Jensen Huang Just Gave AI Investors Great News

Nvidia CEO Jensen Huang said on the earnings call that "tokens are now profitable," suggesting AI workloads are becoming more efficient and monetizable. Nvidia CFO Colette Kress added that customers are generating profitable revenue beyond the depreciable life of their GPUs, a bullish signal for long-term chip demand. The article remains cautious on valuation, noting Nvidia trades at about 25x estimated future profits and that many AI stocks still look expensive.

Analysis

The real signal here is not “AI is working” but that utilization is crossing a threshold where inferential workloads can justify premium silicon on a cash-flow basis. That matters because it reduces the probability of a capex air pocket: if customers can now monetize tokens faster than GPU depreciation, hyperscalers and model developers are less likely to slash spend after one weak quarter. The second-order effect is a widening moat for the top-end accelerator ecosystem, since once teams optimize around the best chips, switching costs rise and the installed base becomes harder to dislodge.

The beneficiaries extend beyond NVDA. Higher ROI on AI workloads should pull through networking, HBM memory, advanced packaging, and datacenter power infrastructure, while pressuring lower-tier GPU vendors and “AI wrapper” software names that still need to prove durable unit economics. The biggest overlooked winner is likely Meta: if model inference becomes cheaper and more profitable, ad targeting, recommendation engines, and content generation can expand without commensurate opex growth, making incremental margins more sensitive to AI throughput than to headline model quality.

Consensus is probably underestimating how much of this is already priced into “AI winner” baskets. A step-up in profitability supports the long-duration AI capex narrative, but it also narrows the gap between hype and fundamentals, which can compress multiple expansion for names that have run ahead of earnings delivery. The risk is not near-term demand collapse; it is a gradual re-rating where better fundamentals merely justify existing valuations rather than drive another leg higher, especially if rate expectations or enterprise IT budgets soften over the next 1-2 quarters.

Key Decisions for Investors

• Long NVDA vs short a basket of weaker AI monetization names over 3-6 months: best risk/reward is owning the cash-flow winner while fading companies whose AI spend is still pre-ROI; use a 1:1 notional pair and trim if NVDA forward multiple re-rates materially above current levels.
• Add META on any post-earnings weakness for a 6-12 month hold: AI cost leverage should show up faster here than in most software names, and the asymmetry favors a rerating if ad efficiency improves even modestly.
• Buy semis infrastructure beneficiaries on pullbacks — especially memory and networking exposure — as a basket trade for the next 1-2 quarters; the setup is better than chasing headline AI software because these names monetize utilization, not narrative.
• Avoid adding to PLTR at current momentum unless you have a 2+ quarter catalyst window: the stock is more exposed to multiple compression if the market decides “profitable tokens” mainly validates the infrastructure layer rather than the application layer.
• For tactical upside, consider NVDA call spreads 2-4 months out rather than outright stock: the thesis is supportive but already broadly recognized, so capped-risk convexity is preferable to paying full delta into an elevated valuation.