AI Hardware Engineering Intern

Job Description: About the Job Intel is seeking exceptional Computer Science, Electrical Engineering, Computer Engineering, and AI/ML students to join an AI Hardware Engineering internship focused on the future of AI inference and AI infrastructure with hardware/software co-design expertise.

This internship approaches AI infrastructure as a hardware/software co-design challenge. The intern will analyze how model architecture, inference behavior, memory movement, data-center constraints, and software stacks influence end-to-end customer outcomes, including throughput, latency, time-to-first-token, tokens per second, power consumption, utilization, total cost of ownership, and deployability.

The successful candidate will be technically hands-on, capable of independent research and analysis, and comfortable working within a highly technical team of researchers and engineers addressing complex AI hardware and infrastructure challenges.

What You Will Do

• Conduct technical research on the next generation of AI infrastructure, including CPUs, GPUs, AI accelerators, ASICs, memory systems, networking, storage, rack-scale architecture, power delivery, cooling, and cluster management.
• Analyze GenAI and agentic AI workloads to understand their hardware implications, including transformer architecture based inference, mixture-of-experts architectures, multimodal models, real-time interactive inference, and distributed serving.
• Evaluate the hardware behavior of AI inference pipelines, including the distinct demands of prompt processing/prefill and token generation/decode, memory capacity, memory bandwidth, KV-cache management, compute utilization, interconnect bandwidth, and latency. For example, cluster architecture around optimizing both prefill and decode through hardware, memory, and system co-design.
• Identify product and system gaps that may limit Intel's ability to serve AI customers effectively, including hardware capabilities, developer tooling, performance libraries, reference architectures, deployment models, observability, and ecosystem enablement.
• Expertise in modern high-performance CPU, GPU, or AI accelerator architecture and hardware/software interactions.
• Support Silicon HW programs through multiple phases of the silicon lifecycle, with end-to-end chip execution from concept to silicon.
• Good understanding with firmware codebases, compilers, kernels, serving stacks and model architectures.
• Familiar with driving silicon design for subsystems and/or top-level functions with ASICs and SOCs from an architecture specification to production silicon including memory subsystems and interconnects: scale up, scale out, scale across, power management , liquid cooling at rack scale level, deep system level observability & telemetry and dependent impact into Gen AI and LLM level performance/tco/execution KPIs
• Develop technical recommendations for what Intel should build, optimize, enable, partner on to meet AI customer requirements based on deep understanding of customer use cases, workload performance, and ML systems research.
• Benchmark and compare relevant AI hardware and infrastructure approaches across performance, power efficiency, cost, programmability, memory hierarchy, scalability, ecosystem maturity, and workload fit.
• Develop technical models, simulations, prototypes, or analysis tools to evaluate hardware/software trade offs for AI workloads.
• Investigate system bottlenecks across compute, memory, networking, storage, and software orchestration, and propose methods to improve end-to-end AI workload performance.
• Work with Intel technical teams and external ISVs to collect data on model requirements, inference bottlenecks, deployment challenges, hardware optimization opportunities, and customer priorities.
• Engage with AI software ecosystem partners, including model providers, inference-engine developers, cloud providers, framework teams, and ISVs, to understand how their stacks interact with underlying hardware.
• Good understanding of modern AI software layers, including PyTorch, Hugging Face Transformers, vLLM, SGLang, model serving, TensorRT-LLM, Triton, Kubernetes, and Neocloud design efficiencies.
• Present findings through architecture diagrams, benchmark analyses, technical reports, design recommendations, and executive-ready presentations.

…