Serendipity35

This week, OpenAI made news with its new browser, Atlas. Wth all their plans and a new cloud-based AI browser, they need to scale their computing power. Here is a news summary (AI-generated, of course)

OpenAI has announced a strategic multiyear partnership with semiconductor giant Broadcom to co-develop custom-built chips and infrastructure. The collaboration aims to deploy 10 gigawatts of specialized AI accelerators by the end of 2029—a staggering amount of compute capacity equivalent to the power consumption of approximately 8 million U.S. households. (Markets Insider)

This deal marks OpenAI’s first venture into designing its own in-house processors, with Broadcom tasked with developing and deploying the systems. The chips—known as AI accelerators—are optimized for parallel processing, enabling them to execute billions of operations simultaneously. These accelerators will be deployed across OpenAI’s facilities and partner data centers, using Broadcom’s Ethernet-based networking solutions to ensure scalability and efficiency. (Broadcom Inc)

OpenAI CEO Sam Altman emphasized the significance of the partnership: “Developing our own accelerators adds to the broader ecosystem of partners all building the capacity required to push the frontier of AI to provide benefits to all humanity.” The Broadcom agreement is the latest in a series of megadeals OpenAI has struck in 2025 to secure the compute power needed for its rapidly expanding AI services. Earlier this year, OpenAI signed a $100 billion deal with Nvidia to deploy 10 gigawatts of Nvidia systems, beginning in the second half of 2026. (Reuters) In another major move, OpenAI partnered with AMD to deploy 6 gigawatts of chips and received a warrant for up to 160 million AMD shares, potentially making it one of AMD’s largest shareholders. (The Motley Fool)

These deals reflect a broader industry trend: major tech players are increasingly investing in custom silicon to reduce reliance on Nvidia’s dominant GPU offerings. Companies like Google, Amazon, and Microsoft have already begun developing their own AI chips, and OpenAI’s latest move places it firmly within this competitive landscape. While OpenAI has not disclosed how it plans to finance the Broadcom deal, analysts estimate that a single gigawatt-scale data center could cost between $50 billion and $60 billion. 

This latest partnership not only strengthens OpenAI’s technical capabilities but also signals a shift toward greater control over its hardware stack—an essential step as the race to develop next-generation AI systems accelerates.

OpenAI's new Atlas browser has an Agent mode, which allows it to navigate sites, fill carts, and perform real-world tasks. Can it do it well for you - and will you trust it to do it for you?

Elyse Betters Picaro, a Senior Contributing Editor at zdnet.com, did a shopping test. For Plus and Pro users (only on Mac IOS for now), it includes a powerful Agent mode so that ChatGPT can take over your browser, click around, and perform tasks for you.

She tried having it do an order from Walmart and reported the process and results. You put n a prompt, just as you would for any chatbot request. Unsurprisingly, the more specific the prompt, the better the result. (GIGO still lives!) Initially, she asked, "Order me wood putty, paintable caulk, and 2-inch screws from Walmart." In a way out of some sci-fi story from the past, the Agent took over the cursor and all while she watched.

Her test is enough of a cautionary tale that I have not tried a similar test myself at this point. Privacy fears...

Walmart's site created a hurdle (a language-selection pop-up) that blocked Atlas' Agent from continuing, even though she had it access to her Chrome data and Keychain. That's frightening, but consider that she had already given the key data to those two other companies. It couldn't log in to Walmart, didn't know her location or default store, even though she has ordered from them before. She revised her prompt (as we have all done) and on the third try she prompted: "Order me 5 wood putty, 5 paintable caulk, and one pack of 2-inch screws. I want them delivered to my house from the Malone, NY, location in an hour. I've ordered these before, so use my past purchases to find the right products and brands I use."

It worked. The agent used her purchase history, searched for the products in past orders, and added them to her cart. It didn't complete the order but paused at the checkout screen so she could select a delivery window, adjust the tip, and confirm payment.

Does this amaze you, excite you, or frighten you? I'm glad that others are testing it. I've hit the pause button.

OpenAI, maker of the world’s most popular chatbot, ChatGPT, launched a web browser, Atlas, this week. Will it make surfing the Internet smarter?

Atlas is available only for computers that run Apple’s MacOS operating system. The company plans to introduce a version for Microsoft Windows and mobile operating systems, including Google’s Android and Apple’s iOS.

I tried it out on my iPad. It doesn't have a traditional address bar. You type the address into the chat window. That essentially removes competing search engines from the process. Google did something similar more than a decade ago with Chrome by integrating the browser and their search engine.

Atlas is very light on using your device's resources because all the heavy lifting is done in the cloud. 

The biggest criticism, or maybe it's a fear, that I've seen early on is that Atlas allows OpenAI to directly gather all user data that can train their future AI technologies. Microsoft (who clearly have a horse in this race) cautions that in exchange for this AI and lighter load, ChatGPT wants permission to watch and remember everything you do online. They say it "out-surveils even Google Chrome, and that’s saying something."

It not only keeps track of which websites you visit. It also stores “memories” of what you look at and do on those sites. It can even control your mouse and browse for you. It could complete an online order for you. (more on that tomorrow)

It is still early to evaluate whether Atlas’s AI capabilities outweigh its data gathering, but the privacy concerns are real and huge. Does OpenAI offer sufficient controls for managing what Atlas remembers? That's unclear. 

This has been the appeal of other browsers, especially DuckDuckGo, which emphasizes its privacy and is also a lighter browser than Chrome or Opera. (I consider Firefox to be somewhere between.) After all, your default browser is your entry point to almost all of your online surfing. (Yes, apps can bypass it.) But Duck Duck Go has a small percentage of the browser market.

Adding AI to browsers is not a new thing that OpenAI invented. Another lesser-known search engine, Perplexity, makes a browser called Comet. Google has added its Gemini bot to Chrome and will soon add "agent" capabilities that let AI do tasks for you, and Atlas has an“Ask ChatGPT” button that lets you chat with the bot about pages you’re viewing. You can ask it to summarize an article, analyze data, or revise your email draft.

OpenAI's response to concerns about privacy and data collection? So far, just a shrug. 

Australia’s nationwide ban on social media use for children under 16 takes effect today, making it the first country to prohibit underage users from major platforms outright. It is a noble and probably necessary thing, but I cannot believe it is doable.

Millions of accounts are expected to be affected as companies, such as TikTok, Instagram, Facebook, Snapchat, YouTube, and X, face fines of up to $33M for serious or repeated noncompliance. However, the law places responsibility on companies rather than families, and platforms must demonstrate that they have taken “reasonable steps," such as implementing age checks and removing suspected underage accounts.

I suspect the companies will say these things have already been put in place. (Have you noticed the increase in ads on TV and in your Instagram feed about their teen accounts?) And how will Australia monitor this? Critics say the law is difficult to enforce. It might push teens onto harder-to-monitor platforms. Enforcement may pose privacy risks. We know that many children who create accounts have already lied about their age. Can that be determined?

The research shows 96% of children aged 10 to 15 had used social media, and a majority had used a communication platform to chat, message, call, or video call others (94%). Anecdotally. many of them report encountering harmful content, grooming, or cyberbullying.

• Australia may have been the first country to pass a social media ban bill to stop children under 16 from accessing the platforms, but 7 other countries have taken similar steps, such as strict youth modes or time limits.
  https://studyinternational.com/news/countries-social-media-ban-children/  
• Read/download the Australian research report
  https://www.esafety.gov.au/sites/default/files/2025-07/Digital-use-and-risk-Online-platform-engagement-10-to-15.pdf