Ai4 bills itself as America's largest AI event, and next week, August 4–6, it takes over The Venetian in Las Vegas.
The timing is very 2026: as I write, the city is under an extreme heat warning. This Sunday the mercury is forecast to hit all-time August records around 115°F (that’s 46 C for all you visiting from the rest of the world).
Suffice it to say we’ll be indoors, talking about a different kind of burn rate.
An agenda fit for an older, wiser AI industry #
Conference agendas are a great way to read the barometer for what’s top of mind in an industry. Scan this year's Ai4 tracks and you’ll see some patterns emerging: four infrastructure tracks, an AI Policy Summit, a dedicated safety and alignment track, and topic lists full of phrases like "cost management and optimization in AI infrastructure" and "governance, logging, and auditability in agent environments."
Even the keynotes, anchored on agentic AI and frontier models, are as much about consequences as capabilities.
In 2026, the industry has largely moved on from asking what AI can do, to how we can do it safely and efficiently at scale. This will likely be a very welcome development for any IT leader still stuck in the (by now well-researched) gap between a dazzling pilot and a production service.
So, three predictions for the week.
- On stage, the most-said word will be "agents." Over coffee, it will be "budget," — because swarming agentic workloads consume tokens with jaw-dropping speed. Sovereigntywill be ubiquitous, as data residency and control-boundary requirements harden across regulated industries of every kind, from telco to healthcare to financial services.- Open-weight models will be discussed as peers of the frontier labs, because for lots of benchmarks, and real workloads, they now very much are. We reckon Kimi K3 will be on plenty of lips, and soon on plenty of on-prem hardware (at least, if you’ve got enough vRAM to hold it).
The next chapter in local inference #
Ai4 lands just two weeks after AMD Advancing AI, where we announced PaletteAI Inference Launchpad: a turnkey software solution that makes a GPU server into a governed, local-first inference platform, aimed at today’s token-hungry agentic developers.
It routes each request to a local (open) or frontier model based on economic, governance and sovereignty policies you set, with token metering and quotas per team and model, and depending on workload mix it can cut token costs by up to 70%.
Cost… sovereignty… open models… forgive us for stating the obvious, but we reckon the Launchpad is a bullseye for those big trends converging at Ai4.
It’s fortunate then that we’ll have something new to share at the event, in partnership with our friends at AMD. Our lips are sealed for now.
The era where the industry picks its terms #
Every technology wave reaches a point where the hype fades away and user reality defines the agenda — often around topics like cost, standards vs lock-in, and security. The web had that moment. Cloud had it. AI's is arriving now, and we’re excited to be at the heart of the ecosystem that’s shaping it.
So this is the charter we'll be making the case for in Vegas: open models and open stacks treated as first-class citizens; transparency and control over what inference costs; and sovereignty, the ability to run AI on infrastructure you govern, available to any organization that wants it.
If any of this maps to your world, then book time with us at Ai4. We're hosting an Experience Zone for private meetings and demos, and we’ll be available to connect in the AMD meeting room, too. Oh, and the drinks will be ice cold. In Vegas next week, that might be the most compelling offer of all.