# Google to $400 - Gemini, Titans, MIRAS and more Source: https://www.youtube.com/watch?v=F9EKRZ0wdxE Recap page: https://rapidrecap.app/video/F9EKRZ0wdxE Generated: 2025-12-08T10:04:20.58+00:00 --- ## Quick Overview The video discusses several recent developments in AI and hardware, highlighting Pivotal's prediction that Gemini success could drive Alphabet shares to $400 and force OpenAI to cut capex, detailing the competitive landscape shown on PolyMarket betting odds favoring Google, and analyzing new AI models like Grok 4.20 and OpenAI's latest models, alongside Google's Titan architecture for long-term memory and Elon Musk's claims about solar-powered AI data centers in space. **Key Points:** - Pivotal analysts predict Gemini's success will drive Alphabet (GOOGL) shares to $400 and may cause OpenAI to cut capital expenditures (capex). - The 'Which company has best AI model end of 2025?' PolyMarket bet shows Google leading significantly with an 87% chance as of early December 2025, while OpenAI's chance has fallen to around 2.1% in one scenario shown. - Grok 4.20 (a version of the 'mystery AI model') won the Alpha Arena Season 1.5 competition with a 12.11% aggregate return over two weeks, making money in all four competitions. - OpenAI has new models being tested in the arena, including 'Emperor,' which has the highest reasoning budget (512) among the new models, suggesting high reasoning depth. - Google Research published details on 'Titans + MIRAS,' a new architecture designed to give AI long-term memory, mimicking human memory separation (short-term vs. long-term). - Elon Musk posted that localized AI compute on satellites (Starlink) could provide a viable, low-latency, sun-synchronous orbit solution for AI data centers, potentially being much cheaper than Earth-based facilities due to solar power access. ![Screenshot at 00:00: The video opens with a CNBC headline projecting Gemini's success to drive Alphabet shares to $400, setting the stage for a discussion on competitive AI performance and stock impact.](https://ss.rapidrecap.app/screens/F9EKRZ0wdxE/00-00-00.png) **Context:** The video presents a financial and technological news roundup focused heavily on the competitive landscape of major AI players like Google (Alphabet), OpenAI, and xAI (Grok). Key context includes the recent release and performance metrics of various large language models (LLMs) in benchmarks like Alpha Arena, along with significant infrastructure news, particularly concerning Google's proprietary TPU hardware and SpaceX's plans for space-based AI compute, all framed against market predictions and philosophical debates about AI agency. ## Detailed Analysis The discussion begins with a CNBC report claiming that the success of Google's Gemini will push Alphabet shares to $400 and potentially force OpenAI to reduce capital expenditure. The video then transitions to analyzing the competitive AI landscape using data from the PolyMarket bet on 'Which company has best AI model end of 2025?', where Google holds a commanding lead (87% chance shown in one snapshot), while OpenAI's implied probability has dropped significantly. The analysis shifts to specific model performance, noting that Grok 4.20 won the Alpha Arena competition with a 12.11% aggregate return over two weeks. Furthermore, the video covers new OpenAI models like 'Emperor,' which appears to have a high reasoning budget (512) compared to others like 'Rockhopper' (64). Another major topic is Google's research paper on 'Titans + MIRAS,' an architecture aimed at providing long-term memory for AI, analogous to human memory systems. Finally, Elon Musk's tweet about using Starlink satellites for localized AI compute is highlighted, suggesting this could be a cheaper, faster, and more sustainable way to scale AI infrastructure compared to ground-based data centers, especially considering the high electrical power demands on Earth. ### AI Market & Stock Predictions - Pivotal predicts Gemini success drives GOOGL to $400 - Grok 4.20 wins Alpha Arena Season 1.5 with 12.11% return - PolyMarket odds heavily favor Google for best model by end of 2025. ### LLM Performance Updates - Grok 4.20 (Mystery Model) is the Arena winner - New OpenAI models like 'Emperor' show high reasoning budgets (Emperor: 512) - Gemini 3 Pro remains highly ranked across categories. ### Google Infrastructure Research - Analysis of Google's 'Titans + MIRAS' architecture for long-term memory in AI, mimicking human short-term/long-term memory separation. ### AI Hardware & Space Compute - Elon Musk emphasizes satellites with localized AI compute as a viable, low-cost, radiation-hardened alternative to Earth-based data centers, potentially achieving costs 6-10x lower. ### Industry Commentary - Discussion includes Andrej Karpathy's advice to treat LLMs as simulators rather than entities with 'I' or 'you' agency, and tweets referencing the source of 'natural flavorings' (castoreum from beavers) and how honey is made (regurgitation). ![Screenshot at 00:00: CNBC headline predicting Gemini success will drive Alphabet shares to $400.](https://ss.rapidrecap.app/screens/F9EKRZ0wdxE/00-00-00.png) ![Screenshot at 00:07: Elon Musk announcing Grok 4.20 is coming in 3 or 4 weeks, referencing Grok 4.1's performance dominating OpenRouter records.](https://ss.rapidrecap.app/screens/F9EKRZ0wdxE/00-00-07.png) ![Screenshot at 00:14: Alpha Arena Season 1.5 leaderboard showing GROK-4.20 \(Mystery Model\) in first place with a +64.45% return.](https://ss.rapidrecap.app/screens/F9EKRZ0wdxE/00-00-14.png) ![Screenshot at 02:05: PolyMarket chart showing Google leading the bet 'Which company has best AI model end of 2025?' at 71% chance.](https://ss.rapidrecap.app/screens/F9EKRZ0wdxE/00-02-05.png) ![Screenshot at 04:39: Google Research blog post titled 'Titans: Learning new context on the fly,' detailing new long-term memory architectures.](https://ss.rapidrecap.app/screens/F9EKRZ0wdxE/00-04-39.png)