The AI agent inflection point
Speakers
Deirdre Bosa — Reporter
— CNBC anchor, tech reporter
Summary
- The AI industry has reached a critical "inflection point," shifting from Chatbots (answering questions) to Agents (executing workflows).
- OpenAI's growth is stalling; new premium subscriptions peaked at 1.7 million/month last year but have flattened, while competitors like Anthropic are surging ($1B run rate in 6 months).
- This shift implies "CapEx exhaustion" is unlikely; in fact, spending must increase because Agents require significantly more inference compute (running tasks for hours vs. seconds).
- The "Agent War" is fragmented and undecided, unlike the "Chatbot War" which OpenAI clearly won.
Trade Ideas
"Gemini Three... faded fast, arguably because it was still a better chatbot in a market that is already moving past chatbots." OpenAI's growth "has been stalling right at that inflection point." The market leaders of the previous cycle (Chatbots) are struggling to pivot effectively to the new cycle (Agents). A "great model launch isn't enough if you're winning the wrong war." This suggests potential stagnation for incumbents relying on chatbot dominance. Watch for signs of successful pivot to agents; currently, they are losing momentum to Anthropic and specialized tools. These companies have massive resources and could acquire or brute-force their way back to the lead.
"Anthropic's Claude Code... hit $1B run rate in just six months." Conversely, "Chinese models are showing us that this can be done a lot more efficiently." Momentum has shifted away from the incumbent (OpenAI) toward Anthropic (performance) and Chinese models (efficiency). As the "Agent War" begins, capital and user growth are flowing to these newer, more specialized or efficient players rather than the legacy chatbot winner. Long the emerging winners of the Agent cycle. (Note: Anthropic is private; look for proxies or private market access. Chinese AI offers an efficiency hedge). Regulatory crackdowns on Chinese tech or API restrictions.
Deirdre Bosa
Anchor/Reporter, CNBC Tech Check
When asked if this shift makes CapEx exhaustion more likely, Bosa states: "It probably makes the case for more spending... Agents take just a huge, huge amount of [inference compute]." She notes agents run for hours across multiple models. The market fears AI spending will slow down. However, the transition to "Agents" (which operate autonomously for long periods) creates an exponential increase in demand for inference compute compared to simple chatbots. This ensures sustained demand for Nvidia's chips and broader AI hardware. Long hardware/compute providers as the "Agent" phase requires significantly more processing power than the "Chatbot" phase. Enterprise adoption of agents could be slower than anticipated due to security or hallucinations.
