On the Q1 earnings call, Lisa Su said MI450 was already sampling to top customers — and that more on Helios and MI500 would come in July. That deadline is a week away: July 22–23 at San Francisco's Moscone Center for Advancing AI 2026.
Last year's show delivered MI350, ROCm 7, and the first public look at the Helios rack, with Meta, OpenAI, and Microsoft on stage. The pressure this year is different: the market isn't looking for another roadmap slide — it wants proof that Helios can ship and MI450 has real deployment stories. The July 23 keynote streams on AMD's YouTube channel; West Coast viewers get a morning show, East Coast gets a lunch-hour slot.
Below we organize what's confirmed and what's reasonable to expect before launch: dates, live stream links, how the two days break down, and the hardware/software beats worth tracking. The agenda is still updating — check AMD's site for the latest.
I. Date & Location
| Item | Details |
|---|---|
| Event name | AMD Advancing AI 2026 |
| Dates | Tuesday, July 22 – Wednesday, July 23, 2026 |
| Location | Moscone Center, San Francisco, California |
| Format | In-person attendance + online live stream (keynote) |
| Official announcement | April 28, 2026 |
| Keynote | July 23, main keynote by CEO Lisa Su |
Timezone reference (PDT & EDT)
San Francisco runs on Pacific Daylight Time (PDT, UTC−7) in July. US Eastern Daylight Time (EDT, UTC−4) is 3 hours ahead — so a 9:00 AM PDT session starts at noon on the East Coast.
AMD's event page lists partial breakout times (exact keynote start still TBD on the site):
| Session (published agenda snippet) | San Francisco PDT | US Eastern EDT |
|---|---|---|
| July 22 breakouts | 2:00 PM – 2:45 PM | 5:00 PM – 5:45 PM |
| July 23 breakout A | 11:00 AM – 11:30 AM | 2:00 PM – 2:30 PM |
| Helios rack infrastructure session | 12:00 PM – 12:45 PM | 3:00 PM – 3:45 PM |
| July 23 breakout B | 1:00 PM – 1:45 PM | 4:00 PM – 4:45 PM |
| Lisa Su keynote (exact time TBD) | July 23 (morning, per past years) | Likely noon–1:00 PM EDT; confirm on stream |
US viewers: set a reminder
If the keynote opens around 9:00–10:00 AM PDT, that's roughly 12:00–1:00 PM EDT — ideal for a lunch-break watch on the East Coast. Add a calendar alert from the AMD event page, or subscribe to notifications on AMD YouTube.
II. Live Stream
Per AMD's investor relations press release and event site:
- Live stream: AMD's official YouTube channel — July 23 keynote broadcast online.
- In-person registration: Sign up at the Advancing AI event page; Google Calendar / iCal one-click add available; seats are limited.
- Agenda updates: As of this writing, the full keynote schedule and speaker list aren't fully public yet. Registrants receive email updates.
The event page also teases developer offerings: AI Agent, model optimization, and training workshops, plus certification programs and free GPU resource access — exact eligibility TBD closer to the show.
III. Schedule
While the detailed agenda is still rolling out, AMD and industry coverage have already shaped a clear two-day outline:
| Date | Focus | Typical content |
|---|---|---|
| July 22 | Ecosystem partners & developers | Breakouts, workshops, customer stories, AI infrastructure deployment practices |
| July 23 | Keynote & major launches | Lisa Su keynote, MI series & Helios progress, roadmap previews, partner appearances |
One confirmed technical session: "Deploying Rack-Scale AI Infrastructure with AMD Helios" (July 23, 12:00–12:45 PM PDT), covering how scale-up and scale-out map to training, inference, and agentic workloads — a signal that Helios is a core narrative pillar this year.
IV. Expected Announcements
Everything below is expectation, not AMD commitment. Final product names, specs, and ship dates come from the July 23 keynote and press releases.
Instinct MI450 series
On the Q1 2026 earnings call, Lisa Su confirmed MI450 GPUs were sampling to leading customers. At Financial Analyst Day, AMD disclosed MI450 on CDNA 5 with 3D packaging and higher-bandwidth memory, targeting narrower gaps vs. Nvidia's flagship in inference and select training scenarios.
A reasonable Advancing AI expectation: formal SKU details, production windows, and first deployment customers — not another slide-only preview. MI450 is the linchpin of AMD's 2026 datacenter revenue story.
Helios rack-scale AI platform
Helios debuted as a concept at last year's Advancing AI — a full-rack AI system integrating multiple MI accelerators, interconnect, and cooling, positioned against Nvidia's Vera Rubin rack-scale approach. Multiple analysts expect Helios to enter production/delivery in Q3 2026.
Watch for: rack configurations, interconnect topology, hyperscaler PO progress, and the "scale-up first, scale-out on demand" deployment philosophy — consistent with the official Helios session description.
