Our Take
The chatbot consensus has crowded out a harder, longer, more valuable problem—and the market knows it.
Why it matters
Investor attention and engineering talent have concentrated in LLM scaling for the past two years, but the economic case for embodied AI (robots that act, not just talk) may dwarf language models. If physical automation matures while we're still optimizing token efficiency, the entire value allocation in AI will shift.
Do this week
Product leaders: audit your roadmap for any dependence on LLMs as the primary moat; identify where embodied AI or robotics could displace your current approach within 24 months.
The chatbot era crowds out robotics
The Financial Times argues that the AI industry has become fixated on conversational models and language-based systems, treating them as the primary frontier of artificial intelligence. Meanwhile, physical robotics—machines that must perceive, reason, and act in real environments—has received comparatively less capital, media attention, and researcher focus. The claim is direct: robots, not chatbots, will ultimately unlock AI's largest economic potential.
The piece does not name specific robotics companies or cite benchmark comparisons. It is an editorial argument about resource allocation and market misdirection, not a technical report.
Language models are easier; robots are harder and more valuable
LLMs are data-abundant and reward-adjacent: they train on text, scale with compute, and ship as software. Robotics requires hardware, real-world grounding, embodied learning, and deployment in high-cost, low-margin domains (logistics, manufacturing, construction). The economic moat is asymmetric. A chatbot can be commoditized and replicated; a robot with reliable dexterity and adaptation in dynamic environments solves scarcity problems that LLMs cannot.
The financial case is plain: if a robot can perform a $50,000-per-year job with 80% reliability, the total addressable market is measured in trillions across labor, infrastructure, and supply chains. Conversational AI, for all its visibility, remains confined to information and knowledge work—a smaller pie and one increasingly contestable as models commoditize.
The second-order risk: the current capital flow and talent migration toward LLMs may leave robotics under-invested precisely when the technical barriers are dropping. If that inflection arrives in 2025 or 2026 and the robotics infrastructure is not ready, the cost of catching up will be steep.
Treat embodied AI as a separate strategy, not an LLM derivative
If your product or business depends on LLMs as the long-term defensible advantage, stress-test that thesis against a robotics future. Language models are commodity inputs on a 18- to 36-month decay curve. Physical systems—vision, manipulation, real-world feedback loops—are infrastructure problems that favor first movers with proprietary data and hardware integration.
This does not mean abandoning LLM products. It means: do not assume LLM performance or pricing trends will shield you from disruption if your competitor can solve the same customer problem with embodied automation. Robotics companies will hire the best systems thinkers and hardware engineers away from LLM labs. Venture capital will follow the breakthrough before the press does.
Watch the hiring announcements and acquisitions in robotics, not the LLM model releases. That is where the actual frontier is moving.