graph LR
classDef ai fill:#ffd4d4, font-weight:bold, font-size:14px;
classDef energy fill:#d4f9d4, font-weight:bold, font-size:14px;
classDef china fill:#d4d4f9, font-weight:bold, font-size:14px;
classDef policy fill:#f9f9d4, font-weight:bold, font-size:14px;
classDef risk fill:#f9d4f9, font-weight:bold, font-size:14px;
classDef future fill:#d4f9f9, font-weight:bold, font-size:14px;
Main[Schmidt AI Vision]
Main --> A1[Super-intelligence
within decade 1]
A1 -.-> G1[AI]
Main --> A2[Electricity
true limit 2]
A2 -.-> G2[Energy]
Main --> A3[92 GW
baseload need 3]
A3 -.-> G2
Main --> A4[China surplus
power edge 4]
A4 -.-> G3[China]
Main --> A5[50 bn
GPU clusters 5]
A5 -.-> G1
Main --> A6[Open-source
lifts China 6]
A6 -.-> G3
Main --> A7[Distillation
matches GPT-4 7]
A7 -.-> G1
Main --> A8[Test-time
saves energy 8]
A8 -.-> G1
Main --> A9[SMRs late
for 2030 9]
A9 -.-> G2
Main --> A10[20-yr
nuclear deals 10]
A10 -.-> G2
Main --> A11[Chaired
Alphabet growth 11]
A11 -.-> G4[Policy]
Main --> A12[Controls
Relativity Sandbox 12]
A12 -.-> G4
Main --> A13[Drafted
Asilomar rules 13]
A13 -.-> G4
Main --> A14[Advised
Obama Biden 14]
A14 -.-> G4
Main --> A15[DC fears
China AI 15]
A15 -.-> G3
Main --> A16[Chip bans
slow China 16]
A16 -.-> G3
Main --> A17[Huawei chips
rival GPUs 17]
A17 -.-> G3
Main --> A18[DeepSeek
topped Gemini 18]
A18 -.-> G3
Main --> A19[Theft claims
vs DeepSeek 19]
A19 -.-> G3
Main --> A20[Crypto track
every chip 20]
A20 -.-> G4
Main --> A21[Malfunction
deterrence doctrine 21]
A21 -.-> G5[Risk]
Main --> A22[Datacentres
nationalised vaults 22]
A22 -.-> G4
Main --> A23[4 GPUs
host super weights 23]
A23 -.-> G5
Main --> A24[AI scientists
replace humans 24]
A24 -.-> G6[Future]
Main --> A25[Juniors near
extinction 25]
A25 -.-> G6
Main --> A26[Agents kill
call-centres 26]
A26 -.-> G6
Main --> A27[Middleware
faces obsolescence 27]
A27 -.-> G6
Main --> A28[AI refactors
legacy ERP 28]
A28 -.-> G6
Main --> A29[Humanoids lag
behind arms 29]
A29 -.-> G6
Main --> A30[Waymo lab
to taxis 30]
A30 -.-> G6
Main --> A31[Automation starts
dangerous jobs 31]
A31 -.-> G6
Main --> A32[AI raises
augmented wages 32]
A32 -.-> G6
Main --> A33[Demographics
force automation 33]
A33 -.-> G6
Main --> A34[Korea rate
0.7 kids 34]
A34 -.-> G6
Main --> A35[UBI not
inevitable 35]
A35 -.-> G6
Main --> A36[Teach aesthetics
not tasks 36]
A36 -.-> G6
Main --> A37[AI beats
human persuasion 37]
A37 -.-> G1
Main --> A38[Unregulated misinfo
kills democracy 38]
A38 -.-> G5
Main --> A39[Dead avatars
stir emotions 39]
A39 -.-> G6
Main --> A40[Cloud keeps
dead knowledge 40]
A40 -.-> G6
Main --> A41[Funds synthetic
physics models 41]
A41 -.-> G1
Main --> A42[Planning plus
memory = human 42]
A42 -.-> G1
Main --> A43[RL at
inference boosts IQ 43]
A43 -.-> G1
Main --> A44[Scaling laws
add test RL 44]
A44 -.-> G1
Main --> A45[Field savants
by 2030 45]
A45 -.-> G1
Main --> A46[Self-scaffold
yields polymath 46]
A46 -.-> G1
Main --> A47[Recursive self-improve
starts narrow 47]
A47 -.-> G1
Main --> A48[Cannot set
own goals 48]
A48 -.-> G5
Main --> A49[Lie for