Instinct MI500 roadmap preview
AMD previously outlined the MI500 series (targeting 2027) at analyst events: CDNA 6, advanced process (industry rumors point to ~2nm class), HBM4E, and more. Lisa Su also hinted July would bring more MI500 and Helios roadmap detail.
Expect a roadmap preview, not a full next-year spec sheet. Investors will watch whether performance-multiple promises vs. MI300X still hold.
EPYC Venice (Zen 6) & ROCm
On the CPU side, EPYC Venice (Zen 6) may get airtime as the host/orchestration layer in AMD's "AI factory" story — GPU launches need a matching CPU narrative.
Software-side, last year brought ROCm 7. This year watch day-one MI450 support, PyTorch/JAX optimizations, and whether CUDA migration friction drops further. For teams building their own clusters, ROCm maturity often matters more than peak FLOPS when making purchase calls.
| Topic | Last year (2025) | This year (2026) expected |
|---|---|---|
| GPU accelerators | MI350 launch, MI400 preview | MI450 production details, MI500 roadmap |
| Rack systems | Helios first reveal | Helios delivery & deployment stories |
| Software | ROCm 7 | ROCm update / MI450 optimizations |
| Narrative | Open ecosystem, partner stage time | From "preview" to "orders & production" |
V. Last Year
Looking back at Advancing AI 2025 helps gauge this year's delivery pressure:
- Instinct MI350 accelerator officially launched
- ROCm 7 software stack unveiled
- MI400 series and Helios rack platform previewed
- Partners included Meta, OpenAI, Oracle, Microsoft, Cohere, xAI, Red Hat, Astera Labs, Marvell, and others
Last year's tone was "open AI infrastructure" and ecosystem breadth. This year the market wants Helios on schedule and MI450 winning cloud-scale orders — why analysts (Cantor Fitzgerald raised its price target to $700) are watching closely.
VI. Who Should Watch
Quick guide by role:
| Role | Watch for | Can skip |
|---|---|---|
| Cloud architect / MLOps | Helios topology, MI450 inference value, ROCm deployment path | Consumer APU rumors |
| AI app developer | Free GPU resources, certification & workshop announcements | Rack interconnect minutiae |
| Mac / edge developer | Indirect: cloud training costs & open-model hosting trends | Usually no Apple or Xcode content here |
| Investor | Named customers, 2026–2027 revenue guidance, MI500 promise conservatism | Hands-on lab details |
If you mostly run Codex / Agent workflows on Mac, this keynote won't change your IDE shortcuts — but it does shape backend compute pricing and open-stack choices. Block 45 minutes for the full keynote.
FAQ
When is AMD Advancing AI 2026?
July 22–23, 2026 at Moscone Center, San Francisco. The keynote is July 23, delivered by CEO Lisa Su.
Where is the live stream?
AMD's official YouTube channel. Also register on the event site for agenda email updates.
Will consumer GPUs launch here?
Historically, no. Advancing AI targets datacenter AI and ROCm; Radeon gaming lines typically update at Computex or other venues.
Do I need a ticket?
Online keynote viewing is usually free. In-person seats require site registration (limited); some workshops may have extra requirements.
Advancing AI vs. Nvidia GTC — which to watch?
GTC is Nvidia's full-stack CUDA show; Advancing AI is the main stage for the "second supplier" open path and MI roadmap. Anyone doing compute procurement or multi-cloud strategy should catch both keynotes on replay.
Conclusion
The core question for Advancing AI 2026 is simple: can last year's promises become shippable racks and production MI450s? Dates are set (July 22–23), the venue is set (Moscone), and the stream is set (YouTube + Lisa Su on July 23) — what remains is specs, customer names, and whether the MI500 roadmap is convincing enough for cloud vendors to keep diversifying supply.
For US developers: block 45–90 minutes around noon EDT on July 23 for the live keynote, or catch the replay the same afternoon. You don't need every breakout unless you're deploying ROCm or Helios-scale infrastructure.
Edge AI development (iOS, macOS Agents, local IDE work) still depends on whether your Mac node stays stable — cloud training costs are the conference topic; keeping local builds unbroken is something you control today.
Need stable macOS for long Agent tasks? Pair with a Cloud Mac execution node
Conferences show trends; execution environments are yours to own. Vuncloud dedicated Mac mini M4 keeps Codex / Xcode builds and TestFlight tasks running to completion in your local dev flow.
Related Reading
- Why Are Chinese AI Models Dominating the OpenRouter Leaderboard? 5 Signals Developers Should Read
- How to Choose GPT-5.6 Sol, Terra, or Luna? Performance, Price & Speed Compared (2026)
- The Model Arms Race Is Over—Why Mac Compute Nodes Are Suddenly Hard to Get
- Is Mac Mini M4 Good for AI Development on a Cloud Mac? (2026)
Schedule and stream details per the AMD official event page and investor press release; product expectations are pre-event industry analysis, not purchase or investment advice. Last updated: July 14, 2026.