In 30 seconds
Musk estimates AI could add $20–30 trillion annually to the global economy, but investor Jo Bhakdi asks how people who lose jobs would benefit. In a discussion of Musk’s G20 clips, Bhakdi and host Herbert Ong debate the transition to higher incomes, a projected AI power shortage and the potential of computing in orbit. Bhakdi also predicts restrictions on advanced AI training, while Ong offers a different view of self-learning systems.
- 1Musk estimates AI could add $20–30 trillion a yearHe frames the number as a rough estimate and predicts major gains in digital work and coding.
- 2Bhakdi asks how displaced workers would share the gainsHe argues that growth alone does not provide income to people without jobs or stock holdings.
- 3Musk predicts a 15-gigawatt AI power shortfallHe attributes the projected 2027 gap to chip production growing faster than available power outside China.
- 4Bhakdi expects orbital computing to scale quicklyHe sees a large SpaceX opportunity but identifies falling AI demand as a risk.
- 5Bhakdi predicts restrictions on advanced trainingHe explicitly presents this as speculation about future oversight, not knowledge of a government plan.
- 6The speakers want a plan for the transitionThey ask who will help people between today’s economy and the abundance Musk predicts.
Takeaways
- Bhakdi argues that economic growth alone does not provide income to displaced workers.
- Musk predicts a 15-gigawatt AI power shortfall in 2027.
- Bhakdi expects orbital computing to expand rapidly, while identifying demand as a risk.
- The speakers ask for a concrete plan to manage the transition toward higher incomes.
Read the edited text by chapter
Musk’s AI forecast and the jobs question
The speakers react to his estimate of economic growth and debate whether displaced workers would share in it.
Host Herbert Ong introduces clips of Elon Musk at the G20 and a discussion with economist and investor Jo Bhakdi. Ong says Musk estimates AI could add $20–30 trillion annually to global output but sees electricity as a major constraint.
In the first clip, Musk estimates AI could increase the global economy by 20–30%. He predicts that by the end of the following year AI could perform digital tasks without physical manipulation and become so good at coding that people cannot compete, comparing it with the chess program Stockfish.
Bhakdi says he expects rapid, compounding AI improvement through the end of 2027. He predicts systems will be capable of doing much white-collar work, though he distinguishes capability from the actual speed of adoption.
Bhakdi worries that rapid automation could cause large-scale unemployment before a way to share the benefits exists. He questions whether the growth implied by AI spending will reach people who lose jobs and do not own stocks.
Ong brings up Musk’s idea of universal high income. Bhakdi questions whether such a system will exist soon enough; Ong replies that new jobs may also emerge, while Bhakdi distinguishes entrepreneurial opportunities from conventional jobs.
Ong says economic growth could help create abundance and support higher incomes. Bhakdi asks for a concrete path from current conditions to that outcome, arguing that a larger economy alone does not ensure displaced workers receive a share.
Both discuss a potentially difficult transition. Bhakdi says he does not see a sufficient plan for managing it and argues that building robots and rockets is a different task from deciding how societies adjust.
Electricity as a constraint
Musk predicts a power shortfall, and Ong discusses efforts to build generating capacity.
Ong says he has used AI agents to carry out computer tasks such as building or fixing a website. He then returns to the power question raised in Musk’s G20 remarks.
In another clip, Musk says analysts expect at least a 15-gigawatt power shortfall for AI chips in 2027. He attributes the gap to chip production growing faster than available electricity outside China.
Ong says data-center developers face delays obtaining power and gas turbines. He describes SpaceX’s effort to make turbine components as a way to bring facilities online sooner.
The investment case for orbital computing
Bhakdi expects space-based computing to grow rapidly, while identifying demand as a risk.
Bhakdi is bullish on orbital computing. He points to SpaceX and Nvidia’s announced work adapting AI hardware for space, and predicts substantial space-based capacity around 2028–29 followed by rapid scaling.
He argues that strong demand, constrained supply and national-security interest could make orbital AI valuable for SpaceX. Bhakdi identifies a collapse in AI demand as a risk, while saying he considers the company’s opportunity unusually large.
Ong describes the history of conflict between Musk and OpenAI, then says AI companies are renting computing capacity from SpaceX. He introduces another G20 clip to explain Musk’s account of the electricity bottleneck.
Renting compute and moving it into space
Musk describes leasing capacity and power plants; Bhakdi discusses inference and training in orbit.
In the clip, Musk says Google, Anthropic and others lease computing from SpaceX, which he says built its own power plants to bring capacity online. He argues that countries able to add power could attract data centers and collect taxes and fees.
Bhakdi says he favors space-based AI over relying on terrestrial power expansion. He distinguishes inference—running a trained model—from training, and argues that links between satellites could introduce delays for very large training clusters.
Could future AI need less training
Bhakdi and Ong offer different predictions about agents, safety and self-learning.
Bhakdi predicts that future AI systems may not need ever-larger training runs to deliver economic value. He imagines networks of agents built on sufficiently capable models and worries that pursuing much greater capabilities could increase risks.
He speculates that governments may eventually restrict advanced AI training and require oversight. Bhakdi acknowledges that this is his prediction rather than firsthand knowledge of a government plan.
Ong offers a different route to a similar conclusion: he expects self-learning systems eventually to reduce dependence on large pretraining runs. He compares the idea to a child learning over time.
Bhakdi compares inherited biological traits with results of an evolutionary learning process and says greater training could still make AI more powerful. His argument is that society may decide the additional capability is not worth the risk.
OpenAI, Cursor and competing ecosystems
The hosts discuss a proposed end to model access and changing company relationships.
The discussion moves to OpenAI and Cursor. Ong says OpenAI proposed ending direct model access after SpaceX acquired Cursor, and recounts reactions from Musk, Cursor’s leadership and Anthropic.
Bhakdi calls the dispute unsurprising and discusses how AI companies can simultaneously compete and collaborate. He speculates about alliances among SpaceX, Anthropic and Nvidia, and whether the market will support a competing ecosystem.
The speakers continue discussing relationships among AI companies and chip suppliers. Bhakdi says those relationships could shape access to models and computing, while acknowledging uncertainty over how partnerships will develop.
Following AI and planning the transition
The speakers discuss X as an information source and return to the unresolved question of how people adjust.
In a final clip, Musk recommends following AI discussion on X, saying that is where he gets much of his news. Ong and Bhakdi say they also use the platform to follow developments and expert debates.
Returning to the social transition, Bhakdi says forecasts of universal high income do not explain how people get through the intervening disruption. He wants more detailed discussion of who should plan that transition and what they should do.
Ong says he hopes to ask Musk about the difficult period between today and abundance. Bhakdi suggests asking who Musk thinks should take responsibility for working out a transition, rather than simply asking him to solve it alone.
More from this channel
21:07EN
36:44EN