The trillion-dollar LLM race is over. The true disruption will be Agentic AI—autonomous, goal-driven systems—a trend set to dominate TechCrunch Disrupt 2025.

When OpenAI’s massive multimodal models were released in the early 2020s, the entire tech world reset. It felt like a gold rush, where the only currency that mattered was GPU access, trillions of tokens, and a parameter count with enough zeroes to humble a Fortune 500 CFO. For years, the narrative has been monolithic: bigger models, better results. The global market for Large Language Models (LLMs) and LLM-powered tools is projected to be worth billions, with worldwide spending on generative AI technologies forecast to hit $644 billion in 2025 alone.

This single-minded pursuit has created a natural monopoly of scale, dominated by the five leading vendors who collectively capture over 88% of the global market revenue. But I’m here to tell you, as an investor on the ground floor of the next wave, that the era of the monolithic LLM is over. It has peaked. The next great platform shift is already here, and it will be confirmed, amplified, and debated on the hallowed stage of TechCrunch Disrupt 2025.

The future of intelligence is not about the model’s size; it’s about its autonomy. The next billion-dollar companies won’t be those building the biggest brains, but those engineering the most competent AI Agents.

🛑 The Unspoken Truth of the Current LLM Market

The current obsession with ever-larger LLMs—models with hundreds of billions or even trillions of parameters—has led to an industrial-scale, yet fragile, ecosystem. While adoption is surging, with 67% of organisations worldwide reportedly using LLMs in some capacity in 2025, the limitations are becoming a structural constraint on true enterprise transformation.

We are seeing a paradox of power: models are capable of generating fluent prose, perfect code snippets, and dazzling synthetic media, yet they fail at the most basic tenets of real-world problem-solving. This is the difference between a hyper-literate savant and a true executive.

Here is the diagnosis, informed by the latest ai news and deep-drives:

• The Cost Cliff is Untenable: Training a state-of-the-art frontier model still requires a multi-billion-dollar fixed investment. For smaller firms, the barrier is staggering; approximately 37% of SMEs are reportedly unable to afford full-scale LLM deployment. Furthermore, the operational (inference) costs, while dramatically lower than before, remain a significant drag on gross margins for any scaled application.
• The Reliability Crisis: A significant portion of users, specifically 35% of LLM users in one survey, identify “reliability and inaccurate output” as their primary concerns. This is the well-known “hallucination problem.” When an LLM optimizes for the most probable next word, it does not optimise for the most successful outcome. This fundamentally limits its utility in high-stakes fields like finance, healthcare, and engineering.
• The Prompt Ceiling: LLMs are intrinsically reactive. They are stunningly sophisticated calculators that require a human to input a clear, perfect equation to get a useful answer. They cannot set their own goals, adapt to failure, or execute a multi-step project without continuous, micro-managed human prompting. This dependence on the prompt limits their scalability in true automation.

We have reached the point of diminishing returns. The incremental performance gain of going from 1.5 trillion parameters to 2.5 trillion parameters is not worth the 27% increase in data center emissions and the billions in training costs. The game is shifting.

🔮 The TechCrunch Disrupt 2025 Crystal Ball: The Agentic Pivot

My definitive prediction for TechCrunch Disrupt 2025 is this: The main stage will not be dominated by the unveiling of a new, larger foundation model. It will be dominated by startups focused entirely on Agentic AI.

What is Agentic AI?

Agentic AI systems don’t just generate text; they act. They are LLMs augmented with a planning module, an execution engine (tool use), persistent memory, and a self-correction loop. They optimise for a long-term goal, not just the next token. They are not merely sophisticated chatbots; they are autonomous problem-solvers. This is the difference between a highly-trained analyst who writes a report and a CEO who executes a multi-quarter strategy.

Here are three fictional, yet highly plausible, startup concepts poised to launch this narrative at TechCrunch Disrupt’s Startup Battlefield:

1. Stratagem

• The Pitch: “We are the first fully autonomous, goal-seeking sales development agent (SDA) for B2B SaaS.”
• The Agentic Hook: Stratagem doesn’t just write cold emails. A human simply inputs the goal: “Close five $50k+ contracts in the FinTech vertical this quarter.” The Agentic AI then autonomously:
  • Reasons: Breaks the goal into steps (Targeting $\rightarrow$ Outreach $\rightarrow$ Qualification $\rightarrow$ Hand-off).
  • Acts: Scrapes real-time financial data to identify companies with specific growth signals (a tool-use capability).
  • Self-Corrects: Sends initial emails, tracks engagement, automatically revises its messaging vector (tone, length, value prop) for non-responders, and books a qualified meeting directly into the human sales rep’s calendar.
  • The LLM is now a component, not the core product.

