The short version
- The students panicking that “AI will take all the jobs before I graduate” and the operators actually building the AI buildout are looking at completely different pictures.
- The four bets these operators name are the physical layer (power, transistors, materials), biology as an engineering discipline, agentic infrastructure and security, and infrastructure finance.
- Three of the four are less crowded than software was a decade ago — Stanford’s own EE enrollment has dwindled while the bottleneck moved straight into hardware.
- None of it is choosable by a 13-year-old. What is choosable now is the floor those fields stand on: circuits and transistors, systems and networks, code and models, and the habit of finishing hard things.
- Our Robotics & AI Mentoring program, grades 4–12, is built on exactly that floor — six engineering disciplines, one student per project, year-round.
If you are a high schooler planning majors and a first career right now, the AI supercycle is producing a set of opportunities that is genuinely counterintuitive. A lot of students have concluded that the sensible response to AI is to get out of the way of it. The people building the thing are saying something close to the opposite — and pointing at fields most students have already ruled out.
The advice below comes from founders and operators speaking to Stanford’s MS&E 435 class: leaders from OpenAI, Anthropic, Databricks, Vercel, Baseten, Applied Compute and Chai Discovery.
Where the operators say the value is going
Four bets, and the majors each one points to
Founders and operators from OpenAI, Anthropic, Vercel, Baseten, Applied Compute and Chai Discovery, speaking to Stanford’s MS&E 435 class.
The physical layer
Power, transistors and chip manufacturing are the real choke points of this era — and enrollment has moved the other way.
Sachin Katti (OpenAI) · Yash Patel (Applied Compute)
- Electrical Engineering
- Materials Science
- Physics
- Mechanical / thermal design
Biology as an engineering discipline
Molecular design is becoming predictable rather than alchemical — which raises, not lowers, the return on running real experiments.
Joshua Meier (Chai Discovery) · Eric Abrams (Anthropic)
- Computational Biology
- Bioinformatics
- Biophysics
- Molecular Biology
Agentic infrastructure and security
Agents that write and run their own code need sandboxes, gateways, networks — and an entirely new security surface to defend.
Guillermo Rauch (Vercel)
- CS — systems & networking
- Cybersecurity
- Computer Engineering
Infrastructure and project finance
Someone has to model how a gigawatt of compute gets funded, structured and amortised. Very few people can.
Tuhin Srivastava (Baseten)
- Finance
- Quantitative Economics
- Management Science & Engineering
1. Electrical engineering, materials science, applied physics
The media coverage has been almost entirely about software. The actual choke points of this technological era are physical: power, transistors, chip manufacturing, and heat.
Sachin Katti — who leads industrial compute at OpenAI and taught at Stanford for years — describes a severe talent shortage in exactly the place the bottleneck landed. Enrollment in electrical engineering dwindled through the software boom. Meanwhile the physical layer is where defensible, durable value is now accumulating. His advice to students is direct: study transistors, materials science, and physical engineering.
Yash Patel of Applied Compute says that if he weren’t running his own company, he would go straight into hardware and energy efficiency.
Majors: Electrical Engineering, Materials Science, Physics, or Mechanical Engineering with a thermal and cooling focus.
2. Computational biology and bioinformatics
Biology is in the middle of a historical transition — from a trial-and-error craft closer to alchemy into a predictable engineering discipline.
Joshua Meier, CEO of Chai Discovery, originally planned to be a doctor. He pivoted into code and biotech on a specific realisation: molecular therapeutics scale globally the way software code does. One good molecule reaches everyone.
The common assumption is that AI will make wet-lab work obsolete. Meier argues the reverse, and it is the more interesting claim. By making molecular design cheaper and better-targeted, AI raises the return on running a physical experiment — which triggers more lab work, not less. This is Jevons’ paradox pointed at a lab bench: efficiency raises consumption.
Eric Abrams, who leads biology work at Anthropic, takes it further. If AI can carry clinical trial design, target selection and regulatory filings, a single engineer could soon run a whole portfolio of drug candidates. He calls it a “pipeline in a person.”
Majors: Computational Biology, Bioinformatics, Biophysics, or Molecular Biology.
3. Computer science, systems, and cybersecurity
“Software is dead” is the loudest myth in this cycle, and it is wrong. Software is more plastic and more active than it has ever been.
Guillermo Rauch, CEO of Vercel, points out that his company is hiring software engineers faster than before, to build what he calls agentic infrastructure. As AI agents write more code, demand explodes for the things that code has to run inside: virtual sandboxes, custom API gateways, specialised networks.
And there is a security consequence that has barely been priced in. When an agent is handed a computer and allowed to compile and execute its own code, the attack surface changes shape entirely. AI-native cybersecurity is going to be one of the best-paid technical careers of the next decade, and there is almost no one trained for it yet.
Fastest-growing skills
The top three are exactly what a fixed robotics kit can’t teach
Skills employers rank as rising fastest in demand between now and 2030.
- 1 AI and big data
- 2 Networks and cybersecurity
- 3 Technological literacy
- 4 Creative thinking
- 5 Resilience, flexibility and agility
Note what the top three are — and note that none of them are learned by watching. They are learned by building and breaking systems.
Majors: Computer Science with a systems, networking or security focus, or Computer Engineering.
4. Infrastructure finance and economics
Building the supercycle is, underneath the technology, a multi-trillion-dollar capital allocation problem. Funding a single gigawatt-scale data centre runs to tens of billions of dollars — Tuhin cites a figure around $70 billion for an OpenAI Stargate-scale build.
