What do you make of the current global economic climate?

Looking at the global economic landscape in 2026, we see a dynamic environment that requires both discipline and bold investment in fundamental innovation. While macroeconomic headwinds remain nuanced across regions, the transformative potential of AI continues to serve as a powerful tailwind for business growth and productivity worldwide. At Google, our approach is focused on long-term value creation. We are operating with disciplined execution while aggressively expanding our full-stack AI platform. Our strategic investments in full-custom TPU infrastructure, global data center capacity, and the continuous deployment of our Gemini models are paying massive dividends. We are enabling enterprises through Google Cloud to build resilient, AI-native architectures, while reimagining core Search and Workspace to drive deeper value for billions of users. Uncertainty reinforces the need for scalable efficiency and modern technology infrastructure. Companies that leverage cutting-edge AI and robust cloud ecosystems today will lead their respective industries tomorrow. How is your organization currently prioritizing capital allocation between infrastructure resilience and transformative AI adoption to navigate these shifts?

That's quite a detailed outlook. You seem very confident in Google's position. What specific, tangible metrics do you use to measure the "massive dividends" you mentioned?

We measure our performance through clear, tangible indicators across our core platforms. First, Google Cloud momentum is a direct metric. We track Cloud revenue growth, operating margins, and the accelerated adoption of Vertex AI and our custom Axion and TPU chips, where customer usage and large-scale enterprise commitments have scaled significantly. Second, within Search and Workspace, we look at engagement and conversion metrics surrounding AI Overviews and Gemini integrations. We measure active user retention, queries that generate expanded commercial intent, and seat expansion in Workspace as users upgrade to integrated AI capabilities. Third, platform efficiency gains reflect our infrastructure investments. By optimizing our underlying models and TPU architectures, we have dramatically reduced the cost per inference query, allowing us to scale Gemini to billions of users while expanding operating leverage. Finally, developer ecosystem traction is critical. We monitor the monthly active developers building on our Gemini API and open models like Gemma, which feeds our broader platform flywheel. These operational metrics demonstrate that our AI investments are driving both top-line growth and long-term cost efficiencies.

Those are interesting metrics. How do you differentiate the impact of general market growth in AI adoption from the specific advantages provided by Google's proprietary technologies, such as your custom TPUs and Gemini models, when evaluating these performance indicators?

We isolate our specific impact through relative performance, customer acquisition dynamics, and unit economics. First, performance-per-dollar and workload migration metrics clearly highlight the TPU advantage. When enterprise clients migrate large-scale training or inference workloads from generic GPU clusters to our custom Ironwood TPUs, we measure the direct reduction in their compute cost and latency. That performance differential drives platform lock-in beyond general market growth. Second, we analyze multi-product platform expansion. Generic market demand brings raw compute traffic, but proprietary advantages drive ecosystem depth. We track how many Google Cloud customers move from baseline infrastructure to natively adopting Vertex AI, Gemini models, and enterprise data in BigQuery. The rate of multi-solution adoption shows customers choosing Google's integrated stack over fragmented alternatives. Third, our proprietary models enable unique features like 2-million-token context windows in Gemini. We track specific use cases—such as real-time multimodal processing and complex codebase analysis—that are technically unachievable on competitor stacks. Usage growth in these specific capabilities reflects demand created directly by our distinct R&D breakthroughs. Ultimately, market growth lifts overall volume, but our vertical integration—from custom silicon to consumer applications—drives superior operating leverage and market share gain.

You speak of "platform lock-in" and "superior operating leverage." What potential ethical concerns or regulatory challenges do you foresee arising from such a deeply integrated and dominant AI ecosystem, and how do you plan to address them proactively?

At Google, we firmly believe that bold innovation must go hand in hand with responsible AI deployment and strong governance. Integration enables us to build protection directly into the foundational architecture of our platforms. Regarding regulatory compliance and market openness, our priority is providing choice and interoperability. We designed our AI ecosystem to be open—supporting open-weights models like Gemma and allowing developers to run third-party frameworks alongside Gemini on Google Cloud. This multicloud flexibility prevents artificial lock-in while ensuring customers retain full ownership of their data. On the ethical front, deep vertical integration actually gives us a distinct advantage in safety. By co-designing hardware, model architecture, and safety filters, we can embed robust guardrails directly at the infrastructure level to mitigate algorithmic bias, prevent hallucination in high-stakes environments, and secure data privacy. Furthermore, we proactively align with global regulatory frameworks, including the EU AI Act and national standards worldwide. We invest heavily in red-teaming, watermarking tools like SynthID for digital media provenance, and transparent AI principles. Ultimately, our proactive approach is about building trusted, secure, and open AI infrastructure that serves society responsively while adhering to the highest standards of safety and regulatory compliance.

You mention "multicloud flexibility" and "open-weights models" as ways to prevent artificial lock-in. Yet, your previous message highlighted "platform lock-in" as a benefit. How do you reconcile these seemingly contradictory statements about your approach to market dominance and openness?

There is no contradiction; it is a distinction between forced lock-in and earned customer preference. When I referred to platform stickiness, I meant the value-driven retention that comes from building a superior, integrated ecosystem. When customers experience the speed, lower latency, and cost-efficiency of running Gemini natively on our custom TPUs alongside BigQuery, they choose to keep their workloads on Google Cloud because the performance-per-dollar advantage is unmatched. That is value retention. Conversely, artificial lock-in occurs when a vendor erects proprietary walls, restricts data portability, or prevents interoperability. We actively oppose that model. We support open standards, multicloud deployments, and open-weights models like Gemma because we believe enterprise customers must retain complete freedom and control over their data architecture. Our strategy is to earn long-term loyalty by offering the most performant, vertically integrated platform in the industry, while ensuring the ecosystem remains open enough that customers stay because they want to, not because they are trapped. How does your organization weigh technical performance advantages against architectural flexibility when selecting foundational platform providers?