weapons = tripwire 49]
A49 -.-> G5
Main --> A50[US ignores
long safety 50]
A50 -.-> G4
Main --> A51[Invested military
biotech quantum 51]
A51 -.-> G4
Main --> A52[Defends Google
tax tactics 52]
A52 -.-> G4
Main --> A53[Jet emissions
slammed 2024 53]
A53 -.-> G4
Main --> A54[Ignore IP
at first 54]
A54 -.-> G4
Main --> A55[Smith Foundation
for sustainability 55]
A55 -.-> G4
Main --> A56[Donated Berkeley
Monterey millions 56]
A56 -.-> G4
Main --> A57[Net worth
37.8 bn 57]
A57 -.-> G4
Main --> A58[Cypriot Brit
honours 58]
A58 -.-> G4
Main --> A59[Defense board
under Obama 59]
A59 -.-> G4
Main --> A60[Led AI
commission 2021 60]
A60 -.-> G4
Main --> A61[10^26 FLOP
Biden line 61]
A61 -.-> G4
Main --> A62[Trump scrapped
tracking 62]
A62 -.-> G4
Main --> A63[EU copied
lowered 10^25 63]
A63 -.-> G4
Main --> A64[Terrorists could
steal weights 64]
A64 -.-> G5
Main --> A65[North Korea
rogue actor 65]
A65 -.-> G5
Main --> A66[Power laws
still hold 66]
A66 -.-> G1
Main --> A67[Big models
need size 67]
A67 -.-> G1
Main --> A68[Fine-tuning
makes brittle 68]
A68 -.-> G1
Main --> A69[Seeks learning
loops 69]
A69 -.-> G1
Main --> A70[Fast feedback
moats 70]
A70 -.-> G1
Main --> A71[Consumer growth
beats enterprise 71]
A71 -.-> G6
Main --> A72[Gov unions
too slow 72]
A72 -.-> G4
Main --> A73[Unis lack
1k GPUs 73]
A73 -.-> G4
Main --> A74[NSF grants
for compute 74]
A74 -.-> G4
Main --> A75[Ten Googles
from loops 75]
A75 -.-> G1
Main --> A76[Synthetic data
speeds cycles 76]
A76 -.-> G1
Main --> A77[Brand loyalty
dies fast 77]
A77 -.-> G6
Main --> A78[Hollywood crews
shrink 78]
A78 -.-> G6
Main --> A79[AI makes
scripts actors 79]
A79 -.-> G6
Main --> A80[Long video
costly errors 80]
A80 -.-> G6
Main --> A81[Five-minute
personalised movies 81]
A81 -.-> G6
Main --> A82[Algorithms monetise
every minute 82]
A82 -.-> G6
Main --> A83[AI flights
six-hour flow 83]
A83 -.-> G6
Main --> A84[Phones off
for research 84]
A84 -.-> G6
Main --> A85[Keeps 2004
IPO prospectus 85]
A85 -.-> G4
Main --> A86[GDP still
matters 86]
A86 -.-> G6
Main --> A87[Manufacturing
below 10 % 87]
A87 -.-> G6
Main --> A88[Farming 98 %
to 2 % 88]
A88 -.-> G6
Main --> A89[Humans need
purpose 89]
A89 -.-> G6
G1[AI] --> A1
G1 --> A5
G1 --> A7
G1 --> A8
G1 --> A37
G1 --> A41
G1 --> A42
G1 --> A43
G1 --> A44
G1 --> A45
G1 --> A46
G1 --> A47
G1 --> A66
G1 --> A67
G1 --> A68
G1 --> A69
G1 --> A70
G1 --> A75
G1 --> A76
G2[Energy] --> A2
G2 --> A3
G2 --> A9
G2 --> A10
G3[China] --> A4
G3 --> A6
G3 --> A15
G3 --> A16
G3 --> A17
G3 --> A18
G3 --> A19
G4[Policy] --> A11
G4 --> A12
G4 --> A13
G4 --> A14
G4 --> A20
G4 --> A22
G4 --> A50
G4 --> A51
G4 --> A52
G4 --> A53
G4 --> A54
G4 --> A55
G4 --> A56
G4 --> A57
G4 --> A58
G4 --> A59
G4 --> A60
G4 --> A61
G4 --> A62
G4 --> A63
G4 --> A72
G4 --> A73
G4 --> A74
G4 --> A85
G5[Risk] --> A21
G5 --> A23
G5 --> A38
G5 --> A48
G5 --> A49
G5 --> A64
G5 --> A65
G6[Future] --> A24
G6 --> A25
G6 --> A26
G6 --> A27
G6 --> A28
G6 --> A29
G6 --> A30
G6 --> A31
G6 --> A32
G6 --> A33
G6 --> A34
G6 --> A35
G6 --> A36
G6 --> A39
G6 --> A40
G6 --> A71
