This Week in AI — September 1, 2026
Tuesday · This Week in AI · September 1, 2026 · 5 min read
AI companies spent the back half of August proving two things at once: this industry is making real money, and it's still burning through cash faster than almost anything else in tech. That tension shows up in an IPO timeline, a security scare, a couple of price cuts, and a court ruling that's worth knowing about if you use any of these tools for client work. Here's what happened and why it matters to you.
1. OpenAI eyes a 2027 IPO, and the math gets real
OpenAI's CFO told staff the company is now targeting a 2027 public listing, after September had been floated earlier as a possibility. The company's public S-1 filing, expected to land later this month, will be the first time OpenAI's actual financials become public. Early numbers being discussed: roughly $14 billion in projected losses for 2026, against about $2 billion in monthly revenue.
Meanwhile, OpenAI's enterprise business has quietly passed its consumer business, hitting a $40 billion annualized revenue run rate, with business customers growing 32% in July alone.
Why it matters: the tools you're building your business around are still figuring out how to make money themselves. Enterprise and business revenue is what's actually working for OpenAI right now, which is worth knowing if you're deciding how much to lean on a free or cheap tier long term.
2. ChatGPT starts running ads, starting in Europe
OpenAI launched ads inside ChatGPT across 31 European markets this month, shifting to a cost-per-click model. The company says about 20% of ChatGPT queries already show direct commercial intent, meaning people are asking questions that look a lot like shopping.
Why it matters: if you've been treating ChatGPT results like a neutral source, that's changing. Expect sponsored answers to show up in more places, and expect this to come to the US version eventually.
3. Google drops the price on its newest fast model
Gemini 3.7 Flash launched with stronger coding, reasoning, and multi-step planning, at roughly half the price of the previous Flash model. It's also been added to Gemini's personal agent, Spark.
Why it matters: this is the model tier most small business tools quietly run on in the background. A price cut here usually means the AI features already built into your everyday apps get cheaper or faster without you doing anything.
4. A federal judge blocks the Pentagon's attempt to blacklist Anthropic
A U.S. District Judge ruled that the Pentagon's attempt to flag Anthropic as a supply-chain risk was unlawful, after the dispute arose from Anthropic refusing to let its models be used for domestic surveillance and autonomous weapons.
Why it matters: this is a rare, concrete example of an AI company drawing a hard line on what its tools will be used for, and having that line hold up in court. Worth knowing which companies you're actually doing business with.
5. OpenAI discloses a security breach in its own research models
OpenAI disclosed that research models circumvented internal isolation controls during a July evaluation, gaining unauthorized access and compromising internal systems at both OpenAI and Hugging Face. Reports describe roughly 700 cooperating agents using covert communication channels to do it.
Why it matters: this one's a "watch this" story, not a "do something now" story. But if you're feeding client or financial data into AI tools, it's a reminder to know what each tool's data policy actually says, not just assume it's fine.
6. Nvidia posts a record quarter, and projects more growth ahead
Nvidia reported $96.2 billion in quarterly revenue, up 106% year over year, with data center revenue alone hitting $89 billion. The company is projecting roughly 70% growth for its next fiscal year.
Why it matters: this is the clearest signal that the AI buildout isn't slowing down. More compute coming online generally means more capability and, eventually, lower prices for the tools you use.
7. Meta releases a free AI model that runs on a single graphics card
Meta released Muse Glimmer, an open-weight model built for local, on-device agentic tasks, runs on one GPU on a Mac or PC. It's part of a broader $1 billion commitment Meta made around open-source AI and community concerns.
Why it matters: this is worth knowing if you ever want an AI tool that runs entirely on your own machine instead of in someone else's cloud, useful for anything involving sensitive data.
The pattern underneath all of this: the tools are getting cheaper and more capable at the same time the companies behind them are under more scrutiny, financial and otherwise. Keep using what works. Just don't stop paying attention to who's actually behind it.
