'No longer a startup playground': Former Google executive flags turning point in AI economy

Alphabet’s latest earnings may be remembered less for headline revenue growth and more for what they reveal about the future shape of artificial intelligence. According to Parminder Singh, former Managing Director at Google (Display, APAC), the numbers mark a decisive shift in how AI is being built, funded and monetised. 

In a detailed post on X (formally twitter), Singh said Google’s results underline that the AI industry is moving rapidly from experimentation to industrialisation — a phase defined by capital intensity, scale and long-term balance-sheet strength. 

At the heart of this transition is capital expenditure. Alphabet has guided for a staggering $175-185 billion in capex for 2026, a figure Singh described as the most important signal in the earnings report. “Capex guidance, not revenue growth, is what matters now,” he noted, arguing that frontier AI is increasingly becoming a game only the largest players can afford.

The implication, he said, is stark: AI is beginning to resemble earlier infrastructure revolutions such as oil, railways, electricity and telecom — industries where only companies with balance sheets “the size of nation-states” could compete at the top. 

Cloud becomes AI spend in disguise 

Another key takeaway from the results was the transformation of cloud computing itself. Google Cloud posted 48% year-on-year growth, reaching $17.7 billion in Q4, with the company explicitly attributing the surge to enterprise AI infrastructure. 

Singh argued this marks a structural change. Cloud buyers, he said, are no longer purchasing storage or virtual machines. Instead, they are buying inference, training capacity and AI-native platforms. “Cloud is no longer IT spend — it’s AI spend wearing a cloud cloak,” he wrote. 

This shift puts pressure on standalone cloud providers without deep AI stacks, which risk being reduced to low-margin utilities as AI workloads become the core driver of demand. 

From demos to economic engines 

Perhaps the most telling numbers in the report relate to usage. Google disclosed that its Gemini APIs are now processing 10 billion tokens per minute, while the Gemini app has reached 750 million monthly active users. 

For Singh, these figures show that AI models are no longer showcases or experiments — they are becoming full-fledged economic engines. Token economics, he said, are now central to revenue mechanics, pushing the industry toward efficiency, reliability and global scale rather than just model intelligence. 

“The winners won’t be those with the smartest models,” he wrote, “but those who can run them cheaply, reliably, everywhere.” 

Search grows alongside AI 

Contrary to fears that generative AI would eat into traditional search, Google’s search revenue grew 17% even as AI features were deeply integrated into the product. Singh said this punctures the popular narrative that AI is cannibalising demand. 

Instead, AI appears to be expanding the “surface area of queries,” increasing how and how often users interact with search. Habit and distribution, he argued, may prove more powerful than novelty in shaping AI adoption. 

A three-layer AI industry emerges 

Singh also outlined what he sees as a clear stratification of the AI ecosystem into three layers. At the top are “AI superpowers” such as Alphabet, which own their models, infrastructure and distribution. In the middle are “AI translators” — SaaS platforms and vertical integrators that build domain-specific intelligence but live or die by unit economics. At the bottom are AI consumers, who pay per token with limited scope for differentiation. 

While moats are forming rapidly at the top, Singh warned that the middle layer will face intense churn as economics tighten. 

AI becomes a balance-sheet business 

One of the most intriguing signals in Alphabet’s report was its decision to issue $24.8 billion in new debt despite holding substantial cash reserves. Singh said this reflects the long-dated, capital-hungry nature of AI returns, drawing parallels with telecom networks, power grids and semiconductor fabs. 

The broader conclusion, he argued, is that the AI industry may be exiting the phase where “any smart startup can win” and entering one where scale, capital and endurance decide the outcome.