‘Tis the season of the blockbuster AI (or AI-adjacent) IPOs. Elon Musk’s SpaceX has already filed for a $2 billion-ish valuation. The official investor roadshow kicks off tomorrow, and trading will commence on the NASDAQ on June 12, under the ticker symbol SPCX.

Now Anthropic, once viewed as OpenAI’s bratty kid brother, has filed confidentially for an initial public offering, before OpenAI could manage to do so, and possibly at a higher valuation than OpenAI will be able to command.

Anthropic is currently the hottest AI model-making company on the planet, outshining the OG, OpenAI, with astounding revenue growth (now $45 billion annualized, 5 times its run rate in December). This growth stems largely from eager adoption of its Claude Code product by an increasing number of large enterprises. Claude Code allows software engineers to delegate almost all of the actual code generation to the AI agent, while the engineer sets goals, decides on overall architecture, and verifies the correctness of the solution.

OpenAI is likely to file for an IPO soon, at a valuation near $1 Trillion. This may mean 3 IPOs of AI companies in 2026, all at or above $1 Trillion market capitalization.

Anthropic files for IPO

Premier AI chipmaker Nvidia became the most valuable publicly traded company in the world (the very first $5 trillion company) by gaining a hammerlock on the market for GPUs, the chips that power AI models like ChatGPT. Now CEO Jensen Huang is looking for new worlds to conquer.

On Sunday he unveiled a new superchip, the RTX Spark, in a bid for the $200 billion-a-year market for the central processing unit in PCs, called CPUs. GPUs, Nvidia’s current mainstay product, are designed to accelerate certain matrix calculations that are important in graphics (hence, GPU), and in AI models. CPUs are the engine for all the usual computer operations, like handling text or spreadsheets, loading data, etc. GPUs do the mathematical heavy-lifting for CPUs in gaming, crypto, and AI.

With the RTX Spark, Nvidia can supply a laptop or server with both the CPU and one of its GPUs. Major PC manufacturers, including Dell, HP, Lenovo, and Microsoft Surface will release Windows PCs this fall that are built around Nvidia CPUs and GPUs.

Jensen’s vision is that in the future, lots of AI will happen locally, on the end user’s own hardware, with the monster AI models that operate in the cloud being accessed only as needed. This has obvious advantages for privacy, security, and cost. Jensen believes apps and user interfaces will fade away, and people will just talk to their computer and tell it what they need or want.

Nvidia announces AI Spark superchip for laptops

Venerable socialist and cranky Vermonter Bernie Sanders thinks that if the AI companies are going to replace all (or most) human jobs, as they claim, then they should cede half of their stock to a US Sovereign Wealth fund. The fund will use those assets to pay for services for the displaced workers, such as training.

Sanders’ argument is that since AI models are trained on humanity’s collective data, the public deserves a cut of the proceeds.

Writing in an Op-Ed for the New York Times, Sanders reveals that he is proposing a new law, the American AI Sovereign Wealth Fund Act. It will create a federally managed fund that would hold stock in major AI companies such as OpenAI and Anthropic. The government will hold equal voting shares and Board representation to the equity shareholders, and thus be able to block any actions deemed harmful to the public. Over time, proceeds from the fund would be able to fund public goods such as education, housing, and job retraining.

Sanders points to successful examples of his approach, including Norway’s $2 trillion Sovereign Wealth Fund, which holds the profits from Norway’s North Sea oil production, and funds approximately one-quarter of the Norwegian government budget. Similarly, Alaska’s Permanent Fund, also based on oil revenues, sends an annual direct cash payment to each and every Alaska resident.

Bernie Sanders wants AI companies to give shares to the US public

A German AI startup named Shift is offering free housecleaning services for apartments in New York City. How are they able to do that? Each housecleaner wears a camera rig that films their movements as they clean. This data is anonymized and put into a dataset that will help train household robots. Shift gets paid by robotics companies for access to this training data.

The company indicates that they intend to expand their free-services-for-robotics-training-data model to other cities worldwide, and to other services such as repairs and errands.

AI startup will clean your apartment for free, if you let robots watch

One niche sector of robotics, powered exoskeletons for mobility assistance, is getting a boost from an unexpected quarter: mountain hikers. Originally conceived of as an assistive device for victims of stroke or other neurological diseases that hamper walking, exoskeletons are typically strap-on, battery powered motors that help lift the legs with each stride. AI is required for the device to adjust to the wearer’s stride and be truly assistive, rather than marching the wearer around like a puppet on a string. Advances in design and technology have made these devices cheaper and better, and they are now starting to show up on mountain trails, helping hikers climb more, or steeper, or higher mountains than they otherwise would be able to manage.

E-hiking is here, with AI-enabled exoskeletons

USC researchers have developed a robotic hand that can reproduce any tune that it hears after just two minutes of preparatory experimentation on the keyboard. Once booted up from scratch, the Musician Hand, which has four robotic fingers to press keys on a keyboard, will begin experimenting with pressing different keys to learn the notes, and then use different finger pressures on each key to learn how that modifies the sound. After two minutes of this directed experimentation, the robotic hand can immediately reproduce any tune that it hears.

In a blinded test, music judges listening to the playing of the robotic hand as well as that of four expert human piano players, could not distinguish between the humans and the robot.

The USC researchers believe that this demonstration project embodies a new paradigm for training robots, a method they call perceptual robotics. Rather than having robots pre-programmed for a task, they advocate giving robots elemental capabilities and the ability to experiment with their capabilities in a purposeful way. In this way, robots that can help with the gait problems of a Parkinson’s patient or collaborate with a construction worker on the job may arrive faster than we now foresee.

The Musician Hand teaches itself to play by ear in two minutes.

Fitness wearables are rapidly moving toward becoming consumer-facing generalized health apps. The latest entrant in this sweepstakes is Oura, the ring-shaped device that is highly prized by women for its light weight and unobtrusive form factor.

Oura has announced Health Radar, a new program which will track the user’s biometric data over time, looking for patterns that may indicate a need for lifestyle modifications, changes in exercise, or even consultation with a physician. To deliver these consultations to those who choose them, Oura has partnered with Counsel Health, a startup that aims to blend advanced medical AI with an in-house network of board certified physicians. The physicians can connect with the Oura user virtually within minutes, and order tests or prescribe medications. For more specialized care, Counsel Health can route the user to appropriate in-network, in-person services, with the goal of reducing visits to the ER or an Urgent Care facility.

Today’s health care system in the US is complex, hard to navigate, and hard to access. This opens the door for more nimble newcomers like Oura and Counsel Health.

Oura fitness ring announces new health initiative

The State of Utah has released a review of the first five months’ performance of the bot from AI startup Doctronic that was licensed under a pilot program to prescribe refills for medications.

They found that the bot recommended refill of the asked-for medication 72% of the time. A reviewing physician agreed with the bot’s recommendation to refill in 91% of the cases. Under this early stage of the program, 100% of the bot’s recommendations are being reviewed by a physician.

The 28% of the cases that were declined by the bot were sent for physician review. In 69% of these instances, the reviewing physician agreed that referral by the bot for physician review was appropriate. In the remaining 31% of cases, physicians felt that the bot was being over-cautious, and no referral to a human was actually needed. This indicates a bias toward conservatism and caution by the bot. The State of Utah wrote an approving comment on this caution, apparently relieved that the bot wasn’t taking risks.

We are still early days in this fascinating pilot program, but the data so far are mildly encouraging that these types of routine physician tasks may be amenable to AI assistance.

Utah releases review of Doctronic prescriber bot