Tuhin Srivastava, CEO of Baseten, argues that studying infrastructure and project finance is one of the highest-leverage paths available. Modelling how to fund, structure and amortise the energy, land and chip capex is now a first-order need at the biggest investment firms — and almost nobody trained for it arrives with the technical fluency to understand what they are financing.
Majors: Finance, Quantitative Economics, or Management Science & Engineering.
The mindset rules matter more than the major
The part that transfers to a 14-year-old
Three mindset rules that outlive any major
Your student picks a major in four to eight years. These are usable on Tuesday.
- 1
“Take a deep breath and calm down”
Ali Ghodsi, CEO, Databricks
Revolutionary technologies take decades to show up in productivity statistics, because organisations have to be rebuilt around them first. The electric dynamo took roughly forty years. Twitter panic is not a planning horizon.
- 2
“Study what is fun, then adapt”
Tuhin Srivastava, CEO, Baseten
A motivated person can get to world-class in a narrow technical niche in about six months. That makes learning-how-to-learn the durable asset and any single undergraduate curriculum a starting point, not a destination.
- 3
“Run toward the hairy problems”
Yash Patel, Applied Compute
When you join anything, volunteer for the messiest work nobody wants — for him at OpenAI that was evals and post-training. It is the fastest route to being indispensable, and the expertise compounds.
Two of the three cut directly against how families are told to plan. “Study what is fun and adapt” is not permission to be unserious — it is a claim that motivation is the scarce input, and that a motivated person reaches world-class in a narrow niche in about six months. “Run toward the hairy problems” is a strategy a student can start using in ninth grade, in a class, on a team, in a project group.
Focus on learning how to learn, rather than treating any undergraduate curriculum as a rigid career destination.
How our Robotics & AI Mentoring program fits this
Here is the honest problem with everything above: a twelve-year-old cannot choose a major, and a fifteen-year-old choosing one has four years of prerequisites in front of them. Advice this good is useless if it arrives the summer before college applications.
What can be built now is the floor these four bets stand on — and, more importantly, the evidence that this particular student can actually do hard technical work without being pushed.
4–12
grades
8
students per cohort, max
2×
sessions weekly, year-round
6
engineering disciplines
From grade 4 to the choke point
What each bet actually requires — and where it gets built
Nobody teaches a middle schooler semiconductor physics or project finance. What can be built early is the floor those fields stand on.
Physical layer
Prerequisite
Comfort with circuits, transistors, power, heat and physical constraint
In the program
Hands-on electrical work — circuits, resistors and transistors, Arduino microcontrollers, motors, torque and gear ratios, 3D-printed parts
Computational biology
Prerequisite
Programming plus real data-and-model literacy applied to messy experiments
In the program
Python, data structures, ML models and computer vision — trained on the student’s own noisy sensor data, not a clean textbook set
Agentic infrastructure
Prerequisite
Systems, networks, operating systems and security fundamentals
In the program
Computer architecture, OS, networks, embedded Linux, cybersecurity, APIs, databases, full-stack and IoT builds
Infrastructure finance
Prerequisite
Quantitative modelling and the ability to defend a number to a skeptic
In the program
Competitive programming and USACO/ACSL math-and-algorithms rigour, plus unscripted project defense in front of judges
Three things about the design are deliberate, and each maps to something a speaker above said.
There is no fixed kit and no seasonal game manual. A competition robotics kit teaches mechanical design inside a fixed parts list — students rarely touch raw electronics. Our students work with circuits, resistors, transistors and microcontrollers directly, which is the exact literacy Katti says has gone missing. The ceiling on a build is the student’s own skill, not a kit specification.
Every student owns and defends their own project. Not a shared team robot — their own, demoed to judges, unscripted, answering questions they didn’t rehearse. That is the “run toward the hairy problems” habit installed early, and it is also the only thing that reliably distinguishes a student later.
The AI is built, not consumed. Computer vision, ML models, generative AI and agents — applied to the student’s own noisy sensor data, on hardware they wired themselves. A twelve-year-old in our program built an object-detection robot car: onboard camera, vision model, real-time driving decisions. The car isn’t the achievement. Hitting four walls and getting past all four is.
The mechanism
One loop, run continuously from grade 4 to grade 12
Every trait in the section above is a by-product of this cycle. Nothing here can be lectured into a child — it has to be run.
- 1 Choose builds Agency
The student picks the project. No rulebook, no assigned game.
- 2 Build builds Craft
Real hardware and real code — sensors, circuits, models they wrote.
- 3 Break builds Resilience
It fails. Usually in front of someone. A mentor sits with them in it.
- 4 Ship builds Obsession
They fix it because they want the thing to exist, not because it is graded.
- 5 Defend builds Fluency
They demo it live to judges and answer hard questions unscripted.
Then back to step one, with a harder problem the student chose themselves.
Two sessions weekly, year-round, grades 4–12, in Sunnyvale and Cupertino or online. One hands-on robotics, AI and hardware/software session; one online session for AP Computer Science, USACO, ACSL and competitive programming — the quantitative rigour that the finance track and every engineering track both sit on. Mentors are practicing Silicon Valley engineers. Cohorts are capped at eight, because the moment a group is large enough to hide in, the outcomes stop being individual.
See the program in full, the High School Tech Hero track for older students, or what our students have built and defended. Related reading: what Michael Moritz looks for in a founder, and why it starts in middle school, and where VEX and LEGO leagues help and where they cap out.
Aasquare Academy — Robotics, AI & Computer Science Mentorship. Grades 4–12. Sunnyvale & Cupertino, CA.
Speaker remarks are drawn from sessions of Stanford’s MS&E 435 course. Titles and affiliations are as given at the time of speaking; the World Economic Forum skills ranking is from the Future of Jobs Report 2025.