This is a short summary based of the YT video by Coldfusion. <iframe width="700" height="350" src="https://www.youtube.com/embed/-q2n5DkDoMQ" title="OpenAI is Suddenly in Trouble" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe> ## The 4 buckets of issues according to Coldfusion ### Scaling problems The scaling law suggests that the more compute and data that is thrown into machines, the better they learn and bigger and better LLMs will be born. Originally published in [scaling Laws for Neural Language Models](https://arxiv.org/abs/2001.08361), this was a key idea that led to GPT-3 ( a much better version of GPT-2). ChatGPT was essentially a wrapper over GPT-3 to democratise "AI" amongst the general public. Then came GPT-4 which was even better and bigger. But, GPT-5 was a bit of a dud in comparison. The scaling law seemed to be falling now. So, it does not look like more resources pumped in will produce a much bigger and better model. The value is not quite clear anymore. This has also in a way led to this feature confusion. Open AI had a bunch of cool releases at one point ([[Unpacking "12 Days of OpenAI" - Insights and Experiments]]), but the value of these have become increasingly unclear. SORA for example just bleeds money. There are going to be "ads" in ChatGPT now, even though Altman had initially suggested that moving onto ads would be a "last resort". One can't help but consider these as "panic releases" to re-capture a market that is seeming to be increasingly drifting away due to a lack of clarity on value. ### Competition When ChatGPT came out first, it was a game changer. OpenAI was hot. Google were nowhere near, their release of Bard (Gemini) was a disaster. But now, the tables have turned. Gemini has quickly taken up market share. Remember the "code red" Altman announced when Nano Banana Pro was released? Personally, I find that a bit weird because its just an "image generator". If OpenAI was committed to solving the "bigger problems" of the world as it did once announce in its younger days, why would it be so bothered about an image generator being better their own models? Then, there are also the cheaper, open-source models from Chinese competitors that are offering some decent punches to OpenAI's customer base. ### Financials The requirements of funding are vast, while the approach to generating "shareholder value" is quite lacking. In Altman's words, there is "no plan to be profitable anytime soon". While this might make sense given how OpenAI is a tech company and [tech companies take time to be profitable, OpenAI is also guzzling up an unprecedented amount of money](https://sherwood.news/markets/openais-planned-cash-burn-unlike-anything-ever-seen-now-doubling-it/). Expected losses in 2026 - $14 billion. Also, its weird how much money is moving around within OpenAI and its partner ecosystem. OpenAI needs money. Needs funding. It's competitor, Google does not need external support. That is a huge advantage. Also, OpenAI moving into serving erotica. That can only be seen as a way to increase revenue. ### The CEO's track record The video also calls to attention Altman's not-so-great track record of keeping promises. Huge claims, lies and an inability to be clear about "value generation" are all character quirks working against him at the moment. ## Worthy angles to dive into in later pieces Here are some areas I would like to dive into later. Putting these out here as signposts for me. On the scaling law plateau - The broader question of what comes after scaling - is it reasoning, memory, agents, something else? This is a live debate with no settled answer, and has huge implications for where the industry goes. On AI monetisation - How do you actually make money from AI? The ad-supported model, the subscription model, API pricing - none have clearly "won." Worth comparing how different players (Google, Anthropic, Meta, Mistral) are approaching this differently. On open-source vs. closed AI - DeepSeek and others have shown that open-source models can punch well above their weight. What does that mean for the moat that closed labs thought they had? Is openness actually a competitive advantage now? On the "value" question - There's a broader piece here about hype vs. real enterprise adoption. A lot of companies are spending on AI but struggling to show ROI. This connects to the scaling plateau — if the models aren't getting dramatically better, the business case gets harder to make. On geopolitics and AI competition - The China angle deserves its own piece. Export controls on chips, DeepSeek's efficiency gains, what it means for the US-China tech rivalry specifically in AI. On AI leadership and accountability - The Altman track record point could generalise into a broader piece about the cult of the AI CEO — Altman, Musk, Hassabis — and whether the industry's credibility problem is partly a leadership credibility problem.