- AI Weekly Wrap-Up
- Posts
- New Post 11-5-2025
New Post 11-5-2025
Top Story
Nvidia is world’s first $5 trillion company
Premiere AI chipmaker Nvidia has become the most valuable publicly traded company in the world, and the first to crack $5 trillion in market capitalization. The AI gold rush has lifted Nvidia, with its 90% market share of advanced AI chips, into stratospheric regions of valuation. The AI tide is lifting many boats, including Microsoft at a $4 trillion market cap, and OpenAI, which is now preparing for a 2027 IPO (initial public offering) at a trillion-dollar valuation. (See story below.)

Clash of the Titans
OpenAI prepares for an eye-popping $1 trillion IPO in 2027
Fresh off its recent restructuring from a nonprofit to a for-profit benefit corporation (see story from last week’s issue), OpenAI has now set its sights on raising the massive amounts of capital that it will need to fulfill its myriad contracts to buy chips and build data centers. Multiple sources within the company have stated that they are preparing for an initial public offering of stock in 2027, with a target value of the company of $1 trillion. The current value of the company is $500 billion, as set by what recent venture capital investors have been willing to pay for its privately traded shares. A mere doubling of its valuation in 2 years is seen by insiders as a sedate pace compared to OpenAI’s meteoric rise from obscurity just 3 years ago to its current status as AI juggernaut.

CEO Sam Altman illustrating the steep ascent of his company’s valuation.
Google plans AI data centers in space
Google has announced Project Suncatcher, an ambitious program to launch AI data centers into low earth orbit. There they will be powered by solar panels, and communicate among themselves via high-speed optical links. Communication to the ground would be by high capacity radio links.
Although highly ambitious, this project, if successful, would solve the two major problems bedeviling current data center construction - access to power, and local resistance to the disruptions attendant on such projects. Solar power is both clean and abundant in space, and making the data centers off-world obviates the need for local permitting and politics.
Google expects to launch its first proof-of-concept satellite data centers in 20027, in collaboration with Planet, a satellite technology company.

Google plans a proof-of-concept satellite launch of in-orbit data centers in 2027.
Amazon says Rufus shopping bot on track to drive $10 billion in sales
Little more than a year after Amazon rolled out an AI-powered shopping assistant named Rufus, CEO Andrew Jassy stated in a recent quarterly earnings call that the bot is on track to drive $10 billion in increased sales annually. The company’s analysis shows that customers that engage with Rufus are 60% more likely to complete a purchase than customers who do not. Rufus helps users find products, compare options, and get quick answers to shopping-related questions. Apparently, this level of AI hand-holding is working.

Amazon CEO Jassy illustrates how much he cares about objections to replacing humans with robots.
Fun News
Reporter tries living 48 hours without AI - finds it’s a struggle
A New York Times reporter tried to live his life for 48 hours, without using any artificial intelligence (including the older and more ubiquitous type of AI known as machine learning.). His struggle revealed how embedded AI is in our daily lives.
Starting from when he awakened on the first day, he couldn’t unlock his phone with facial recognition. Once he unlocked it with a passcode, he couldn’t use any social media apps (they use AI to decide on what gets put into your feed), read email (they use AI-powered spam filters), or even read the news (70% of journalists use AI to help research and draft their stories.) He also couldn’t check the weather report, hail an Uber or taxi, ride the subway, pay with his credit card, or even use his laptop to write the story (its trackpad uses AI to disregard accidental touches.) He finished his story on a manual typewriter he had purchased days before especially for this article, typing by candlelight.

NYT reporter had to use a paper map to get around because GPS uses AI.
AI data centers may triple your electricity bill
The upswing in construction of power-gobbling AI data centers is starting to put pressure on local electricity rates all over the country. Power companies generally set their rates to spread infrastructure costs more or less equally among customers. When a huge new customer like an AI data center comes on the grid, the infrastructure improvements necessary to handle the increased load are spread over all ratepayers, even though just one massive customer is driving the need for those costs. In some areas, the increase in rates can be over 200% over a 5-year period.

