Public transcript

How AI is Reshaping Modern Warfare: Eric Schmidt Explains

By: Catalaxis
Length: 17 min
Language: English
Text from: the video's captions
Transcribed: 14 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Former Google CEO Eric Schmidt discusses how AI is transforming warfare, using the Ukraine war as a test bed where sensors leave no place to hide, big platforms like tanks and carriers are sitting ducks, and the fastest learner wins. He warns that China is driving robot costs to thousands of dollars and will dominate low-end robotics, while the US leads in expensive frontier software but is hampered by slow procurement and a 17-year weapon development cycle. He closes by urging legal reform, mass production of dual-use defenses, and allied manufacturing so startups can supply the military.

Key points

Eric Schmidt argues that modern sensors leave no place to hide, making big platforms like tanks and aircraft carriers sitting ducks for drones.
He believes the fastest learner wins in war, crediting Ukraine's rapid innovation for holding off a numerically superior Russia.
Reinforcement learning AI can simulate millions of battles and discover strategies no human planner would think of, as shown by AlphaGo's move 37.
Schmidt says China is driving robot costs down to thousands of dollars via EV-derived hardware at scale, and will dominate low-end robotics.
The US leads in general intelligence software at huge capital and power cost, while China has the power but lacks chips and deep financial markets.
Because of China's civil-military fusion, Schmidt assumes every civilian product he saw has a hidden military counterpart.
The US military's 17-year average weapon development cycle and outdated systems mean equipment is obsolete by deployment.
Schmidt calls for mass-producible dual-use defensive systems, US and allied factories, and legal reform so the military can work with startups.

Questions it answers

00:00China's robotics lead

Why does Eric Schmidt think China will dominate low-end robotics?

Schmidt says China is driving humanoid and nonhumanoid robot costs from $10,000-$20,000 down to thousands of dollars using EV-derived hardware that scales, while America lacks access to those capabilities.

Chinese EV-derived robot hardware is extraordinary and scales
Robot costs dropping to thousands of dollars

00:47Ukraine as an AI test bed

What did Schmidt observe about AI in the Ukraine war?

Modern sensors see everything, leaving no place to hide; soldiers can only move at dawn and dusk, and big platforms like tanks and aircraft carriers become sitting ducks for drones.

Daytime and infrared cameras detect any movement
New US tank and carriers are vulnerable to next-gen drones

03:04Fastest learner wins

Why is the fight in Ukraine balanced despite Russia's numbers?

The fastest learner wins: Ukraine, outnumbered three or four to one, innovates faster than Russia, which is why Russia has failed to achieve its objectives.

Ukraine out-innovates Russia
Front lines empty of tanks and APCs because they're too dangerous

04:35Reinforcement learning and move 37

How does reinforcement learning change battle planning?

Reinforcement learning systems simulate millions of battles, invent strategies no human would think of, and execute them at machine speed, illustrated by AlphaGo's move 37 which stunned human experts.

Move 37 was the first computer-invented strategy invisible to human experts
War becomes millions of varied drones running learned combinations
Goal is mathematically raising win probability toward 100%

07:00US vs China AI split

How do the US and China compare in AI capabilities?

China wins on scaled, cheap hardware; the US leads in general intelligence models at huge cost. China will focus on consumer and enterprise AI for the rest of the world, and Chinese software may power most global interfaces.

Ideal but impossible combo: Chinese hardware plus American software
US needs 92 gigawatts more power; China has power but lacks chips
Chinese military AI is hidden behind civil-military fusion

10:46US procurement failures

Why is the US falling behind in military AI?

Procurement is broken: it acts as a jobs subsidy for legacy industries, takes about 17 years to field weapons, and wastes money, like million-dollar missiles against $27,000 Shahed drones.

Weapons obsolete by deployment; a Burroughs 6600 still runs in the Pentagon
99.9% of money goes to things Schmidt considers unimportant

15:20Fixing the system

What does Schmidt want America to change?

Mass-producible dual-use defenses, US factories with stockpiled parts plus allied factories in the Philippines, Japan and Korea, and legal changes so startups can cross the valley of death and the military knows who to call.

Change comes from crisis or collective will, not crisis preferred
Enormous revenue exists for startups in defense

Notes

How AI is changing war

Eric Schmidt calls the Ukraine war a test bed for AI-driven warfare: modern sensors mean there is no place to hide, with daytime cameras and infrared night cameras so good that soldiers can only move at dawn and dusk.
Big legacy platforms like the new US tank and aircraft carriers are vulnerable to next-generation drones and may be obsolete by the time they deploy.
His principle: the fastest learner wins. Ukraine is outnumbered three or four to one but innovates faster, which is why Russia has not achieved its objectives.

Reinforcement learning as a war advantage

Reinforcement learning systems simulate millions of battles and can invent strategies no human planner would consider, executing them at machine speed.
Schmidt cites AlphaGo's move 37 in Korea as the first gaming example of a computer discovering a completely new strategy invisible to human experts.
Future wars would need millions of drones of many kinds, with algorithms learning optimal combinations against the enemy's force projection.

China vs the US

Schmidt visited Chinese robot factories and concluded China will dominate low-end robotics, cutting humanoid and nonhumanoid robot costs from $10,000-$20,000 to thousands of dollars using EV-derived hardware at scale.
He describes the ideal scenario as Chinese hardware plus American software, which is geopolitically impossible.
The US pursues general intelligence at enormous capital and energy cost (he testified Congress needs 92 gigawatts more power), while China has power from solar but lacks chips and deep financial markets, so it will likely focus on consumer and enterprise AI for the rest of the world.
Because of civil-military fusion, he assumes every Chinese civilian product he saw has a military counterpart in a building he was not allowed into.

US failings and fixes

The US military is effectively a jobs subsidy program for existing industries, spends roughly a trillion dollars a year, and takes about 17 years to field a weapon system that is obsolete by then.
Examples of waste: million-dollar S-300/S-400 missiles shooting down $27,000 Shahed drones, and jets with Pentium 4 processors and a Burroughs 6600 still running in the Pentagon.
He wants mass-producible dual-use defensive systems, large US manufacturing with stockpiled parts, and allied factories in the Philippines, Japan and Korea.
Change requires either a crisis or collective will, plus legal reform so the single-source military customer can work with startups through the valley of death.

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