AI2Day Daily Brief — 27 Aug 2026

Good morning — it is Thursday the twenty-seventh of August, and we are starting with autonomous excavators doing real work on real US construction sites right now, no demo, no test track, no human in the cab.

Bedrock Robotics, a San Francisco startup founded by veterans of Waymo, has deployed fully autonomous excavators on three live construction projects in Texas and Nevada. The machines are standard heavy excavators fitted with a sensor and computer kit, and they are digging and moving earth without an operator. Bedrock says more than forty percent of the US construction workforce is expected to retire within five years, which helps explain the pitch. The industry also runs an average of twenty percent over schedule and eighty percent over budget, according to the company. Construction workers make up seven and a half percent of the US workforce but account for twenty percent of all workplace deaths. The company has raised three hundred and fifty million dollars in venture capital and is not describing this as a pilot.

Worth noting that those schedule and budget figures are Bedrock's own. Moving to industrial vision — a German camera maker has released a new three-dimensional sensor aimed squarely at factory floors and logistics warehouses.

IDS Imaging, based in Obersulm, Germany, has launched the Nion: a compact three-D camera now available in full production quantities. It delivers one-point-two-megapixel depth images at thirty frames per second, using time-of-flight sensing — the camera fires infrared light pulses and measures how long they take to bounce back, giving every pixel a distance reading rather than just a colour. It handles fast-moving objects and difficult lighting conditions, covers a range of point-three to seven-point-five metres, carries an IP sixty-seven weatherproof rating, and connects through a single Power over Ethernet cable. IDS positions it as a practical entry point for factories already running two-dimensional camera systems that want to add depth awareness without a major infrastructure overhaul.

Practical and unglamorous, which in industrial robotics is usually a good sign. Now to OpenAI, where the subscriber numbers are going up and the executive roster keeps thinning out.

More than a dozen senior OpenAI leaders have left since January, including the chief operating officer, the chief revenue officer, and the chief marketing officer. Last week, Chris Malone, the executive who oversaw the company's data centres, departed after roughly fifteen months in the role — OpenAI cited an internal reorganisation. Co-founder and president Greg Brockman is picking up the pieces: both the infrastructure and product teams now report to him. Meanwhile, OpenAI's desktop app added roughly fifteen million subscribers in two months, the company filed confidential paperwork with the SEC to go public in June, but the IPO is not now expected until twenty-twenty-seven. Revenue is rising, and by multiple accounts, losses are rising alongside it.

A company that large shedding that many senior people in that short a window is worth watching closely. Next, Nvidia just posted the biggest quarterly revenue in its history.

Nvidia reported ninety-six-point-two billion dollars in revenue for its most recent quarter — up from roughly eighty-six billion the previous quarter. Nearly all of it came from data centre sales: the specialised chips that tech companies, cloud providers, and governments buy to build and run AI systems hit eighty-nine billion dollars, more than double the figure from the same quarter a year earlier. Quarterly profit came in at fifty-nine-point-seven billion. The company is now forecasting one hundred and eight billion in revenue next quarter, which would put it alongside Amazon, Apple, and Alphabet as companies that routinely clear a hundred billion in three months. On the consumer side, component shortages are still pushing graphics card prices up, and Nvidia has flagged further price increases on its AI chips.

One hundred and eight billion in a single quarter is a forecast, not a result — but the direction of travel is hard to argue with. Finally, Perplexity has released an AI agent that runs entirely on your own machine.

Perplexity has launched a product called Portable Computer, built in partnership with Nvidia. It is a version of the company's existing agent platform — software that can carry out multi-step tasks such as reading a folder of documents, identifying problems, and writing a summary — but it runs locally on hardware you already own rather than on a remote server. That means no per-task cloud fees and no automatic upload of your files. It runs on Nvidia DGX Spark desktops and Linux machines with a compatible Nvidia RTX graphics card — minimum a thirty-ninety with twenty-four gigabytes of video memory — and Windows support is coming in September. On Perplexity's own fifty-three-task benchmark, the agent scored eighty-five-point-four percent accuracy. The nearest open-source rival scored seventy-seven-point-six. Those benchmarks are the company's own.

Today's sponsor is Zero2Hero: an AI poker coach sits at your table, and you can ask it mid hand whether the player across from you is bluffing. The free tier is real: 50 hands a day against AI opponents, no card. Not a trial. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

For the deeper look at this week's stories, the Monday show — AI Today Weekly — has the full analysis. That is your AI briefing for today. Every story is at A-I-2-Day dot live. That is A, I, the number two, D-A-Y, dot live. We are back tomorrow morning. If this was useful, hit the thumbs up and subscribe, so the next one finds you.