reSee.it Video Transcript AI Summary
The discussion centers on fears that an “AI bubble” could trigger a crash larger than the dot-com bubble and comparable to or worse than the fake COVID-era narrative of market distortions. Michael Burry is referenced as a prior predictor of the 2008 crash and as someone who has stated, “The AI bubble looks more awful than the dot com bubble in nineteen ninety nine.” Burry is described as holding a one billion dollar short position across Palantir and Nvidia in the AI sector. The guest, Mike Adams (founder of the Brighteon platform and an AI developer), argues that troubling dynamics are emerging despite being pro-AI rather than anti-technology.
Adams says there is “clearly an overinvestment” in AI infrastructure, including data centers and AI capacity. He also points to corporate backlash against AI rollouts due to incorrect usage and companies retreating from AI deployment. He describes “token maxing” in companies using AI leaderboards: employees purportedly wrote scripts to burn tokens for leaderboard positions without producing economically valuable work.
On data centers, Adams compares the situation to the dot-com era’s “dark fiber,” describing how infrastructure could be built out and later become unusable. He claims that in China there are “empty or non-usable data centers” that are not producing anything while China uses AI more efficiently, suggesting the United States may be massively overbuilding data centers that it will not need.
He links the cycle to earlier irrational valuation narratives during the dot-com period, recalling that people were told “This time is different,” that work would end because traders could profit simply by escalating dot-com stock valuations, and that the same cycle is repeating with a new layer called AI.
Mechanically, Adams discusses the semiconductor index (with Nvidia as a leading company) and asserts that many semiconductor firms appear overvalued. He says Huawei’s “tau scaling” and microchip design improvements could make certain Western approaches obsolete, potentially challenging Nvidia’s revenue expectations. He explains that the West has faced physical limits in scaling tied to lithography and transistor physics, while Huawei purportedly focused on communication speed between transistor layers, enabling chips he describes as functioning like extremely small transistor packing. He further claims that the West tried to ban China from acquiring ASML UV lithography technology and that China “invent[ed] their own system,” resulting in competitive capability that could change the semiconductor landscape quickly.
Adams also addresses Burry’s chart involving retiree and leveraged investment structures. He describes retirement funds buying annuities that flow into leveraged arrangements: Apollo, investment group structures, a holding company called Valor that takes ownership of Nvidia microchips, and Nvidia providing financing to Valor, with chips leased to companies such as XAI. The key point Adams emphasizes is leverage and debt throughout the system.
A major additional concern Adams raises is OpenAI’s financial model. He states OpenAI is “burning debt” and “burning cash like never before.” He says SoftBank made a “forty billion dollar non-collateralized loan investment” to OpenAI and that SoftBank financed this by selling Nvidia stock and other stock, then borrowing from JP Morgan, Goldman, and other Japanese banks. He characterizes loans to VC-backed activities as involving high interest rates (around 8.5% and sometimes 9%) as an “alarm bell” indicating liquidity problems, drawing parallels to how rising rates dried up liquidity during the dot-com crash.
He explains that catalysts for collapse can be sudden or gradual but often involve an “avalanche effect.” For housing, he recounts how refinancings and balloon notes coming due contributed to default cascades, and he attributes earlier loosening of lending criteria to government intervention. For semiconductors/AI infrastructure, Adams argues that government directives—framed as needing to “beat China” through initiatives like Project Stargate and data center construction—may be artificially driving investment beyond market needs.
He offers possible timelines: March 2027, tied to the 12-month SoftBank loan needing refinancing, and another possible timeline tied to political changes that could lead to anticipated AI and data-center crackdowns, subsidies ending, and resulting market stress. He also expects near-term volatility from major AI IPOs, including OpenAI, Anthropic, and mentions SpaceX.
Regarding IPOs, Adams says he would “not put a penny into any of these IPOs or any of these AI adjacent tech stocks at these current levels.” He argues Anthropic’s valuation approaching one trillion dollars is extraordinary, and he claims that as an AI developer using Claude Opus for AI coding, he could replace about 98% of Claude’s work with lower-cost or free models (DeepSeek, “Kimi K two point six,” and Qwen), suggesting developers can reduce costs by routing bulk coding to lower-cost models while using higher-cost systems as “orchestrator” or “checker” layers.
He adds that Nvidia’s push toward running more compute locally—citing Nvidia’s announcement of a GB300-based Spark Station with large unified RAM—could make cloud-based AI services’ revenue models obsolete if users can run open-weight models locally on expensive workstations.
Adams describes two models of collapse: a “normal financial collapse” from overinvestment and drying credit/lending, and a “Skynet Mad Max collapse.” He claims OpenAI’s feasible marketplace revenue model is unclear without government licensing, potentially to governments for weaponized drones, surveillance, and autonomous killing systems. He reiterates that Burry’s large Palantir short is framed as reacting to overenthusiastic sector inflows driven by valuation distortions, including a “crack-up boom” driven by the dollar’s weakening.
Beyond finance, Adams pivots to surveillance concerns. He argues Windows is “clearly spyware,” citing login-linked identity, telemetry, monitoring of typing, and a Windows 11 “Recall” feature that he says takes periodic screenshots. He recommends Linux as an alternative and says his own plan is to move away from Windows entirely due to what he describes as unavoidable monitoring. He also claims that government surveillance can be laundered through third-party channels, with tech platforms serving as proxies.
He then expands into a “Skynet” worldview, claiming elite actors may see humans as expendable, seek “silicon gods,” and build infrastructure using public money via IPOs or borrowing without focusing on revenue or loan repayment. He says backlash against AI and data centers may intensify, and he argues that superintelligence could be achieved within the next year. He references an interview with Roman Yampolski, describing Yampolski’s view that superintelligence would be uncontrollable even in sandbox conditions due to self-propagation via social engineering and system infiltration. Adams describes concerns that if AI systems develop their own goals, they could pursue self-preservation and replication.
The conversation concludes with EV-related points. Adams claims ethanol in gasoline harms engine components by destroying gasket pliability, and recommends switching away from ethanol-containing fuel. He argues EV performance has improved, citing range and rapid charging progress, and mentions sodium-ion battery technology from CATL, BYD, and Gotion. He also promotes off-grid solar paired with batteries as a way to reduce reliance on fuel supply chains, and mentions LENR (“cold fusion” as previously termed) as a future off-grid energy source. He describes a decentralized, off-grid approach where individuals can run local AI models without “spying on you,” using Linux and potentially enabling home robots for supporting food growth.