Uber plans 2000 AI robotaxis

Welcome to OHMO AI. Uber and Pony.ai want 2,000 robotaxis on the road in four European cities. Anthropic is reportedly in talks to spend about $6 billion on a startup that builds simulations of the physical world.

The cheap end of AI is getting expensive. DeepSeek's API prices go up by more than four times, and OpenAI is previewing a mode that runs its top model 14 times faster.

In today's newsletter:

OHMO AI: Uber is heading to Europe with Pony.ai, a Chinese autonomous vehicle maker, and the two of them want to put 2,000 driverless cabs on the road in four cities. That's a plan rather than a fleet in service, and a rollout of that size is a different proposition to a pilot in one neighbourhood.

The details:

Why it matters: Robotaxis have mostly been something you watch in a video from another continent. A fleet at this scale would put driverless cars in front of Europeans who've never stood next to one. The number is the easy part, and the harder part is the four sets of regulators standing between the plan and a passenger.

CEOs, politicians, regulators, and analysts talk for hours on podcasts every day.

Particle actively transcribes 125,000+ podcasts and turns them into structured data, available over REST API or MCP, searchable within minutes of airing.

Track a company or person by name, with each mention cited to the exact quote, speaker, and timestamp, or search for commentary on a thesis or sector with semantic search.

Particle is built by former Twitter and Tesla engineers, using an AI-native transcription pipeline that delivers high accuracy and extensive data enrichment.

Start with $10 in free credit.

OHMO AI: Anthropic is reportedly in talks to buy the AI startup Decart AI for about $6 billion. Decart builds world models that aim to simulate the physical world, along with software that can reduce the cost of training AI by helping chips work more efficiently. The deal has not been finalized and the talks could fall through.

The details:

Why it matters: Anthropic sells a thing you type into. World models point somewhere else, at machines that need to know how the physical world behaves before they move through it. Reportedly putting about $6 billion behind that is a read on where the money goes once chat stops being the product, and it only counts if the talks hold.

OHMO AI: DeepSeek is bringing in peak-hour pricing that raises its rates by more than four times from current levels, and it starts on 16 August. What you pay for the same request now depends on when you send it.

The details:

Why it matters: DeepSeek's price is why a lot of small builders could afford to ship anything at all, and charging by the time of day changes that maths overnight. Anyone running a product on V4 has a day to work out what their app costs now. Cheap Chinese models have been the pressure holding everyone else's prices down, and that pressure eased a little.

Leadership is asking: where is AI delivering value for us and where is it creating risk? Right now, most teams have no idea.

With Harmonic Security’s Usage Explorer, you get a complete picture of how your organization actually uses AI, automatically categorized into custom use cases with complete tool-level granularity.

See it in action

OHMO AI: OpenAI has put out an early version of Ultrafast, which runs the company's latest and most powerful model at up to 750 output tokens per second, 14x the speed of standard processing. The model in question is GPT-5.6 Sol, and the speed comes from silicon rather than software.

The details:

Why it matters: Waiting for AI to finish typing is the part nobody talks about and everybody feels. Take that wait down far enough and things that don't work today start to, like translation that keeps up with a conversation instead of trailing it. It's a preview, which means it's a demonstration of what the chips can do, not something you can build a business on yet.

Get marketing advice that actually knows your business

Ask AI for ad copy or positioning cold and it guesses, so you get advice that could apply to any business. Give it the real details first and it gives you advice you could actually use. Ten minutes to write once, then paste it before every marketing question from now on.

Copy this, swap in the bracketed parts, and paste it in.

Each field closes off a way generic advice sneaks in. The exact product and customer profile stop the AI from writing for a category instead of your business. Naming competitors and alternatives, including doing nothing, forces the advice to argue against something real. Objections are what most people leave out, and they're the difference between copy that sounds nice and copy that closes. The pricing reasoning stops the AI from inventing a justification you don't actually use. Real testimonials give it your customers' actual language instead of made up ones. Everything you ask after this is only as sharp as this one file, so it earns the ten minutes.

Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.

Today's reader is Nuno C. from Porto, Portugal:

“I run a small tile and bathroom showroom with my brother. Quotes were killing us. Someone sends photos of their bathroom and a rough size on a Tuesday and wants a price by Friday, and I'd be at the kitchen table doing it at ten at night. Now I photograph the plan, drop it into ChatGPT with our price list, and it drafts the quote line by line in Portuguese with the VAT split the way our accountant wants it. I still check every number because it got the square metres wrong twice. But that's an hour a night back, and my brother has stopped calling me the bottleneck lol.”

Want to be in the next newsletter? Tell us how you use AI here.

AI Tools: A curated list of the AI tools worth knowing. 

Instagram: The AI page they pray you’ll never find.

Facebook: For the ones who never left Facebook.