EY says its 'invisible' AI router has helped cut token consumption by up to 60%
THE SO WHAT
If a Big Four can cut token spend by up to 60% with an “invisible” routing layer, AI cost optimization is becoming its own discipline. You should be instrumenting and routing across models based on price-performance, not hardwiring everything to a single flagship API.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIChatGPT’s first answer is usually the most boring — here’s how I get better ones
Prompt literacy is quietly becoming a performance differentiator—teams that know how to push assistants past safe, generic answers get more leverage from the same tools. Worth monitoring for ops leaders: the gap between “we use ChatGPT” and “we know how to drive it” is now a real productivity spread.
Applied AIThe lineage behind 69% of open models was never verified. Cisco just fingerprinted almost 900 for free
Model provenance is moving from GitHub etiquette to a security control—69% unverifiable lineage is a non-starter for regulated deployments. If you’re approving open models, add fingerprinting and lineage verification to your checklist this week or accept that you’re flying blind on IP and supply-chain risk.
Applied AIGoogle integrates Nano Banana 2 into Google Earth via a new Create Image button, letting users prompt visual alterations of locations or infographics
When even Google Earth gets a “Create Image” button, generative UX becomes a default expectation in consumer products. For product teams, the question isn’t whether to add gen visuals, but where they actually change engagement or understanding versus being a novelty.
Applied AIMeta Reports $279 Billion in Future Data Center Leases for AI
A $279B pipeline of AI data center leases makes clear: hyperscale AI is now an infrastructure commitment on the order of utilities, not IT. For anyone building on these platforms, this level of locked-in capacity suggests pricing and access will be shaped more by power and regulation than by raw demand.