Public transcript

Dr. Eric Schmidt: The Future of Technology at 300

By: Special Competitive Studies Project
Length: 41 min
Language: English
Text from: the video's captions
Transcribed: 14 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

In a conversation with Tom Shanker at a Special Competitive Studies Project event, Dr. Eric Schmidt discusses AI and the future of warfare, drawing on his trips to Ukraine, where he says the transparent drone battlefield has frozen the front and made traditional armor obsolete. He assesses that China is now within about six months of the US in AI despite chip export controls, and argues that traditional programming is essentially over, replaced by managing AI coding tools. He closes with views on energy and data center costs, the risks of open-source Chinese models, and a proposal for mandatory AI-tools education in universities and high schools.

Key points

Ukraine is the laboratory for future war: soldiers stay in foxholes for 50 days to months, supplied by drones, because anything that moves is destroyed.
The battle has shifted to the mid-strike area roughly 40 to 100 km from the front, where supply lines are targeted; Ukraine has hit a monthly quota of 35,000 Russian targets.
China released Deep Seek V4 and, using the slower Ascend chip and new software techniques, is now within about 6 months of the top American AI models.
Schmidt argues that exquisite, big, slow, heat-emitting platforms like tanks, ships, and artillery are now vulnerable, and the US should shift toward automation and robotic systems.
Traditional programming is over: programmers now manage AI tools like Claude and Gemini with objective functions, a change that started around October of last year.
A gigawatt data center costs on the order of 50 billion dollars; the real limit on AI is cash, not energy, and only a few countries can finance it.
Open-source Chinese models like Deep Seek V4 could be modified by bad actors, creating risks that proprietary American models do not pose.
Schmidt proposes a mandatory freshman-year university class on using AI tools, and says education should shift from rote learning to asking creative questions.

Questions it answers

00:00Ukraine and the drone war

What did Schmidt observe about the drone war in Ukraine?

The front is largely frozen; soldiers live in foxholes for 50 days to months, supplied by drones, because anything that moves is destroyed. The war is now fought largely in the mid-strike area 40-100 km away, where supply lines are bombed.

Ukraine is the laboratory for future war on a transparent battlefield.
Ukraine hit a monthly quota of 35,000 Russian targets.
Drones work without GPS and through electronic warfare.

06:12US-China AI competition

Who is ahead in AI, the US or China?

China's Deep Seek V4, built on the slower Ascend chip with new software techniques, puts China within about 6 months of the top US models. Only a few countries can compete at this level, but China is fully committed to AI leadership.

US chip export controls worked but are beginning to fail.
China's models are open source and globally diffused.
China is within 6 months, a nanosecond in Schmidt's world.

08:48Robotic future of warfare

How will AI reshape combat and US defense planning?

Anything exquisite, big, slow, and heat-emitting is now vulnerable, so the US should automate land, sea, and air. Budgets will likely go up, not down, but spent on different things, like drone equivalents of 155 artillery.

Future combat will be largely robotic and automated under human control.
Traditional platforms like tanks and artillery are effectively obsolete.
America has a clear model of human control that Schmidt helped write.

11:00The end of programming as we knew it

What is happening to programming as a career?

Traditional coding is over; programmers now manage AI tools like 10 Claude or Gemini friends with objective functions and verification. The top programmers who manage this productivity will earn enormous returns.

The shift started around October of last year.
Anyone coding the traditional way should stop.
Top software people will be fewer but far more valuable.

15:30Training data and energy

What does AI need to advance — training data and energy?

AI has moved from language to action with VLA models, and war requires predictive battlefield analysis and reinforcement learning, which need massive training data. Data centers are moving behind the meter with their own power, and a gigawatt facility costs about 50 billion dollars.

Ukraine has the best training data but keeps it classified.
The real limit to AI is cash, not energy.
The military is building its own classified data centers.

23:00Kissinger, open-source risk, and global order

What did Kissinger worry about, and what are the risks of open-source AI?

Kissinger feared US-China AI competition could destabilize the world and started a track-two dialogue. Open-source Chinese models could be weaponized by a lone actor, and the world lacks agreed language on limiting powerful models, though the Trump administration wants to start.

Roughly 10 very powerful models will exist across the US and China.
American proprietary models let authorities know who to call.
Multi-stakeholder agreement is needed to get ahead of the danger.

27:00Audience Q&A

What did Schmidt say about jobs, human values, science, and education?

He expects the economy to grow and labor shortages rather than mass unemployment, says humanists and values alignment are entering AI companies, defends non-partisan science funding, and proposes a mandatory freshman AI-tools class plus creative rather than rote teaching.

AI job losses are concentrated in entry software and low-paid customer service.
American-values models like Gemma 4 and Nemo Tron are emerging.
Universities should teach students how to use these powerful tools.

Notes

Ukraine as the Laboratory for Future War

Dr. Eric Schmidt describes the front line as largely frozen: soldiers live in foxholes for 50 days to 3-4 months, supplied by drones, because anything that moves is destroyed.
Ukraine recently announced hitting its monthly quota of 35,000 Russian targets; the war has moved to the mid-strike area roughly 40-100 km deep, where supply lines are attacked.
Drones now operate without GPS, communicate through electronic warfare, and will increasingly act in coordinated swarms attacking a target from 360 degrees.

US-China AI Competition

A week before the talk, China released Deep Seek V4, built on the slower Ascend chip with new software approaches to work around hardware limits.
A year ago Schmidt estimated China was 1-2 years behind; the latest analysis puts them within 6 months.
Chinese models are open source and broadly diffused globally, which Schmidt flags as largely uncontrolled and potentially dangerous in the hands of a lone bad actor.

The End of Traditional Programming

Schmidt says he is "in mourning" for his identity as a programmer: traditional coding is essentially over.
Programmers now set up 10 Claude or Gemini "friends" with objective functions, verify their output, and let them work overnight.
He says this shift started around October of last year and tells anyone writing code the old way to stop.

Energy and Data Centers

The AI boom has contributed 1-2% of US GDP; Schmidt believes AI is under hyped, not over hyped.
The grid cannot handle the load, so companies build remote, behind-the-meter, air-cooled data centers with their own power plants.
A gigawatt data center costs roughly 50 billion dollars; the real limit to AI is cash, not energy.

Values, Science, and Education

Kissinger worried US-China AI competition could destabilize the world and started a track-two dialogue.
Human values in AI matter: American proprietary models allow authorities to "know who to call," unlike open-source Chinese models.
Schmidt proposes a mandatory freshman AI-tools class in every university, then high schools, and says teaching should move from rote memorization to creative questioning.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.