2. Phage Labs

• The Pitch: “We have decoupled molecular synthesis from human-led R&D, leveraging multi-agent systems to discover novel materials.”
• The Agentic Hook: This startup brings the “Agent Swarm” model to material science. A scientist inputs the desired material properties (e.g., “A polymer with a tensile strength 15% higher than Kevlar and 50% lighter”). A swarm of specialised AI Agents then coordinates:
  • The Generator Agent proposes millions of novel molecular structures.
  • The Simulator Agent runs millions of physics-based tests concurrently in a cloud environment.
  • The Refiner Agent identifies the 100 most promising candidates, and most crucially, writes the robotics instructions to synthesise and test the top five in a wet lab.
  • The system operates 24/7, with zero human intervention until a successful material is confirmed.

3. The Data-Moat Architectures (DMA)

• The Pitch: “We eliminate the infrastructure cost of LLMs by orchestrating open-source models with proprietary data moats.”
• The Agentic Hook: This addresses the cost problem head-on. The core technology is an intelligent Orchestrator Agent. Instead of relying on a single, expensive, trillion-parameter model, the Orchestrator intelligently routes complex queries to a highly efficient network of smaller, specialized, open-source models (e.g., one for code, one for summarization, one for RAG queries). This dramatically reduces latency and inference costs while achieving a higher reliability score than any single black-box LLM. By routing a question to the most appropriate, fine-tuned, and low-cost model, they are fundamentally destroying the Big Tech LLM moat.

🏆 Why TechCrunch is the Bellwether

The shift from the LLM race to Agentic AI is a classic platform disruption—and a debut at Tech Crunch is still the unparalleled launchpad. Why? Because the conference isn’t just about technology; it’s about market validation.

History is our guide. Companies that launched at TechCrunch Disrupt didn’t just have clever tech; they had a credible narrative for how they would fundamentally change human behaviour, capture mindshare, and dominate a market. The intensity of the Startup Battlefield 200, where over 200 hand-selected, early-stage entrepreneurs compete, forces founders to distil their vision into a five-minute pitch that is laser-focused on value.

This focus is the very thing that the venture capital community is desperate for right now. Investors are no longer underwriting the risk of building a foundational LLM—that race is lost to a handful of giants. They are now hunting for the applications that will generate massive ROI on top of that infrastructure. When a respected publication like techcrunch.com reports on a debut, it signals to the world’s most influential VCs—who are all in attendance—that this isn’t science fiction; it’s a Series A waiting to happen.

The successful TechCrunch Disrupt 2025 startup will not have a “better model.” It will have a better system—a goal-driven Agent that can execute, self-correct, and deliver measurable business outcomes without constant human hand-holding. This is the transition from AI as a fancy word processor to AI as a hyper-competent, autonomous employee.

Conclusion: The Era of Doing

For years, the LLM kings have commanded us with the promise of intelligence. We’ve been wowed by their ability to write sonnets, simulate conversations, and generate images. But a truly disruptive technology doesn’t just talk about solving a problem; it solves it.

The Agentic AI revolution marks the transition from the Era of Talking to the Era of Doing.

The biggest LLM is now just a powerful but inert, brain—a resource to be leveraged. The true innovation is in the nervous system, the memory, and the self-correction loop that transforms that raw intelligence into measurable, scalable, and autonomous value.

Will this new era, defined by goal-driven, Agentic AI, be the one that finally breaks the LLM monopoly and truly disrupts Silicon Valley? Let us know your thoughts below.

The commercial space race is heating up, and at its epicenter is Blue Origin, the aerospace company founded by Jeff Bezos. All eyes are on their massive heavy-lift vehicle, the New Glenn rocket, as it undertakes a pivotal mission—NASA’s groundbreaking ESCAPADE mission to Mars. This launch isn’t just a technical feat; it’s a statement about the future of reusable rockets and Blue Origin‘s challenge to the industry’s established giants.

Why the New Glenn Launch Matters

The New Glenn launch (specifically the NG-2 mission) marks a critical second flight for the colossal, 320-foot-tall rocket. Named after the first American to orbit Earth, John Glenn, this vehicle is foundational to Blue Origin‘s vision of millions of people living and working in space.

