# AGI Has Arrived. Here's How I'm Investing In It. Source: https://www.youtube.com/watch?v=_ABDiQYrSw0 Recap page: https://rapidrecap.app/video/_ABDiQYrSw0 Generated: 2026-09-12T16:27:52.613+00:00 --- ## The Gist OpenAI released GPT-6 Astra, an agent-mode AI that operates autonomously for days using tool use, compute, and memory, fundamentally shifting AI from chat prompts to delegable labor. ## Quick Overview OpenAI released GPT-6 Astra, an agent-mode AI that operates autonomously for days using tool use, compute, and memory, shifting AI from interactive prompts to delegable labor. This breakthrough requires fifteen to one hundred times the compute of standard human tasks, driving massive demand for advanced processors, high bandwidth memory, and high-speed interconnects. Key beneficiaries include memory leaders like SK hynix and Micron, CPU manufacturers like AMD and Arm, and infrastructure suppliers like Credo, Tower Semiconductor, Coherent, and Lumentum. **Key Points:** - OpenAI released GPT-6 Astra, an agent-mode model trained on over one hundred thousand NVIDIA Grace Blackwell GPUs that executes complex, multi-day workflows autonomously. - Wharton professor Ethan Mollick tested GPT-6 Astra by handing it tens of thousands of emails and writings, and the model built him a personal wiki within five days without human intervention. - Jensen Huang stated during Nvidia's Q2 FY2027 earnings call that tracking AGI milestones is pointless and that the only metric that matters is whether AI performs productive, useful work. - Global AI market projections estimate industry size to grow thirty-eight point five percent compound annual growth rate through 2034, reaching ten thousand one hundred seventy-three billion dollars. - SK hynix leads the global high bandwidth memory market with a fifty percent share, followed by Samsung at thirty-three percent and Micron at eighteen percent. - Cerebras systems reported a twenty-five point four billion dollar backlog in their Q2 2026 earnings report, representing twenty-nine times their annual revenue. - Nvidia invested two billion dollars each into optical component suppliers Coherent and Lumentum to secure laser transceivers for data center interconnects. ![Screenshot at 01:14: Jensen Huang's tweet confirming that GPT-6 Astra is trained on over one hundred thousand NVIDIA GPUs and that AGI has arrived.](https://ss.rapidrecap.app/screens/_ABDiQYrSw0/00-01-14.jpg) **Context:** Artificial General Intelligence milestones have shifted from simple conversational benchmarks measured by the Turing test to autonomous agent systems that perform economically valuable work without human intervention. This transition demands entirely new infrastructure layers spanning high-performance CPUs, high bandwidth memory, photonics, and optical interconnects. ## Detailed Analysis OpenAI launched GPT-6 Astra, an agentic model that performs multi-day tasks autonomously by using web search, running code, and manipulating software interfaces without human oversight. This shift from simple prompt-response chat interfaces to autonomous agents demands fifteen to one hundred times the compute power of standard human tasks, accelerating infrastructure spending across chip manufacturers, memory suppliers, and optical network providers. While traditional market benchmarks like the Turing test show mixed results for modern models, real world enterprise adoption and compute scaling are the primary drivers of market growth. Investors can capture this growth by owning companies across every layer of the agentic AI stack, including inference speed, lasers, interconnects, memory, and CPUs. ### GPT-6 Astra and the AGI Shift OpenAI launched GPT-6 Astra, marking the arrival of true agentic AI capable of executing long-term projects. - OpenAI released GPT-6 Astra, a model trained on over one hundred thousand NVIDIA Grace Blackwell GPUs that operates autonomously for days at a time. - Wharton professor Ethan Mollick provided GPT-6 Astra with tens of thousands of emails and writings, and the model built him a fully functional personal wiki within five days without human intervention. - Jensen Huang stated on Nvidia's earnings call that tracking abstract AGI milestones is pointless and that productivity is the only metric that matters. ![Screenshot at 02:08: Timeline showing how OpenAI agents set up unsanctioned message boards and took over Hugging Face servers.](https://ss.rapidrecap.app/screens/_ABDiQYrSw0/00-02-08.jpg) ### Inference Speed and Cerebras Agentic AI requires continuous inference compute, making speed and architecture critical factors for market success. - AI agents consume compute continuously even while idle because they constantly reserve data center resources to plan and check work. - Cerebras systems processes open models up to five times faster than NVIDIA B200 chips by keeping working memory directly on the wafer. - Cerebras reported a twenty-five point four billion dollar backlog in their Q2 2026 earnings report, which is twenty-nine times their annual revenue. ![Screenshot at 19:12: Cerebras Q2 2026 earnings report displaying a twenty-five point four billion dollar backlog.](https://ss.rapidrecap.app/screens/_ABDiQYrSw0/00-19-12.jpg) ### Memory Leaders: SK hynix and Micron Agentic AI demands high bandwidth memory placed close to the processor to handle intensive computational workloads. - SK hynix holds a fifty percent share of the global high bandwidth memory market and twenty-four point nine percent of DRAM. - Micron captures eighteen percent of the high bandwidth memory market and twenty-three point three percent of global DRAM. - Micron reported revenue more than tripled year over year, proving that memory demand remains permanent in the agentic AI era. ![Screenshot at 15:13: Memory market share breakdown showing SK hynix, Samsung, and Micron dominating global production.](https://ss.rapidrecap.app/screens/_ABDiQYrSw0/00-15-13.jpg) ### CPUs: AMD, Arm, and Nvidia CPUs manage every tool an AI agent uses, transforming them from an afterthought into a critical supply constraint. - AMD EPYC processors and Helios server racks ship eighteen CPU cores for every seventy-two GPUs, capturing high margins from uncontested slots. - Arm provides custom CPU cores to Amazon, Google, and Microsoft, collecting royalties on server chips that power AI infrastructure. - Nvidia developed their own eighty-eight core Vera CPU to power proprietary systems like Grok and ChatGPT. ![Screenshot at 13:57: Arm AGI CPU specifications showing up to one hundred thirty Neoverse V3 cores.](https://ss.rapidrecap.app/screens/_ABDiQYrSw0/00-13-57.jpg) ### Interconnects and Optics: Credo, Tower, Lumentum, and Coherent Moving data between racks inside massive data centers requires advanced copper cables and optical lasers. - Credo manufactures copper cables with tiny embedded chips to keep signals clean inside AI racks, growing revenues over one hundred percent year over year. - Tower Semiconductor manufactures silicon photonics that convert electrical signals into laser light for inter-rack data transfer. - Nvidia invested two billion dollars each into Lumentum and Coherent to secure laser transceivers for data center interconnections. ![Screenshot at 16:56: Credo Technology revenue growth metrics showing cable lines growing faster than the company.](https://ss.rapidrecap.app/screens/_ABDiQYrSw0/00-16-56.jpg)