In partnership with
Welcome to OHMO AI. Meta is spending hundreds of millions of dollars a year to reach AI models through Microsoft's Azure, and it runs trillions of tokens through the platform every week. Demand for AI is still concentrated in the tech industry, and that's what the concentration costs.
Meta AI landed on the Mac. Nvidia's H200 chips reached ByteDance and Tencent. OpenAI previewed a way to catch misuse without keeping what you typed.
In today's newsletter:
Meta turns out to be one of Microsoft's biggest AI customers
Meta AI lands on the Mac as a desktop app
Nvidia's H200 chips reach ByteDance and Tencent
OpenAI previews a misuse check that keeps none of your data
LATEST IN AI
META
Meta is one of Microsoft's biggest AI customers
Image source: OHMO AI
OHMO AI: Meta has become one of Microsoft's biggest AI customers. It pays for access to AI models on Microsoft's Azure cloud, and each week it runs trillions of tokens through the platform, tokens being the unit AI computing gets consumed in. Demand for AI, it turns out, is still coming mostly from inside the tech industry itself.
The details:
The spending runs to hundreds of millions of dollars a year.
Each week Meta burns through trillions of tokens on the platform, which is a measure of raw usage rather than headcount.
The arrangement is a straight customer relationship. Meta is buying access to AI models hosted on someone else's cloud.
It underlines how concentrated AI demand still is in the tech industry rather than outside it.
Why it matters: The AI boom is usually described as everyone buying in. But one of Microsoft's biggest AI customers is Meta, spending hundreds of millions a year, and demand for the technology remains concentrated in the tech industry. If that holds, the growth being priced into all of this rests on a small circle of customers, not a broad one.
TOGETHER WITH BELAY
The New Rules of Online Visibility
Your customers are searching in places your strategy doesn’t reach.
So before your business is buried and left behind, you need to understand the new rules of SEO.
BELAY's SEO in the Age of AI report explains how search is changing, what AI means for your visibility, and the practical steps small businesses like yours can take to stay visible.
BELAY’s U.S.-based Marketing Assistants turn strategy into execution, helping your business stay visible, credible, and competitive in every search.
META
Meta AI lands on the Mac as a desktop app
Image source: OHMO AI
OHMO AI: Meta has launched a Mac desktop app for its Meta AI chatbot, which puts it in direct competition with the AI assistants people already keep open all day. The difference it's selling is business data. Meta says you can ask about your own account engagement and ad performance and get answers drawn from context only it holds.
The details:
It's a standalone desktop build of the Meta AI chatbot, running on Mac.
Meta frames it as a potential competitor to the other popular AI chatbots.
The business pitch is answers pulled from data Meta already holds on an advertiser, naming account engagement and ad performance as the examples.
The company set the change out in a blog post, describing it as an update to what its assistant can answer.
Why it matters: Most AI assistants know only what you type into them. This one arrives attached to the ad account and the engagement numbers a small business already runs on, which is a real reason to open it instead of the chatbot you use now. It also means the assistant sitting on your desktop belongs to the company holding your advertising data, and that's a trade worth making on purpose.
NVIDIA
Nvidia's H200 chips reach ByteDance and Tencent
Image source: OHMO AI
OHMO AI: Nvidia's H200 AI chips have started landing in mainland China. ByteDance and Tencent have each taken delivery of a batch in recent weeks, under US clearance that lets both of them buy far more than they've received so far.
The details:
ByteDance and Tencent have each received about 10,000 H200 processors over the past few weeks.
US clearance goes much higher. Each company is approved to purchase up to 100,000 of the chips.
The approvals weren't limited to those two. Ten Chinese firms were given permission to buy H200 processors.
Why it matters: Export policy usually plays out as speeches, rules and rumours about rules. This is hardware physically arriving, with paperwork approving far more of it. What China's AI companies can build next now depends less on the argument and more on how many of those chips actually get on the plane.
PRESENTED BY PARTICLE
Glean insights from across the podcast landscape.
Search by company, by theme, or by entity, and get alerted the moment something airs. Each result is cited back to the exact quote, speaker, and episode.
OPENAI
OpenAI previews a misuse check that keeps no data
Image source: OHMO AI
OHMO AI: OpenAI has previewed Private Safety Processing, a monitoring system that looks for abuse patterns while holding on to none of the customer's data. It widens the scope of the company's zero data retention offering, and instead of judging one chat on its own it reads across several, which is where OpenAI says the serious risks tend to surface.
The details:
OpenAI calls it a form of long-horizon safety monitoring, assessing inputs and outputs across multiple conversations instead of one.
The company previewed it as a system that "hunts for misuse across several interactions at once".
It's pitched as widening the scope of zero data retention rather than replacing it.
OpenAI's reasoning: "The most serious AI safety risks are not always visible in a single interaction."
Why it matters: Privacy and safety are normally sold as a trade. To catch the bad thing, somebody has to read your chats. This is an attempt to have both, watching the shape of behaviour without keeping the content. It's a preview, so the things to watch are who gets it and whether a system that spots risk across conversations can do that without quietly becoming a system that remembers them.
AI GUIDES
See every tradeoff at once instead of a single ranked list
Ask AI to rank your options and it flattens everything that matters into one score, hiding the tradeoffs that actually made the decision hard. Ask it to plot the options instead and you can see all of them at once: which one is the long shot, which one is the safe bet, and which one quietly beats both. Costs ten seconds and works in any chatbot with charts.
Copy this, swap in the bracketed parts, and paste it in.
I'm deciding between these options: [list your options, e.g. Job A, Job B, Job C]. The real tension is [name the tension, e.g. higher pay versus better hours]. Don't rank them for me. Instead, plot every option on a chart, using [variable 1, e.g. salary] for one axis and [variable 2, e.g. hours per week] for the other axis. Show [variable 3, e.g. company size or vacation days] as the size or color of each point. Let me click into any point for more detail, and be ready to re-plot if I add an option or change what matters most.Naming the actual tension up front stops the AI from quietly picking a winner for you, since it now knows there are two things pulling in opposite directions, not one. Putting the two biggest variables on the two axes means you see both at once, side by side, instead of buried inside a single number. Adding a third variable as size or color gets you more information without turning the chart into a wall of text. And asking for a re-plot means this isn't a one-time answer: as new options come in or your priorities shift, you update the picture instead of starting over.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Mary S. from Porto, Portugal:
“I run a small tile and stone shop with my brother. The quoting is what kills us. Someone sends a photo of a bathroom at 9pm and wants a price before bed. Now I take my own photos, type in the measurements, drop in the supplier price list, and get AI to work out the square metres, the waste allowance and a quote laid out the way we send them. It gets the maths right more often than I do that late. Last month it caught that I'd priced adhesive per bag instead of per pallet, which would have come out of my own pocket. My brother still checks every single one lol.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
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.