Here’s what makes this event so significant:

• NASA’s ESCAPADE Mission: The primary payload is NASA’s twin ESCAPADE (Escape and Plasma Acceleration and Dynamics Explorers) probes. These small spacecraft, nicknamed “Blue” and “Gold,” are headed to Mars to study how solar wind interacts with the Red Planet’s magnetosphere, an essential step for future human missions. This is New Glenn‘s first operational flight for NASA, demonstrating critical confidence in the burgeoning commercial launch sector.
• The Reusability Challenge: A key objective of the mission is the propulsive landing of the first-stage booster on the “Jacklyn” landing platform vessel in the Atlantic Ocean. The reusable first stage, powered by seven BE-4 engines, is designed for a minimum of 25 flights. A successful landing would be a huge leap for Blue Origin, positioning it as only the second company to achieve this feat with a heavy-lift orbital rocket, directly challenging the cost efficiency of competitors.
• Clearing the Backlog: Following its maiden flight in January, which successfully reached orbit but missed the booster landing, a successful NG-2 mission is vital for Blue Origin to accelerate its launch cadence. It is crucial for tackling a reported multi-billion-dollar backlog of customer contracts, including missions for satellite constellations like Amazon’s Project Kuiper.

The New Glenn Rocket: A Closer Look

The New Glenn is a giant, two-stage-to-orbit vehicle meticulously designed for maximum performance and cost-effectiveness:

The commitment to reusability is the core of Blue Origin‘s strategy. By recovering and reflashing the most expensive part of the rocket, the company aims to dramatically lower the cost of accessing space, making frequent and sustainable launches a reality.

The Road Ahead: Blue Origin and the Future of Space

The impending Blue Origin launch of New Glenn is more than just a single event; it’s a testament to the tenacity of the private space industry. With a successful launch and, more importantly, a recovered booster, Blue Origin will prove the operational maturity of their technology.

The success of the ESCAPADE mission will cement Blue Origin’s role as a trusted partner for deep-space exploration, demonstrating that commercial providers can reliably handle complex interplanetary missions for NASA and other global customers. As the countdown continues from Cape Canaveral, the space community holds its breath, waiting for New Glenn to further solidify its place in the history of spaceflight.

Introduction

Artificial Intelligence (AI) is rapidly advancing, and its impact on society is becoming more profound. The technology has the potential to revolutionize industries, improve healthcare, and even help solve global challenges like climate change. However, as with any powerful technology, there are concerns about its potential misuse and the need for regulation. The problem is, that the world is divided over AI, and geopolitics is getting in the way of global regulation.

The Geopolitical Divide

The divide over AI is not just about the technology itself, but also about the geopolitical implications of its development and use. The United States, China, and Europe are the three major players in AI, and each has its interests and priorities. The US is focused on maintaining its technological edge, while China is focused on becoming a world leader in AI. Europe, on the other hand, is focused on ensuring that AI is developed and used in a way that respects human rights and values.

The US and China are in a race to develop AI, and this competition is driving the development of the technology. However, this competition is also creating a divide over AI, as each country is focused on its interests and priorities. The US and China are not interested in global regulation, as they see it as a threat to their technological edge.

The European Union, on the other hand, is pushing for global regulation of AI. The EU has proposed a set of ethical guidelines for AI, which include principles like transparency, accountability, and non-discrimination. However, these guidelines are not legally binding, and there is no mechanism for enforcing them.

The Need for Global Regulation

The lack of global regulation of AI is a major concern. The technology has the potential to be used for both good and bad purposes, and without regulation, there is a risk that it will be used to harm people and society. For example, AI could be used to create deepfakes, which could be used to spread misinformation and manipulate public opinion.

Regulation is also needed to ensure that AI is developed and used in a way that respects human rights and values. For example, AI could be used to discriminate against certain groups of people, such as women or ethnic minorities. Regulation is needed to ensure that AI is developed and used in a way that is fair and inclusive.

The Challenges of Global Regulation

The challenge of global regulation of AI is that it is difficult to agree on a set of principles that are acceptable to all countries. The US and China are unlikely to agree to regulations that would limit their technological edge, while Europe is unlikely to agree to regulations that would compromise its values.

Another challenge is that AI is a rapidly evolving technology, and it is difficult to keep up with its development. Regulations that are put in place today may be outdated tomorrow, and there is a risk that they will stifle innovation.

Conclusion

The world is divided over AI, and geopolitics is getting in the way of global regulation of this powerful technology. However, the need for regulation is clear, as AI has the potential to be used for both good and bad purposes. The challenge is to find a way to regulate AI in a way that balances the interests of all countries and ensures that the technology is developed and used in a way that respects human rights and values.