Awakend.ai tracks where AI value is moving: off trophy models and into the machines and balance sheets that run them.
This week, Yole forecasted 181,000 robotaxis by 2031, Boston Dynamics cut a finger from Atlas’s new hand, and Tesla trimmed memory in its Optimus chips to build more robots.
The longer look, last, is DNA. An AI read through 200,000 enzymes in 21 hours and flagged a pattern scientists had never confirmed before. Whether it can edit a gene is still unknown.
Today's Highlights (3 min read)
Robotaxi fleets are on track to reach 181,000 vehicles by 2031
Boston Dynamics removed a finger to make Atlas easier to train
Tesla cut chip memory to build more Optimus robots faster
AI flagged a new DNA pattern that looks like a gene-editing tool
Gene therapy is on track to be 5x by 2035
THE AI UPDATE
Yole Forecasts Robotaxi Fleets Growing to 181,000 by 2031
The global robotaxi fleet is on track to grow from about 10,000 vehicles today to 181,000 by 2031, an average growth rate of 78% a year, according to Yole Group. The sensors and onboard computers that let these cars drive themselves are forecast to get cheaper as mass production and simpler hardware scale up.
The bigger shift is that robotaxis are becoming profitable, not just more common. Yole found US operators break even once a car is running more than 40% of the time, since there’s no driver salary to cover. That combination, cheaper hardware and profitable operations, is why Yole expects robotaxis to start eating into new car sales directly, by 4% in China and 1% in the US by the mid-2030s.
Boston Dynamics Removes a Finger to Make Atlas Easier to Train
Boston Dynamics unveiled a new four-fingered hand for Atlas, dropping the pinky to simplify the design. The hand carries 13 degrees of freedom, with the thumb driving 4 and each remaining finger driving 3, plus tactile sensors across the fingertips and palm. The team tested the pinky’s usefulness by taping their own pinky and ring fingers together for a day, concluding the extra dexterity didn’t justify three more actuators, added cost, and more points of failure.
The hand is built for simulation-based reinforcement learning, where the AI trains in a virtual environment first, then transfers its learning onto the real robot. Fewer actuators make that transfer more reliable, according to Boston Dynamics. Hyundai says Atlas will start on factory jobs like parts sequencing in 2028, and reach component assembly by 2030. The company also aims to build 30,000 robots a year by 2028.
Tesla Trades Robot Brainpower for Mass Production
Tesla cut the memory in the chips that will power its next Optimus robots, aiming to build more robots faster. The next chip, called AI5, was first planned with far more memory, then cut in half to 72GB so it could be manufactured in higher volume, before Musk raised it back up to 96GB as a compromise. The following chip, AI6, was also cut by about a third to 144GB. Musk said the lower specs were “the only way to get enough volume for Optimus production," which also cut costs significantly.
The company claims the smaller memory won’t hurt performance, since what actually slows a robot down is how fast it pulls data from memory, not how much it can hold. Micron, a major memory chip maker, told investors that humanoid robots are expected to need about as much memory as driverless cars, more than 200 gigabytes and about double the AI5’s memory.
THE ANALYSIS
Did Artificial Intelligence Just Find a New Way to Edit DNA and Cure Genetic Diseases?
More than 300 million people worldwide live with a rare disease, and most of them are genetic. Only a small number have a treatment, because fixing a faulty gene safely requires a tool built exactly for that job, and those tools are rare. CRISPR is the best one scientists have. It took 36 years to turn it into medicine.
Finding a second tool like it has meant scientists combing through raw DNA by hand, one enzyme at a time, for decades. But Medical AI just did that search in 21 hours, and it found something nobody had confirmed before.
How CRISPR Started
CRISPR works like a pair of molecular scissors. Scientists point it at an exact spot in a person’s DNA and cut, removing a faulty gene or fixing one that causes disease. It’s already used to treat sickle cell disease and is being tested against cancer, blindness, and dozens of other conditions.
But CRISPR didn’t start that way. Scientists first noticed it in 1987, a strange, repeating pattern inside bacteria with no idea what it did. It took 25 years, until 2012, before anyone figured out how to turn that pattern into scissors humans could control.
