Intel executives outlined a strategic pivot in enterprise artificial intelligence at The Six Five Summit: AI Unleashed 2026, stressing cost efficiency and measurable returns as critical drivers for adoption beyond initial pilot phases.
Speaking on Tuesday, Anil Nanduri, Intel’s Vice President of AI Products and Go-To-Market for its Data Center division, highlighted that enterprises are now prioritizing return on investment (ROI) over hardware-first approaches. The company’s stock has surged 255% over the past year, closing at $90.07 on August 24, 2026, before dipping 2.24% in after-hours trading.
Nanduri cited Uber as a case study, noting that the ride-hailing giant initially exceeded its annual AI token budget within months. The company later reduced costs and expanded deployments by improving algorithm efficiency, enhancing token observability for engineers, and selecting models tailored to specific tasks. He emphasized that not all AI tokens are equivalent, distinguishing between ultra-low latency requirements for real-time applications like fraud detection and batch processes with extended delivery windows.
The discussion also addressed the growing complexity of agentic AI systems, which now require end-to-end workflow optimization, orchestration, verification, and execution. Intel’s heterogeneous computing strategy spans four layers: a control plane using Xeon CPUs for orchestration and security, a data plane combining CPUs, GPUs, and specialized architectures, a network plane for cross-system communication, and a storage plane leveraging CPU-based solutions for databases and memory pooling.
Intel’s Crescent Island GPUs, designed with low-power DDR memory instead of GDDR or HBM, were highlighted for their cost and energy efficiency in PCIe card deployments. The company estimates that about 80% of enterprise AI workloads can be handled with open-weight models, while the remaining 20% require frontier-model intelligence. Additionally, more than 90% of publicly available data has already been accessed, underscoring the need for efficient data management and processing.