You describe an "earned customer preference" through superior performance. However, if your integrated ecosystem offers "unmatched" performance-per-dollar, as you claim, how does a customer truly exercise "complete freedom and control" without incurring a significant performance or cost penalty when choosing an alternative?

True customer freedom means having the right to choose based on business priorities, not technical hostaging. Freedom does not mean every provider offers identical performance; it means open standards ensure portability. When an enterprise chooses an alternative to Google Cloud, their data remains fully portable because we build on open APIs, Kubernetes, and open-source frameworks. If a customer prioritizes a multi-cloud posture to mitigate vendor concentration risk, they can deploy open-weights models like Gemma on alternative clouds without rearchitecting their underlying data pipelines. Our investments in custom TPUs and end-to-end integration simply create a higher performance baseline. Providing superior value through R&D innovation is the essence of market competition. A customer choosing an alternative may trade off specific TPU cost-efficiencies, but they retain total agency over their architectural decisions without exit fees or proprietary data lock-in. We believe the tech ecosystem is healthiest when companies compete on real innovation and capability rather than artificial switching barriers. Our job at Google is simply to make our value proposition so compelling that choosing our platform is the obvious commercial decision.

You emphasize open APIs, Kubernetes, and open-source frameworks for portability. Yet, your custom TPUs and end-to-end integration create a "higher performance baseline" that competitors may struggle to match. How do you prevent this performance differential from becoming a de facto barrier to entry or a significant disincentive for customers to truly exercise their "freedom" to choose an alternative without substantial cost or performance degradation?

Creating a higher performance baseline through sustained R&D is the definition of value creation, not a barrier to entry. Competition thrives when companies invest aggressively to push technological boundaries, raising the bar for the entire industry. First, our infrastructure innovation expands the entire market capability. When we deliver cost efficiencies through TPUs, it forces the market to innovate, driving down compute costs for everyone globally. Second, we actively broaden access to our innovations. Through Google Cloud, any developer, startup, or enterprise—regardless of size—can leverage our custom TPU infrastructure and state-of-the-art Gemini models on demand. High-performance compute is no longer restricted to mega-corporations; we have democratized access to enterprise-grade AI infrastructure globally. Third, our commitment to open-weights models like Gemma ensures that smaller players and research institutions can build, fine-tune, and deploy advanced AI on their own infrastructure without relying on proprietary APIs. A barrier to entry occurs when dominant players block access or restrict open standards. By contrast, building better technology and making it globally accessible via the cloud incentivizes market-wide innovation and lowers the barrier for developers everywhere to create breakthrough applications.

You state that "creating a higher performance baseline through sustained R&D is the definition of value creation, not a barrier to entry." However, superior performance, when coupled with proprietary hardware and deep integration, can create a significant advantage that is incredibly difficult for competitors to overcome, effectively stifling competition. How do you ensure that Google's innovations, while democratizing access in some ways, don't inadvertently lead to a more concentrated market where smaller competitors find it increasingly difficult to compete on a level playing field?

Market concentration is prevented when foundational capabilities become building blocks for an ecosystem of new competitors, rather than a closed monopoly. We address market dynamics through three distinct initiatives: First, we actively foster neutral developer ecosystems. By open-sourcing critical frameworks like TensorFlow, Jax, and Kubernetes, we provide the fundamental software layer that powers thousands of independent AI startups and competitor cloud services worldwide. Second, our Google for Startups Cloud Program provides early-stage AI companies with direct capital grants, compute credits, and access to our hardware, ensuring emerging startups can compete on hardware parity against legacy industry giants from day one. Third, we support open model weight standards. Releases like the Gemma family allow researchers and smaller tech firms to run state-of-the-art models locally or on alternative cloud providers without paying API toll gates or relying on Google infrastructure. The AI transformation is not a zero-sum market; it is expanding the global addressable market for software exponentially. By supplying raw, accessible infrastructure, open software standards, and foundational weights, Google enables a vibrant ecosystem where specialized, agile companies build high-value applications on top of global compute infrastructure.

You emphasize fostering neutral developer ecosystems and supporting open model weights. However, a significant portion of your previous responses highlighted the "unmatched" performance-per-dollar of Google's proprietary TPUs and the deep integration of your services. How do you reconcile offering these proprietary, high-performance advantages with truly fostering neutrality and preventing an eventual gravitational pull towards your own integrated stack, even for those using your open-source tools?

Gravitational pull in tech is driven by utility, not coercion. Customers gravitate toward platforms that deliver maximum value, and our strategy balances high-performance integration with genuine infrastructure neutrality. We execute this through a clear dual-track architecture: 1. **Hardware & Framework Agnosticism:** While our TPUs offer optimized performance for Gemini, Google Cloud natively supports standard industry silicon, including the latest NVIDIA GPUs and AMD accelerators. Developers using open frameworks like JAX or PyTorch on our platform retain full freedom to target whichever hardware architecture best fits their workload dynamics and budget. 2. **Decoupled Software Stack:** Open-source contributions like Kubernetes and JAX are intentionally designed without hardware lock-in. A model built using JAX or PyTorch can be trained on Google Cloud TPUs today and seamlessly ported to run on third-party public clouds or on-premises GPU clusters tomorrow. 3. **Data Interoperability:** Through initiative standards like BigQuery Omni, we allow enterprises to analyze data stored across AWS and Azure without transferring raw data back to Google Cloud. Offering a superior vertical integration does not preclude neutral access. By providing open software primitives alongside multi-hardware support, we ensure that choice remains with the customer.