSouza its permanent chief executive officer effective August 10, 2026, as the company accelerates its applications business toward overtaking core data work. He succeeds interim CEO Jason Droege, who has led since founder Alexandr Wang departed for Meta after the social giant’s $14.3 billion investment for a 49 percent non-voting stake.

The move installs a three-decade enterprise operator at a firm now valued near $29 billion and already powering frontier labs, Fortune 500 companies and U.S. government programs. The timing aligns leadership change with a commercial pivot already under way: applications revenue is climbing fast enough to challenge the original data engine inside roughly a year and a half.

The August 10 Hand-off and Board Mandate

Scale’s board made the appointment official on July 30. Droege, formerly chief strategy officer and an Uber Eats co-founder, will stay on for months to smooth the transition. deSouza arrives from the Google Cloud COO and security products role, where he drove enterprise AI and cloud adoption.

• June 2025: Meta closes $14.3 billion deal; Wang exits day-to-day role for Meta Superintelligence Labs while remaining Scale chairman and board member.
• June 2025 onward: Droege serves as interim CEO.
• July 30, 2026: Scale announces deSouza; start date set for August 10.
• Coming months: Droege supports hand-over.

In the Scale AI board appointment announcement, Wang called deSouza the right steward for a company that has become essential infrastructure. deSouza said his focus will be getting solutions into more businesses and governments and delivering the best data for AI labs through provable outcomes.

The multi-month overlap is deliberate. Droege carried the firm through the post-deal period when some frontier labs paused or reduced work over data-confidentiality worries. Keeping him in place gives deSouza time to absorb customer relationships, government contract cadence, and the internal split between the still-profitable data unit and the faster-growing applications side.

Wang Stays Chairman While Meta Holds the Stake

Wang founded Scale at 19 and built it into a backbone supplier of training data and evaluations. The Meta deal gave the company capital and left it independent, with Meta holding no voting power. Wang now leads Meta’s superintelligence efforts, including Wang’s health AI focus at Meta, yet keeps the Scale chair seat.

Welcome @fdesouza to @scale_AI as CEO! I started Scale a decade ago at 19, and it has grown to be the backbone of the AI world, powering frontier labs, Fortune 500 companies, and the US Government. Francis is an exceptional leader and the right steward. After spending a lot of time together, I have no doubts he’ll take Scale to greater heights. Thank you @jdroege for your leadership as interim CEO

Wang posted that welcome on X, drawing more than 900 likes and 164,000 views. The public endorsement underscores continuity even as operational leadership professionalizes. Some frontier labs had paused or reduced work after the Meta stake over data-confidentiality worries; the rebound under Droege shows the dual-engine model can absorb the friction.

Independence matters to the customer base that remains. Labs that buy training data and evaluations want assurance that a large strategic investor cannot steer product roadmaps or access competitive work. The non-voting structure and Wang’s continued chairmanship are the formal answers to that concern. deSouza’s task is to keep those assurances credible in day-to-day operations.

deSouza’s Path From Startups to Genomics to Cloud

deSouza brings a résumé built for Scale’s next stage. He holds bachelor’s and master’s degrees in electrical engineering and computer science from MIT. He founded two companies later acquired by Microsoft and Symantec, then rose to president of products and services at Symantec.

At Illumina he ran one of the world’s largest genomic data platforms. At Google Cloud he saw firsthand how large organizations actually deploy AI at scale. That combination matches Scale’s claim to sit at the intersection of frontier model work and real-world production systems.

The through-line is regulated, high-stakes data. Genomics platforms and enterprise security products both demand auditability, global compliance, and measurable outcomes. Those same constraints now define how governments and Fortune 500 buyers evaluate AI vendors. deSouza has already operated inside that constraint set at two prior companies of significant scale.

Data Engine Still Runs While Applications Double

Scale’s original business supplies high-quality data and human evaluations that train frontier models. That unit grew strongly in 2025, turned profitable in the second half, and added robotics customers with more than 150,000 hours of physical-AI data. Its Safety Evaluation and Alignment Lab (SEAL) published hundreds of evaluations and new leaderboards that labs now treat as reference points.

The newer applications side is the growth story. It more than doubled revenue in the second half of 2025 and is expected to double again. Company statements put the applications business on track to overtake the data business within roughly 18 months. Enterprise closed its strongest bookings quarter ever in Q4 2025. Overall the firm logged over $1B in new business in 2025, with nearly half arriving in the final quarter.

• 2025 new business: well over $1 billion across data and applications
• Applications trajectory: more than doubled H2 2025; expected to double again in 2026
• Data unit: profitable second half; robotics expansion
• Headcount: more than 500 hires; highest offer-acceptance rate in company history

deSouza told Axios that organizations understand AI’s potential but need help realizing it responsibly and reliably. He sees labeling, deployment and infrastructure management all continuing to expand as new use cases appear. That view treats applications as additive, not a replacement for the data foundation.

The headcount surge and record offer-acceptance rate matter for execution. Applications work is delivery-heavy: integrations, change management, and ongoing evaluation loops all require people who can operate inside customer environments. Hiring velocity in 2025 suggests the pipeline is already sized for the next doubling cycle.