950 Agents, One Day
Most searches like this work the same way. Compare a new sequence against a list of ones already known, and move on when nothing matches.
But Anthropic tried something different. About 950 AI agents spent 21 hours reading through more than 200,000 enzymes, not checking them off a list but reading the raw DNA itself, one letter at a time. Out of all of them, one agent stopped on a single stretch of code. In its own notes, it wrote: “I can see by eye a tandem repeat array.” Something was repeating, over and over, right next to an enzyme. It flagged the pattern for human scientists to check.
Meet ART
Think of it like three train cars linked together. The first car is an enzyme called a reverse transcriptase, which can copy RNA into DNA. Next to it sits a gene that controls when the enzyme switches on. Behind both sits a short stretch of DNA that repeats itself over and over, like a pattern stamped into a strip of fabric.
Scientists call this combination ART, short for array-associated reverse transcriptase. On its own, none of the three pieces means much. But every DNA-editing tool scientists have ever found, CRISPR included, is built from this exact same three-piece combination, an enzyme, a neighboring gene, and a repeating pattern. So when those same three pieces turn up together somewhere new, it’s a strong hint there might be another working tool hiding inside it.
Is It the Next CRISPR?
Maybe. Stanford’s Stanley Qi, a CRISPR researcher, said what stood out to him was the AI’s ability to spot “an unusual biological pattern that was difficult to detect before.”
But CRISPR only turned useful once scientists learned to reprogram it, swapping out its guide piece so it could cut any gene they chose. That took 25 years after it was first found. Nobody has shown ART can be redirected the same way yet, and no lab has used it to edit a gene. Until that happens, ART is a strong candidate, not a confirmed next CRISPR.
Why It Matters: The real story here isn’t ART. It’s how fast it showed up. Searching 200,000 enzymes by hand, the way scientists used to, could take years. An AI did it in 21 hours, and one of its search agents flagged ART by reading the raw DNA sequence directly, the same way a scientist would scan a page and notice a pattern.
That speed is what actually matters to patients. CRISPR took 36 years to go from a strange pattern in bacteria to a treatment that’s now curing sickle cell disease in real people. If AI can cut that kind of search down from years to hours, the next gene therapy, the next cure for an inherited disease, doesn’t have to wait a generation to get found.
THE MARKET
Gene Therapy Is on Track to 5x by 2035
Growth
The gene therapy market is worth about $13.68 billion in 2026 and is on track to reach $70.59 billion by 2035, more than five times larger in nine years. Genetic disorders are the fastest-growing disease indication in that forecast. North America holds the largest share, at 45%.Pricing
Gene therapies sell as one-time treatments, and the prices run into the millions. Casgevy, a CRISPR-based treatment for sickle cell disease, costs $2.2 million. Its rival Lyfgenia costs $3.1 million, and the hemophilia therapy Hemgenix costs $3.5 million a dose. ART has no price because no treatment exists yet.Adoption
In its second quarter 2026 report, the American Society of Gene & Cell Therapy counts 46 approved gene therapies and 2,217 more in the pipeline. Casgevy sales reached $76.4 million in the second quarter, up 78% from the first. On July 1, the FDA expanded Casgevy to children as young as 2 with sickle cell disease or a related inherited blood disorder, which makes about 5,500 more US children eligible. The FDA’s approved list also includes gene therapies for hemophilia and inherited blindness.
Takeaway:
The need is far larger than the supply. The NIH counts more than 10,000 known rare diseases, most of them genetic, and only a few hundred have safe, effective treatments. Gene therapy has 46 approved products and 71 more in the last stages before approval. Casgevy sold $119.3 million in the first half of 2026, more than the $115.8 million it sold in all of 2025. Sales are rising, the pipeline is deep, and each patient still costs $2.2 million.
The Timeline Guide:
Each timeline and graph represents the realistic stage of the covered technology, plotted from concept to scale, capturing where it stands today and when broad deployment is likely.
What did you think of today's issue?
If you think we could do better, where can we improve?
Hit reply with anything else on your mind; we actually read every note.
See you in the next upload
Note: Images in this newsletter are often AI generated for illustrative purposes only. Market forecasts are hypothetical, inherently uncertain, represent our best-guess estimates only, and should not be considered investment advice.
