Oracle Struggles with Internal AI Adoption Despite Billions Spent on Infrastructure

Despite spending billions of dollars providing AI infrastructure for external customers, Oracle executives revealed that internal rollout of AI tools like ChatGPT Enterprise and Codex faced significant friction. While AI accelerated code generation from quarters to a single week, it exposed severe bottleneck shifts in testing, deployment, and model cost management. Oracle's experience highlights a critical reality in enterprise AI adoption: accelerating code generation does not automatically translate to faster product releases. Companies must overhaul downstream software engineering pipelines and institute granular usage controls to prevent ballooning LLM API costs. Oracle implemented internal cost-monitoring mechanisms after discovering that top-tier models like GPT-6 Astra cost 2.5 times more than standard models. Additionally, early testing of Anthropic's Mythos Preview security tool flagged more code vulnerabilities in two weeks than in the previous entire year, but 60% to 70% were false positives requiring manual verification.