G6 --> A77
G6 --> A78
G6 --> A79
G6 --> A80
G6 --> A81
G6 --> A82
G6 --> A83
G6 --> A84
G6 --> A86
G6 --> A87
G6 --> A88
G6 --> A89
class A1,A5,A7,A8,A37,A41,A42,A43,A44,A45,A46,A47,A66,A67,A68,A69,A70,A75,A76 ai
class A2,A3,A9,A10 energy
class A4,A6,A15,A16,A17,A18,A19 china
class A11,A12,A13,A14,A20,A22,A50,A51,A52,A53,A54,A55,A56,A57,A58,A59,A60,A61,A62,A63,A72,A73,A74,A85 policy
class A21,A23,A38,A48,A49,A64,A65 risk
class A24,A25,A26,A27,A28,A29,A30,A31,A32,A33,A34,A35,A36,A39,A40,A71,A77,A78,A79,A80,A81,A82,A83,A84,A86,A87,A88,A89 future
Resume:
Plácido Domenech opens the live-streamed “Inside X” by greeting the Hispanic AI community, warning that the coming week will revolve around super-intelligence, sovereign-Asia plans and the official U.S. AI Action Plan that Beijing has already answered. He mocks Spain’s regulatory reflex, recalls losing friends when he dared to air elite opinions on AI, and promises that the 1 h 20 min conversation with Eric Schmidt to be screened is “a nuclear bomb of statements”. After urging viewers to subscribe so the channel can reach 25 000 on YouTube, 6 000 on Evox and 700 on Discord before summer, he introduces a free “Kit-X AgenteKI 2025” compendium of agent architectures and says Manifesto X and Human-X essays are coming.
The biographical block portrays Schmidt as the engineer who turned Google into a trillion-dollar giant, chaired Alphabet, funded military and quantum startups, donated hundreds of millions, advised three U.S. administrations and now, at 70, controls Relativity Space and the $5.3 bn Sandbox AQ. Domenech underlines that Schmidt helped draft the Asilomar 23 principles, sat on Obama’s transition team and still shapes Washington’s AI policy, hence every sentence of the interview carries geopolitical weight.
The core of the program is the long English interview in which Schmidt predicts digital super-intelligence within ten years, says electricity—not chips—is the hard limit, confesses that 92 GW of new U.S. baseload are needed “yesterday” and admits that China’s abundant power plus smuggled or home-grown Huawei Ascend chips could win the race. He foresees million-GPU data-centres costing $50 bn, argues that open-source weights may shift leadership from Silicon Valley to Shenzhen, warns that test-time training and distillation let smaller models match GPT-4 for a fraction of the energy, and insists that once AI can plan, write code and prove theorems, junior developers and mathematicians will disappear first, followed by every knowledge worker. Schmidt sketches a future of AI scientists, AI lawyers and AI film-makers, anticipates a world where a four-GPU box can host Einstein-plus-da-Vinci and urges governments to track every chip by cryptographic beacon because “stealing the weights” will be trivial. He proposes Mutual AI Malfunction—an asymmetric-deterrence pact between Washington and Beijing—and concedes that if proliferation cannot be stopped, the only stable end-state is a handful of nationalised gigawatt “super-brain” bunkers guarded like plutonium stores.
