Hot Chips 2026 (HC38) — A Symposium on High Performance Chips

Hot Chips 2026 (HC38) is the IEEE symposium on high-performance chips, running 23-25 August 2026 at Memorial Auditorium on the Stanford University campus in Palo Alto, California, with a virtual option. A Sunday tutorial day covers memory technology and RISC-V; Monday and Tuesday are single-track sessions on CPU, GPU, AI, memory, networking and interconnect, FPGA and automotive silicon, plus a keynote from Waymo's Daniel Rosenband on compute for autonomous driving. This edition is unusually heavy on hyperscaler custom silicon: Google presents its eighth-generation TPU family, Microsoft its MAIA 200 accelerator, Meta its custom AI chips and OpenAI its own chip programme, alongside NVIDIA Rubin, AMD Instinct MI400 and Intel's Crescent Island inference GPU.

Date

August 23–25, 2026

Location

Memorial Auditorium, Stanford University, Palo Alto, California

Format

Hybrid

Event Type

Conference

Status

Completed

Pricing

Paid

Organizer

Hot Chips (IEEE Symposium on High Performance Chips)

Added

Jul 14, 2026

Updated

Aug 22, 2026

• ✓Chip and system architects
• ✓Data center infrastructure leaders
• ✓AI platform and ML systems engineers
• ✓CTOs at semiconductor and hyperscale companies
• ✓Hardware performance engineers
• ✓Compiler and runtime engineers
• ✓Technology analysts and investors

For anyone sizing an AI build-out, this is where the next 18 months of silicon is disclosed by the architects who designed it rather than by a marketing team. Three days gets you directly comparable detail on NVIDIA Rubin, AMD Instinct MI400 and Intel Crescent Island in a single GPU session, the hyperscalers' custom accelerators in another, and the memory and Ethernet-fabric roadmaps that determine whether your cluster ends up compute-bound or bandwidth-bound. It is the cheapest available way to pressure-test a vendor's performance claims before a procurement cycle commits budget.

• ✓Keynote — 'Compute in Motion: Challenges of Autonomous Driving', Daniel Rosenband, Waymo
• ✓Tutorial day (23 Aug), Memory: HBM base dies on advanced logic (Samsung), HBM advanced packaging (SK hynix), 3D DRAM accelerators for generative inference (d-Matrix and Meta), and High Bandwidth Flash
• ✓Tutorial day (23 Aug), RISC-V: RVA23 profiles and adoption (SiFive), enterprise open source (Canonical), RISC-V interoperability with NVIDIA GPUs, and RISC-V for automotive (Infineon)
• ✓AI sessions 1 and 2: Google's eighth-generation TPU family, Microsoft MAIA 200, Meta's custom AI silicon, the Cerebras wafer-scale rack architecture, SambaNova's SN50 RDU and OpenAI on building its own chips
• ✓GPU session: NVIDIA Rubin, AMD Instinct MI400 series architecture and system design, and Intel Crescent Island for agentic AI inference
• ✓CPU sessions: Intel Diamond Rapids Xeon and Wildcat Lake, NVIDIA's Vera CPU, Arm's chiplet-based AGI server SoC, Fujitsu-MONAKA and the next IBM Z / LinuxONE processor with its AI inference chipset
• ✓Networking and Interconnect session: Broadcom's Thor Ultra AI/HPC Ethernet NIC, NVIDIA BlueField-4 and the Spectrum-X multiplane network architecture
• ✓Automotive session: Waymo's sensor-fusion processor and BOS Semiconductors' chiplet-based Eagle-N SoC
• ✓Memory session: Samsung LPDDR5X-PIM processing-in-memory for inference and XCENA's MX1 CXL computational memory device

• ★Vendors present their own unreleased silicon in architectural detail — Hot Chips is where the engineers who designed a chip disclose its microarchitecture, and it is frequently a part's first public technical disclosure.
• ★The hyperscaler custom-accelerator programmes are all in one room this year: Google's eighth-generation TPU family, Microsoft MAIA 200, Meta's custom AI silicon and an OpenAI talk titled 'You Can Just Build Things … Chips'.
• ★The GPU session puts NVIDIA Rubin, AMD Instinct MI400 (architecture and system) and Intel's Crescent Island back to back, which is as close to a directly comparable head-to-head as the industry gets.
• ★Run by IEEE volunteers on a university campus rather than by a vendor — there is no sponsor keynote, and the two conference days are a single peer-selected track everyone attends together.
• ★Small and consistently oversubscribed: in-person passes for 2026 sold out ahead of the event, while a virtual pass costs $210-255 and includes slide downloads and a Slack channel with the presenters.

In-person registration is SOLD OUT — both the conference-days-only option (24-25 Aug) and the tutorial-plus-conference option (23-25 Aug) show sold out at every rate. Virtual attendance is still open and covers all three days: $210 professional/IEEE member, $255 professional/non-member, $90 student/IEEE member, $115 student/non-member, and $5 for students in developing countries (rates as listed for registration before 1 August 2026). A virtual pass includes live streaming of every presentation, downloadable slides, a private Slack workspace for Q&A with presenters, and session replays. The refund request deadline was 11:59pm PDT on 19 August 2026 and has passed.

When is Hot Chips 2026 (HC38)?

Hot Chips 2026 (HC38) takes place August 23–25, 2026 in Palo Alto, California.

Where is Hot Chips 2026 (HC38) held?

Hot Chips 2026 (HC38) is held in Palo Alto, California. The event format is hybrid.

How much does Hot Chips 2026 (HC38) cost?

Hot Chips 2026 (HC38) is paid. In-person registration is SOLD OUT — both the conference-days-only option (24-25 Aug) and the tutorial-plus-conference option (23-25 Aug) show sold out at every rate. Virtual attendance is still open and covers all three days: $210 professional/IEEE member, $255 professional/non-member, $90 student/IEEE member, $115 student/non-member, and $5 for students in developing countries (rates as listed for registration before 1 August 2026). A virtual pass includes live streaming of every presentation, downloadable slides, a private Slack workspace for Q&A with presenters, and session replays. The refund request deadline was 11:59pm PDT on 19 August 2026 and has passed.

Who should attend Hot Chips 2026 (HC38)?

Hot Chips 2026 (HC38) is aimed at Chip and system architects, Data center infrastructure leaders, AI platform and ML systems engineers, CTOs at semiconductor and hyperscale companies, Hardware performance engineers, Compiler and runtime engineers, Technology analysts and investors.

What topics does Hot Chips 2026 (HC38) cover?

Hot Chips 2026 (HC38) covers AI accelerators, GPU architecture, CPU architecture, HBM and memory systems, RISC-V, Chiplets and advanced packaging, AI networking and interconnect, Processing-in-memory, Automotive and autonomous-driving compute, Data center infrastructure.

This page was written from 4 sources, 2 on domains other than hc2026.hotchips.org.

1. 1.hc2026.hotchips.org — programvendor
2. 2.hc2026.hotchips.org — registrationvendor
3. 3.hotchipssymposium.regfox.com — hot chips 2026
4. 4.nvidia.com — hot chips conference