Data center power needs are putting pressure on electricity rates.
Wharton report on AI in business - the CEOs are happy
A recent report out of MIT indicated that 95% of corporate AI projects were failing. Wharton’s survey of 800 business leaders has a different take. The Wharton researchers found that AI usage has gone mainstream, with 46% of business leaders using AI daily, and 82% using AI at least weekly. Nearly 75% of business leaders say they are already seeing positive return on investment in AI.
How can 95% of AI projects be failing (MIT’s report) while 75% of business leaders are seeing positive ROI? The 2 studies had radically different methodologies. MIT classified an AI project as successful only if it showed positive ROI in 6 months, a very high bar that most business software would fail. Wharton asked the CEOs if they were happy - a perhaps too-subjective criterion. What do the researchers say? That more research is needed, of course.

Korean Institute develops robots that you can wear like clothes
The Korea Institute of Machinery and Materials (KIMM) has developed an automated weaving system that allows the continuous mass production of “fabric muscle”, a lightweight, wearable textile that can assist the wearer’s physical movements. The textile is made from ultra-thin filaments of shape-memory alloy, metal that “remembers” and can return to a previous shape after being deformed. The lightweight fabric can lift 1000 times its own weight. This allows it to be incorporated into wearable robots that can assist lifting, bending, and standing. Such wearable robots will be highly useful for factory workers that need to repetitively lift heavy objects, or for patients with neuromuscular weakness, who need assistance to perform the daily activities of life.

Lightweight wearable shoulder-robot assists lifting.
Robots
Aurora opens second autonomous truck route
Autonomous vehicle startup Aurora Innovation has announced a second driverless truck route, from Fort Worth to El Paso. The company added this second, 600-mile route just six months after initiating its first route, the 240-mile trip from Dallas to Houston. The company was founded in 2017 by alumni of Google’s self-driving research effort, which eventually morphed into the leading robotaxi company, Waymo. Aurora has focused on autonomous vehicles for hauling freight, not people, and has started its rollout in Texas, which has a friendly, low-regulation posture toward autonomous vehicles on the highways. (Having driven in Texas, I can say from personal experience that I would be less apprehensive about robot drivers than many of the human ones I encountered. I expect that the robots would carry fewer firearms.)

Aurora’s self-driving trucks have a suite of sensors, including cameras, radar, and lidar (lasers).
WPI researchers build drones with echolocation like bats
Researchers from the Worcester Polytechnic Institute in central Massachusetts have taken inspiration from bats and developed drones that navigate using echolocation to avoid barriers. The goal is to make small (less than 4 inches to a side), lightweight (less than 4 ounces) autonomous drones that can fly in smoke, dust, and dark, where traditional camera-based navigation fails. They are designed to be released in swarms for search and rescue in hazardous and challenging environments, such as fires and other disasters.

This tiny drone uses echolocation to avoid a plexiglass barrier in a smoke-filled test lab.
AI in Medicine
Google releases personalized AI health coach for Fitbit users
Almost half-a-billion people wear fitness watches around the world, generating a massive real-time trove of physiologic data - and almost all of it is wasted. Insights from research on this data might yield new treatments for disease, or forewarnings of diseases to come. And often the users themselves don’t quite know how to use the data their fitness watch is so tirelessly producing. Google is taking a step to address this second problem by rolling out a personalized AI health coach for users of their Fitbit fitness watch who subscribe to their Fitbit Premium data analytics package. Users will get personalized feedback and advice, and they can ask health and fitness questions of the AI coach.

Subscribers to Google’s Fitbit Premium data analysis app can get AI-generated personalized health and fitness coaching.
World’s smallest 3-D bioprinter helps repair vocal cords
McGill University biomechanical engineers and surgeons have developed a prototype of a tiny 3-D printer that deposits hydrogels on vocal cords which have had lesions surgically removed. The hydrogels promote healing, and precise deposition of the gels helps restore the original shape and function of the damaged vocal cord. The prototype is mostly manually controlled, but the team is working on a semi-autonomous version that will increase the precision of gel deposition to improve the end result, and more closely restore the patient’s natural voice.

A 2.7-mm bioprinter delivers healing hydrogels to vocal cord lesions.