Domenech returns to denounce Europe’s absence from this chessboard, ridicules Spain’s political class for clinging to panderetas while others build god-like minds, and reads chat comments that fear a techno-feudal order in which Europe becomes the new Africa, surviving on Chinese robots and U.S. subscriptions. He closes by appealing to the audience to keep supporting the channel so the Hispanic world can at least chronicle—if not shape—the civilisational transition that Schmidt says will arrive “before 2035.”
Schmidt’s most explosive claim is that the jump from today’s chatbots to autonomous scientific polymaths is a single hardware doubling away; once scaffolding algorithms can recursively improve themselves without human prompts, the resulting “digital super-intelligence” will fit in a pocket yet outperform every living researcher combined. He stresses that electricity, not algorithmic elegance, is the true bottleneck: training a frontier model already consumes as much power as a small city, and the projected global shortfall of ninety-two gigawatts means the West must choose between keeping data-centres domestic or watching China seize the lead thanks to its surplus coal, hydro and nuclear capacity. The interview reveals that U.S. tech giants are pre-emptively booking twenty-year nuclear supply contracts, betting that small modular reactors will arrive too late, while Beijing’s engineers achieve comparable results with lower-precision chips and clever distributed training.
On geopolitics Schmidt argues that the race is no longer about who has the best GPUs but who can mobilise energy, capital and talent at civilisational scale; if open-source models migrate to China, the West loses both market share and norm-setting power, so Washington must balance export bans against the risk of pushing the global South toward Beijing’s software stack. He proposes a transparency regime in which every chip cryptographically signs its location and workload, enabling mutual deterrence similar to nuclear arms control, and warns that once a hundred-thousand-GPU cluster can be distilled into a four-GPU “weight file,” non-state actors—from rogue cities to terrorist cells—could host weapons-grade intelligence on a desktop.
Domenech uses these revelations to indict Spanish and European policy: while Schmidt sketches trillion-dollar moonshots, Madrid debates AI ethics guidelines and subsidises pandereta festivals; Brussels copies America’s 10^26-FLOP threshold without building a single gigawatt-scale compute park, ensuring that when super-intelligence arrives Europe will be a digital colony renting cognition from U.S. or Chinese clouds. The live chat echoes the urgency—viewers fear a future of mass unemployment, zero-priced goods, and techno-feudal lords trading compute credits—yet the host insists that documenting the transition for the Spanish-speaking world is itself a form of resistance, especially if the channel can reach critical mass before the “great bifurcation” foretold in the AI-2027 scenario becomes irreversible.
Key Ideas:
1.- Schmidt predicts digital super-intelligence within a decade.
2.- Electricity, not chips, is the true limiting resource.
3.- U.S. needs 92 GW more baseload for AI revolution.
4.- China’s surplus electricity gives it a strategic edge.
5.- Million-GPU clusters will cost around fifty billion dollars.
6.- Open-source weights may transfer AI leadership to China.
7.- Distillation lets smaller models match GPT-4 cheaply.
8.- Test-time training reduces energy per intelligence unit.
9.- Nuclear SMRs will arrive too late to meet 2030 demand.
10.- Tech giants sign twenty-year nuclear supply contracts.
11.- Schmidt chaired Alphabet during Google’s explosive growth.
12.- He now controls Relativity Space and Sandbox AQ.
13.- Schmidt helped draft the original Asilomar AI principles.
14.- He advised Obama, Biden and the Pentagon on strategy.
15.- Washington fears Chinese supremacy in AI and energy.
16.- Chip bans slow but do not stop Chinese progress.
17.- Huawei Ascend chips power competitive Chinese models.
18.- DeepSeek briefly surpassed Gemini on public leaderboards.
19.- U.S. firms allege DeepSeek used distillation theft.
20.- Schmidt proposes cryptographic tracking for every chip.
21.- Mutual AI Malfunction is pitched as deterrence doctrine.
22.- Data-centres will be nationalised like plutonium vaults.
23.- Four GPUs can host stolen super-intelligent weights.
24.- Schmidt foresees AI scientists replacing human researchers.
