This GTC feels structurally different from past ones because the narrative is no longer “more training compute”, but economics of inference. Markets will likely react to proof of monetisation, not just technological ambition.

Why GTC could be a catalyst

1. Inference is the larger TAM

Training is episodic; inference is continuous. If Nvidia demonstrates meaningful latency and cost-per-token improvements via SRAM + 3D stacking, it reinforces NVDA as the toll collector of everyday AI usage, not just model creation.

2. Architecture transition matters

A Feynman-class inference chip signals Nvidia defending against specialised challengers (Groq, custom ASICs, hyperscaler silicon). Integration rather than displacement strengthens ecosystem lock-in.

3. Customer validation

OpenAI purchase commitments and Meta scaling inference workloads suggest demand is shifting from experimentation to production deployment. Markets reward visibility of recurring compute demand.

Why it may become “sell the news”

1. Expectations are already elevated

NVDA trades on forward dominance. If gains are incremental rather than step-change economics, positioning could unwind.

2. Inference diversification risk

Meta’s CPU-based inference and hyperscaler in-house chips imply optimisation pressure on GPU pricing over time.

3. Capex fatigue narrative Investors increasingly ask about ROI, not capacity expansion.

Base case

GTC likely drives volatility, not a trend reversal.

Strong technical roadmap = medium-term bullish confirmation.

Short-term reaction depends on whether Nvidia proves one thing: AI inference lowers cost enough to expand total demand rather than compress margins.

In short: catalyst fundamentally, but tactically prone to profit-taking.