NVIDIA’s latest quarterly results were impressive by any measure. Revenue surged again, data-centre sales hit record levels, and forward guidance pushed expectations even higher. The market reacted with predictable enthusiasm: if the biggest supplier of AI infrastructure is still experiencing rapid demand, many assume the broader AI boom remains intact.
But the reality is more complicated. NVIDIA’s strength confirms that AI investment is real, not speculative fantasy. Yet it also exposes a widening divide inside the AI ecosystem, one that matters far more to investors than the headline numbers.
The simple version: AI is generating revenue, but only in certain parts of the value chain. Infrastructure suppliers and cloud platforms are making money today. Foundation model labs are generating revenue but still wrestling with heavy costs. Downstream application companies, start-ups, and AI-themed stocks are mostly selling narratives rather than financial results.
Where the Money Is Actually Being Made
Infrastructure remains the most reliable part of the AI trade. NVIDIA, AMD, Broadcom and TMSC sit in the slipstream of an enormous capex cycle. Hyper-scalers are spending to build out clusters for model training, inference and enterprise employments. These companies sell hardware and networking gear that customers need today, not someday.
Cloud providers (Microsoft, Google, AWS) are also benefitting. AI workloads are lifting cloud growth, and the economics of selling compute scale better than most assume. Microsoft’s integration of OpenAI into Azure has created an early lead, but Google is catching up as Gemini spreads across its product suit.
This part of the market is already monetising AI in a meaningful way. For traditional equity investors, these remain the most credible and predictable ways to gain exposure.
The Middle Layer: Foundation Models With Real Revenue but Heavy Costs
OpenAI, Anthropic, and Google DeepMind sit in a different position. They generate revenue through enterprise subscriptions, API usage and model integrations. Demand is strong and improving. But profitability is less clear.
Training large models remains extremely expensive. Running them at scale is also costly. Model pricing is under competitive pressure, and the industry hasn’t yet reached a stable equilibrium between capability, cost and willingness to pay.
Still, these companies are the most likely long-term winners outside of infrastructure. They have access to compute, proprietary data and distribution channels. Scale matters, and few smaller labs can match it.
Where the Bubble Risk Is Rising
The long tail of the AI market is where valuation risk has quietly intensified.
Dozens of Ai start-ups with billion-dollar valuations have little revenue and no clear path to monetisation. Many public small-cap “AI companies” have seen significant share price appreciation despite negligible AI-driven income. The pattern resembles past cycles: the companies closet to the picks-and-shovels make the money, and the ones furthest away tend to over-promise and under-deliver.
This doesn’t mean innovation will stall. It means investors need to be selective. The biggest mistake today is assuming the entire industry will monetise at the same pace.
Positioning: What Matters Most for Investors
From a portfolio perspective, the takeaway is straightforward.
- Prioritise infrastructure and cloud: These businesses capture revenue from the primary wave of AI adoption. Their earnings reflect real usage, not projected future value.
- Treat foundation model companies as strategic long-term plays: They may not be profitable yet, but they own the scarce assets: compute, data and distribution.
- Approach downstream applications and small-cap AI names with caution: Most lack pricing power, face high competition and rely heavily on marketing narratives.
- Watch for the shift from model training to enterprise deployment: Over the next two years, the key question becomes whether businesses actually integrate AI into workflows at scale. Hardware spending alone can’t sustain valuations forever.
Read More
Outlook: The Boom Is Real, but Uneven
The AI sector is not moving as a single unit. Some parts are already delivering strong, repeatable earnings. Others are still experimenting with monetisation. That gap will define investor returns over the next decade.
The next phase of the AI cycle hinges on enterprise adoption. If businesses convert pilots into full operational deployment, the second wave will lift cloud providers, model labs, cybersecurity firms, and selective enterprise software companies.
If adoption stalls, or if model costs fall faster than revenue grows, valuations for many smaller firms will come under pressure.
The broader AI story is still compelling. But investors should avoid treating “AI” as a single theme. The winners are emerging through earnings, not narratives. NVIDIA’s success confirms the boom is real; it doesn’t confirm that everyone will share equally in its reward.