Enterprise Logos and Government Contracts Stack Up

Recent enterprise wins include BP, Mayo Clinic and Allianz. Public-sector work delivered its best U.S. quarter in Q3 2025 with two nine-figure awards. The Department of Defense (referred to in company materials as Department of War in some contexts) awarded Scale two major contracts totaling nearly $200 million. Scale continues to lead Project Thunderforge, a flagship effort to put AI agents into military mission planning alongside Anduril and Microsoft.

International public-sector revenue doubled in 2025. The Middle East is the fastest-growing region. In Qatar the firm has deployed applications for education, the Council of Ministers and museums, working with partners that include Google Cloud, Microsoft and AWS. Scale projects the international business will double again in 2026. Governments are moving past pilots and buying full-stack sovereign AI infrastructure they can verify and trust.

• Enterprise logos: BP, Mayo Clinic, Allianz
• U.S. public sector: best quarter in Q3 2025; two nine-figure awards
• Defense: nearly $200 million across two major contracts; Project Thunderforge lead with Anduril and Microsoft
• International: doubled in 2025; Middle East fastest-growing; Qatar deployments across education, ministers, museums

That same enterprise demand for production systems shows up in broader market signals, including the enterprise demand for agentic AI infrastructure that Google Cloud has documented. Scale’s dual offering positions it to capture both the data that trains models and the applications that put them to work on high-stakes decisions.

Sovereign infrastructure buys differ from pilot projects in contract length and verification depth. Buyers want systems they can audit, retrain, and run under their own security boundaries. Scale’s pairing of proprietary data pipelines with deployable applications is built for that shift, and the 2025 public-sector numbers show the shift is already funded.

Why an Enterprise Operator Fits the Reliability Mission

Scale updated its mission in 2025 to “develop reliable AI systems for the world’s most important decisions.” The phrasing is deliberate. After years of hype and pilot fatigue, buyers want systems that perform under real conditions, with audit trails and measurable outcomes. deSouza’s background running regulated, data-heavy businesses at Illumina and security-focused cloud work at Google maps directly onto that requirement.

On X, deSouza wrote that Scale works with top labs to push the frontier while helping enterprises and governments deploy AI they can verify and trust. Crowd conversation around the hire quickly framed it as moving Scale from demo supplier to infrastructure seller. Earlier public criticism that labeled some of Scale’s human-data operations a “sweatshop” model only sharpens the current emphasis on trust and production reliability; a CEO who has run global, compliance-heavy platforms is the logical next steward.

The Meta stake supplies capital and a high-profile chairman without day-to-day control. The apps engine is already scaling. Government and enterprise logos are landing. deSouza’s job is to convert that momentum into durable, multi-year production contracts while keeping the data flywheel spinning for the labs that still need it.

The Dual Engine Keeps Labs and Buyers Aligned

Scale’s structure now runs on two linked engines. The data and evaluation business still feeds frontier labs with labeled training sets, SEAL leaderboards, and robotics hours. The applications business takes similar tooling into enterprises and governments that need production systems, not research inputs.

Those engines reinforce each other. Labs that treat SEAL evaluations as reference points create a quality signal that enterprise buyers can cite when they procure. Enterprise and government deployments, in turn, surface new edge cases and domain data that flow back into the evaluation stack. The post-Meta rebound under Droege showed the model can hold even when some lab relationships cool temporarily.

deSouza inherits both sides already in motion. Applications more than doubled in the second half of 2025 and are expected to double again; the data unit is profitable and still adding physical-AI volume. His stated priority (solutions into more businesses and governments, plus provable outcomes for labs) tracks the dual mandate rather than choosing one engine over the other.

What deSouza Must Convert After August 10

By the August 10 start date the raw materials are already on the table: over $1 billion in 2025 new business, a strongest-ever enterprise bookings quarter in Q4, two defense awards near $200 million, international revenue that doubled once and is projected to double again, and more than 500 recent hires at a record acceptance rate.

Conversion means turning that booking momentum into multi-year production contracts with the audit trails and measurable outcomes the 2025 mission statement promises. It also means protecting the data flywheel so frontier labs continue to treat Scale as essential infrastructure even while Meta holds a large non-voting stake and Wang chairs from a parallel role.

Droege’s multi-month transition window is the bridge. deSouza’s enterprise and regulated-industry background is the filter through which those contracts will be shaped. The board’s July 30 appointment sets the clock; the commercial numbers already on the books set the bar.

Frequently Asked Questions

When does Francis deSouza officially start as Scale AI CEO?

He starts August 10, 2026. Jason Droege remains through a multi-month transition period to support the hand-over after serving as interim CEO since June 2025.

What was Alexandr Wang’s role after the Meta investment?

Wang left day-to-day leadership of Scale to become chief AI officer and head of Meta Superintelligence Labs. He retained the Scale board chairmanship and a board seat; Meta’s 49 percent stake is non-voting, leaving Scale independent.

How large was Meta’s investment in Scale AI?

Meta invested $14.3 billion for a 49 percent non-voting stake in June 2025, valuing Scale at approximately $29 billion at the time of the deal.

What experience does deSouza bring from Illumina and Google Cloud?

As Illumina CEO he grew revenue above $4.5 billion and expanded operations to more than 150 countries while running a major genomic data platform. At Google Cloud he served as COO and president of security products during a period of rapid enterprise AI and cloud growth.

Is Scale AI still primarily a data-labeling company?

Data labeling and model evaluation remain core and profitable, but the applications business more than doubled in the second half of 2025 and is projected to overtake the data business within about 18 months as enterprises and governments buy production systems.