At Google Cloud Next in Las Vegas, Thomas Kurian, chief executive of Google Cloud, responded to a question from Asharq Al-Awsat about attacks on hyperscale cloud data centers amid regional tensions by moving quickly beyond physical protection. The issue, he suggested, is no longer simply how to defend infrastructure, but how to ensure customers are not left dependent on one location when disruption occurs.
Kurian said Google Cloud has managed through global conflict scenarios for many years and has built not only physical safeguards, but also a private global network with extensive redundancy linking its data centers.
The company can shift workloads away from affected locations and replicate them globally because its cloud regions operate as a unified and consistently synchronized architecture, he explained. For customers, he argued, that means they are not tied to a single physical site.
His response moved the discussion from infrastructure protection toward a broader strategic question: whether cloud architecture itself has become part of business continuity planning.
From experimentation to operations
That framing also offered one of the clearest ways to understand Google Cloud’s broader message at Next 2026. Throughout the event, attended by more than 30,000 participants, the company sought to underscore that enterprise AI is moving from experimentation into what it calls the agentic enterprise.
Google Cloud said roughly 75 percent of its customers already use its AI-powered products. Some 330 customers processed more than one trillion tokens over the past 12 months, while more than 35 customers surpassed 10 trillion tokens. The company also said its frontier models now process more than 16 billion tokens per minute, up from 10 billion in the previous quarter.
The purpose of those figures was to signal that AI is no longer a side experiment, but an operational layer companies want to use across their businesses.
Integration and openness together
Perhaps most revealing in the private Q&A with Kurian was what he suggested about where competition is heading. He argued that Google Cloud’s distinguishing advantage lies in combining proprietary chips, frontier models, infrastructure and tools, allowing the company to optimize the entire stack, from computing power to the efficiency of AI agents.
The broader argument was that the next phase of AI will not be determined only by who has the strongest model, but by who can design the broader system around it most effectively. At the same time, Kurian paired this with another point equally important to enterprise customers: openness. He stressed that he does not expect companies to rely exclusively on Google Cloud and said the company has deliberately kept its architecture open.
He pointed to support for multiple models, Google’s own chips, close collaboration with NVIDIA, compatibility with different data platforms and partnerships with third parties in security.
That matters because enterprises want the efficiency of deep integration without being locked into a closed environment. Google Cloud is signaling it can provide a vertically integrated stack while still operating across diverse enterprise technology environments.
Sovereignty at the forefront
Sovereignty also emerged as a major theme. Asked whether European customers would receive the full product offering, Kurian said the broader product is already available in Europe in compliance with sovereignty regulations, hosted across multiple sites including Germany, France, Belgium, the Netherlands, Finland and the United Kingdom.
Though the answer focused on Europe, its significance extends beyond the continent. Enterprise customers, including Saudi Arabia, increasingly want advanced AI services without giving up control over where their data is hosted and processed. That is not a side issue, but part of the architecture of trust itself.
Connectors make the difference
Kurian also addressed another practical issue tied to one of enterprise AI’s real bottlenecks.
Asked who would build the connections between Gemini Enterprise and the many applications companies already use, he said Google Cloud is doing so itself. The company already offers more than 100 connectors covering document repositories, software-as-a-service applications and databases.
He added that Google Cloud also provides a framework for building connectors and supports standards such as Bring Your Own MCP for custom-built systems.
The significance of that point lies at the heart of why many enterprise AI projects struggle: a model may be impressive in isolation, but it only becomes useful when it connects to where work actually happens — documents, business applications, records and databases.
AI and defense
The cybersecurity portion of the discussion was no less significant.
Kurian said Google Cloud recognized some time ago that as models improve at understanding software, malicious actors would use them to analyze code, discover vulnerabilities and attack systems. In his view, the response must also be driven by AI.
He described one layer focused on analyzing and repairing a company’s own code, pointing to a new model called Code Defender that helps fix vulnerabilities.
A second layer focuses on external threats, including threat hunting and threat intelligence. He pointed to Dark Web Intelligence announced at the conference, saying it can prioritize the threats customers should defend against with about 90 percent accuracy.
He also linked this logic to Google Cloud’s acquisition of Wiz, describing a layered model in which a red agent probes systems for weaknesses, a blue team identifies the needed fixes and a green layer carries out remediation.
