From the studio

Video: Data with Donovan: Jobs data and the Fed's next move (8 mins)
Video: TRIOs for 2026: CIO’s Delwin Limas on AI, power, and longevity (7 mins)
Video:Why gold isn’t the only commodity you need, with CIO’s Wayne Gordon (6 mins)

 

Thought of the day

DeepSeek this week unveiled two new versions of AI models that the Chinese startup says match the performance of OpenAI’s GPT-5 and Google’s Gemini-3 Pro on certain metrics. The debut follows Alibaba’s latest upgrade to its AI model Qwen last month, highlighting the competitive global race toward next-generation artificial intelligence.

The announcement comes as market concerns over AI capex and earnings persist, three years after ChatGPT was launched and fundamentally changed the dynamics of the technology industry.

But while the ebb and flow of market fears over the future of AI may trigger periods of volatility, we think investors should focus on the fundamentals to assess the durability of the AI rally.

We see genuine demand for AI-related products and services. Leading tech companies have indicated that the demand for AI-related products and services supports increased capital expenditures, and we think aggregate token usage stands out as the most reliable metric to track. Tokens serve as the basic unit of compute in AI models—each time a model processes a word, instruction, or line of code, it consumes tokens. Google has offered some insights into token usage across its products: monthly usage has climbed from 480 trillion tokens in May to 980 trillion in July, and further to 1,300 trillion in October. This represents more than a 130-fold increase over the past 18 months, underscoring robust and sustained usage across AI platforms.

Big tech corporate margins remain healthy despite higher capital expenditure. As demand is expected to grow, companies must allocate greater capital expenditure toward AI infrastructure projects. The ability to maintain healthy profit margins has therefore become a key concern if these investments are not carefully managed. But while big tech companies’ capex intensity is expected to stay on an upward trend amid rising investment and an increase in depreciation expenses, we expect their margins to remain relatively resilient, at around 27%, as the bulk of other operating costs are growing at a slower pace than the anticipated revenue growth. Our analysis suggests that, even with our projection of USD 1.3tr in global AI capex by 2030, big tech companies can collectively maintain stable margins.

Monetization continues to show improvements. Companies across the AI value chain eamploy a range of business models to generate revenue, and we continue to see encouraging signs of improving monetization. For example, hyperscalers primarily convert AI capex into revenue by renting out GPU compute capacity—an approach with attractive economics. The fully depreciated cost of a GPU is about USD 1.22 per hour, while rental prices range from USD 2 to USD 7 per hour. For AI model developers, sustained revenue growth will depend on consumer and enterprise subscriptions, enterprise APIs (digital connections that let businesses plug AI features directly into their own products), as well as monetization from free users via advertising and/or e-commerce revenues. The potential is large. OpenAI CEO Sam Altman projects an annual recurring revenue of USD 100bn by 2027, compared to an estimated USD 13bn this year. For those upstream in the supply chain, we believe the strong compute demand and the durability of AI infrastructure cycle should support continued earnings growth. For example, NVIDIA’s Hopper suite of chips still generated USD 2bn in quarterly revenue 13 quarters since their inception.

So, considering the balance between medium-term risks and robust near-term fundamentals, we maintain our recommendation to diversify investments across the AI value chain, encompassing enabling technologies, intelligence, and applications. Our strategic focus will increasingly favor the applications layer, as we anticipate that companies operating within this segment will benefit most from ongoing AI-related capital expenditures.

For more details, refer to Intelligence Weekly #91: FAQs on the AI trade, published on 24 November 2025.