From Model Labs to Infrastructure-First AI Strategies
The new infrastructure-first AI strategy is an enterprise approach where software companies prioritize long-term cloud computing investment, compute access, and integrations over building proprietary foundational AI models, treating AI as a core utility rather than a standalone product. This marks a shift from earlier experiments that focused on in-house model development toward large-scale AI infrastructure spending tied to hyperscale cloud providers. Instead of trying to own every layer of the stack, vendors now allocate capital to reserved compute, GPU capacity, and deployment services so AI workloads can scale reliably and predictably. The goal is to embed AI into standard business operations—finance, legal, tax, risk, HR—by ensuring secure, always-on infrastructure and pre-integrated services. As AI workload scaling becomes a board-level topic, this infrastructure-first stance shapes balance sheets, product roadmaps, and software vendor partnerships with platforms such as AWS, Azure, and other cloud ecosystems.
Snowflake’s USD 6B Bet: AI Compute on the Balance Sheet
Snowflake’s USD 6 billion (approx. RM27.6 billion) commitment to Amazon Web Services over five years is one of the clearest signs that AI infrastructure spending is now central to software economics. The agreement covers AWS Graviton compute and GPU-accelerated EC2 instances, which Snowflake will use for AI model training and inference while also powering its core data warehousing services. According to AWS, Snowflake has surpassed USD 7 billion (approx. RM32.2 billion) in lifetime sales through AWS Marketplace, tying infrastructure and distribution into a single cloud relationship. Market commentary notes that this multibillion-dollar cloud computing investment helps Snowflake secure dependable compute for agentic AI workloads that plan, retrieve data, and call tools across business applications. In practice, Snowflake is repositioning itself as “the platform for the AI era,” not by owning custom chips or models, but by reserving massive shared infrastructure capacity and tightening its software vendor partnership with AWS.
EY and Microsoft: AI as an Operational Cost, Not a New Product
Professional services are also shifting to an infrastructure-first enterprise AI strategy. EY and Microsoft have committed more than USD 1 billion (approx. RM4.6 billion) over five years to move AI from pilots into day-to-day operations, focusing on deployment rather than new foundational models. Their joint offer embeds Microsoft engineers and EY industry teams inside client operations to connect tools like Copilot and Azure AI Document Intelligence directly into live workflows. According to Microsoft, finance modernization work using their tools produced 95% faster lead times, while Azure AI Document Intelligence reduced manual workload by up to 90% on EY’s Global Tax Platform. These numbers may be self-reported, but they show how AI workload scaling is treated as a process and infrastructure problem. The large budget funds people, integration, and cloud capacity instead of bespoke models, reinforcing Microsoft’s position as the underlying AI infrastructure provider.
Law Firms’ Custom AI: Build the Experience, Rent the Infrastructure
While many software vendors avoid model-building, some law firms are investing heavily in custom AI front-ends while relying on outside technology partners for infrastructure. Kirkland & Ellis plans to spend USD 500 million (approx. RM2.3 billion) over three to four years on its own AI tools and services, funded from revenue of USD 10.6 billion (approx. RM48.7 billion). The platform is designed as a broad, firm-wide system rather than a patchwork of tools, with external companies helping to build the technology but barred from selling it to competitors. Other firms like Simmons & Simmons and Allen & Gledhill have taken more in-house routes, building platforms such as Percy and A&GEL using their own LLM teams and on-premise hosting for confidentiality. Yet even these “build” strategies lean on cloud or internal infrastructure stacks instead of training massive new models, reflecting a preference to control user experience and legal workflows while treating compute and core AI infrastructure as a shared, utility-style layer.
AI as a Utility: Vendor Lock-In and the New Cost Center
Across data platforms, consulting firms, and legal practices, a pattern is emerging: AI is being treated as an infrastructure cost center rather than the main product differentiator. Snowflake’s AWS commitment shows how software vendors now sign multiyear, multibillion-dollar deals to secure compute before revenue fully materializes, trading flexibility for priority access to GPUs and ARM compute. EY’s alliance with Microsoft frames AI as embedded support for finance, tax, risk, HR, and supply chain workflows, with Copilot rollouts measured in hundreds of thousands of users instead of experimental pilots. Even firms building custom tools are doing so on top of cloud or controlled infrastructure. This infrastructure-first mindset tightens vendor lock-in with providers like AWS and Microsoft, but it also gives enterprises predictable AI workload scaling. Competitive advantage is shifting from owning proprietary models to how reliably and securely vendors integrate AI into existing cloud computing investment and business processes.