Video Summary
Signs of an AI bubble bursting: price hikes, access limits and dropping model quality.
Companies face steep GPU and data‑center costs that undermine subsidized pricing plans.
Anthropic and others are moving to per‑use billing, raising costs for power users.
User complaints and reduced engagement show AI is less sticky for general audiences.
Open‑source and local models can lessen dependence on costly cloud AI but trade performance.
He cites company pricing changes (Anthropic moving to per‑use billing), users reporting reduced access and falling model quality, rising GPU/data‑center costs, delayed/cancelled data‑center projects, and declining public engagement with AI products.
Companies are shifting from subsidized flat subscriptions to usage‑based billing (token or per‑agent charges). That makes heavy use much more expensive, exposing the true operational cost of running large models and reducing the viability of cheap plans.
Mudahar warns that allowing private AI services access to your computer raises privacy and security concerns, since companies could control or monitor devices—creating risks regardless of the provider's nationality.
Yes—open‑source and local models can run on consumer GPUs and reduce cloud dependence. They typically underperform top commercial models but are improving (e.g., quantization like Google's Turbo Quant) and lower recurring cost exposure.
"There is, in fact, a serious bubble that needs to pop for the benefit of all of us who enjoy tech."
Mudahar discusses the ongoing AI bubble that many have begun to recognize, contrasting the previous skepticism he faced a year ago. He emphasizes that something as overvalued as AI technology requires correction.
The negative impact of AI's rapid development on established systems, such as the Internet Archive, is highlighted as he expresses concern over large AI companies ruining valuable resources.
Mudahar believes that the eventual popping of the bubble, while potentially economically damaging, could lead to long-term benefits. Drawing a comparison to the 2008 financial crisis, he assures that it will be tough initially but often results in improvements post-crisis.
"When it pops, the damages... will probably resemble that of like 2008."
Mudahar warns that the economy might face severe repercussions when the AI bubble bursts, due to the extensive investment that both individuals and companies have made in these technologies.
He acknowledges the potential rough period that could ensue, reminding viewers that during hard times, perseverance often leads to better outcomes.
The increases in operational expenses for AI companies, driven by a demand for high-performance GPUs, are discussed. He elaborates on the financial strain these companies face as they provide access to their services.
"Anthropic changed their pricing based on AI use, which is a sign of shifting costs."
He delves into Anthropic's changes to their billing model, which reflects the rising costs associated with AI usage. This change signifies a move from subsidized pricing to a more sustainable approach for companies that utilize AI.
The conversation emphasizes how this adjustment could alarm users, as they may no longer benefit from the previously low-cost access to premium models.
The implications of this pricing strategy lay bare the reality that many power users may not find the service effective if they are suddenly facing soaring fees.
"Many users have been complaining that their usage is getting cut in half, and the quality of the model seems to be dropping."
Users on Anthropic's subreddit have raised concerns about the reduction in model efficiency and usability. The perception is that the AI's performance is degrading as costs rise, which is not ideal for those relying on its capabilities.
Mudahar notes a potential underperformance or "labotomization" of the AI to minimize expenses, suggesting that cost-saving measures may compromise the technology’s quality.
The growth trajectory of AI companies is examined critically, acknowledging the unsustainable levels of expansion that may lead to eventual failures if not managed properly.
"I think that's a scary world to live in, letting a private company access your computer."
Mudahar expresses his concerns over the integration of AI into personal technology, cautioning that granting companies access to private systems is fundamentally risky.
He urges caution when allowing any entity, regardless of nationality, to operate within one's personal computing environment, highlighting the importance of privacy and security.
The increasing reliance on AI in workplace settings poses challenges in scaling the user base while maintaining service quality, which requires robust infrastructure and investment.
"This is a technology that cannot scale to the levels that a lot of these guys are billing you."
Mudahar points out the steep costs associated with utilizing advanced AI models, referencing specific pricing structures from platforms like Open Router that complicate budget considerations for users.
He emphasizes that while many companies promote the accessibility of their models, the actual expense for high-performance AI systems can become prohibitive.
Despite his appreciation for companies like Anthropic for holding ethical standards, he reiterates the financial implications that come with such advanced technologies, signaling a disconnect between user expectations and reality.