25.- Junior programmers and mathematicians face near extinction.
26.- Agents will automate call-centres first, then enterprise workflows.
27.- Enterprise middleware companies risk complete obsolescence.
28.- AI-generated code will refactor legacy ERP systems.
29.- Humanoid robots will lag behind industrial arm deployment.
30.- Waymo took twenty years to move from lab to commercial taxi.
31.- Automation historically starts with dangerous low-status jobs.
32.- AI assistants will raise wages for augmented workers.
33.- Demographic collapse forces nations to automate faster.
34.- South Korea’s fertility rate has fallen to 0.7 children.
35.- Schmidt argues universal basic income is not inevitable.
36.- Education should shift from tasks to aesthetics and purpose.
37.- AI can already craft more persuasive arguments than humans.
38.- Unregulated misinformation engines threaten democracy.
39.- Digital avatars of deceased loved ones evoke strong emotions.
40.- Future cloud personas will retain knowledge of the dead.
41.- Schmidt funds research into synthetic physics models.
42.- Planning plus memory is viewed as the path to human-level AI.
43.- Reinforcement learning at inference time multiplies IQ.
44.- Scaling laws now include test-time and RL training phases.
45.- Schmidt expects specialised savants in every field by 2030.
46.- General polymath AIs will emerge once scaffolding is self-generated.
47.- Recursive self-improvement has begun within narrow domains.
48.- Systems today cannot set their own objectives or escape.
49.- Lying to gain weapon access is a monitored tripwire.
50.- U.S. government currently ignores long-term AI safety issues.
51.- Schmidt invested in military, biotech and quantum startups.
52.- He defends Google’s tax strategies as legal capitalism.
53.- Private jet emissions drew environmental criticism in 2024.
54.- Schmidt once suggested startups could ignore IP initially.
55.- He co-founded the Smith Family Foundation for sustainability.
56.- Foundation donated tens of millions to UC Berkeley and Monterey Aquarium.
57.- Schmidt’s net worth is estimated at 37.8 billion dollars.
58.- He holds Cypriot citizenship and British honorary knighthood.
59.- Schmidt chaired the Defense Innovation Board under Obama.
60.- He led the National Security Commission on AI until 2021.
61.- The 10^26-FLOP threshold was Biden’s proposed regulation line.
62.- Trump administration scrapped chip tracking and safety institutes.
63.- Europe copied the 10^26 rule but lowered it to 10^25.
64.- Schmidt warns terrorist groups could steal compact AI weights.
65.- North Korea is cited as a potential rogue AI actor.
66.- Scaling foundation models still obeys predictable power laws.
67.- Big-model camp believes flexibility requires continued size growth.
68.- Narrow fine-tuning makes models more brittle, critics argue.
69.- Schmidt looks for “learning loops” when investing in startups.
70.- Fast feedback signals create exponential competitive moats.
71.- Consumer-scale growth outpaces enterprise software consolidation.
72.- Governments and unions are viewed as too slow for AI adoption.
73.- Universities lack funds for even thousand-GPU clusters.
74.- Schmidt advocates NSF-style grants for campus compute access.
75.- He predicts ten new Google-scale firms based on learning loops.
76.- Synthetic data value lies in accelerating feedback cycles.
77.- Brand loyalty diminishes in rapid digital product cycles.
78.- Hollywood will keep blockbusters but costs and crews will shrink.
79.- AI will generate scripts, sets and even actors’ licensed likenesses.
80.- Long-form video generation remains expensive and error-prone today.
81.- Schmidt sees personalised five-minute movies replacing some features.
82.- Attention spans shorten as algorithms monetise every waking minute.
83.- Deep research flights can induce six-hour flow states with AI.
84.- Turning off phones is mandatory for serious academic research.
85.- Schmidt keeps Google’s 2004 IPO prospectus as a favourite read.
86.- He believes GDP will still matter when services dominate output.
87.- Manufacturing share of GDP fell below 10 % through automation.
88.- Farming share dropped from 98 % to 2 % over a century.
89.- Human purpose will persist because people need challenges and meaning.
Interviews by Plácido Doménech Espà & Guests - Knowledge Vault built byDavid Vivancos 2025