"So these bills, depending on how much you use this AI, do tend to rack up, especially if you're going by the usage when you're paying per use."
Companies charging for AI services, like Anthropic, often have significant costs associated with token usage, leading to high bills if one is utilizing these services frequently.
For instance, there's a charge of $5 per million input tokens and around $25 for the million output tokens, making the overall price quite expensive.
The analogy of buying gas illustrates this cost structure as well; paying for premium gas translates to a higher price, akin to the higher costs of advanced AI capabilities.
"The reason why I talk about this AI bubble kind of crashing is that I've been following the news carefully."
The rapid growth of AI technology has created unrealistic expectations, and the current market is beginning to experience a reality check.
Many AI models seem to be receiving less functionality over time, suggesting they are being "lobotomized" in order to sustain economically viable operations.
Companies may initially operate at a loss to gain market share; however, we are now seeing price increases and a potential plateau in growth as the costs of providing AI services become apparent.
"When it comes to the U.S. alone, about half of the data centers that were supposed to start this year have either been delayed or have been cancelled."
Data center developments face significant challenges, including delays and cancellations, which affects the growth of AI capability.
Amid rising data center costs, companies like Microsoft may promise stable local pricing, but the infrastructure development expenses inevitably lead to increases in overall costs for consumers.
The integration of data centers into local communities can produce a sense of alienation as they often do not benefit the local population but rather profit large tech corporations.
"There's a scenario where current wars... are making things more expensive for tech investments."
External factors such as global conflicts are putting additional strain on tech industries, particularly AI, by increasing the prices of essential materials like helium, which is necessary for semiconductor production.
The AI market is showing signs of a significant downturn, with a growing realization that the initial hype is dissipating and sustainability is uncertain.
"Unfortunately, a lot of these tech companies try to build a million different things, and it seems like many people just aren't really sticking around with it."
Despite the initial novelty of AI-generated content and platforms, the public is exhibiting fatigue toward AI technologies, leading to a decline in user engagement.
People prefer authentic human interaction and content over AI-generated material, which lacks the personality that users crave.
Even in practical applications like coding, the audience for sophisticated AI tools remains relatively niche, which may not be sufficient to maintain the long-term viability of such offerings in the market.
"Why not just Google search yourself and read the top 10 results without burning through tokens to begin with?"
This statement criticizes the practicality of using AI for information retrieval, suggesting that for most users, traditional search engines may be more efficient.
The speaker conveys a belief that AI's usability has stagnated for the general audience, only appealing to specialized power users who are willing to expend resources on advanced applications.
There's an acknowledgment that while AI technology remains available and evolving, its application for the average person is limited and may not provide substantial advantages over standard search methods.
"There's a lot of open-source stuff you can download to generate text, videos, code."
The speaker emphasizes the availability of open-source AI tools that enable users to perform a variety of tasks without relying on large, costly AI systems.
He notes that while these local AI solutions may not match the performance of larger, commercial AI models, they still offer a viable option for many users to conduct basic functions and explore AI capabilities.
Advancements like Google's Turbo Quant demonstrate the feasibility of running substantial models on a single GPU, showing that there is progress in making powerful AI applications more accessible.
"Once people realize we're not getting the real return on investment, that crash is going to suck, but it's a much needed correction."
The speaker reflects on the impending economic corrections within the tech sector as companies face challenges in delivering sustainable value.
He asserts that as costs rise and consumer technology becomes exorbitant, especially in gaming hardware, there will be pressure for a re-evaluation of market expectations and product pricing.
This situation suggests a future where consumers might struggle with ownership of technology, potentially leading towards a reliance on cloud computing provided by large AI corporations.
"I just hate the fact that these companies have basically priced everything I love out of existence."
The frustration expressed indicates that while technological progression is essential, the current trajectory may lead to unaffordability and exclusivity for many innovations.
The speaker's view posits that AI and advanced technologies often come at a cost that overlooks the needs of the average user, leading to a loss of beneficial tools in everyday life.
Ultimately, the sentiment conveys a desire for a shift in the industry to ensure that technological advancements genuinely serve the wider public rather than create barriers to access.
