Page 28 - gs260301

Insights and Expertise

The AI wave hasn't broken, but

undercurrents are shifting

supplies the computing power in a tightly linked ecosys-
tem that includes OpenAI and Oracle, while many smaller
startups are already struggling with product maturity
and commercial traction.

The structural warning signs are real; this concentration
increases systemic risk (see https://www.ineteconomics.org/
perspectives/blog/the-ai-bubble-and-the-u-s-economy-how-
long-do-hallucinations-last?utm_source=chatgpt.com). When
valuations run far ahead of fundamentals in a narrow part
of the stack, any slowdown can ripple outward quickly.

That does not mean the bubble has burst. It suggests the
system is fragile.

The long view
One of the clearest contrasts in global AI development lies
in strategic intent. Much of the Western focus has centered
on individual productivity, convenience and competitive
By Scott Dawson edge—think faster content creation, smarter assistants,
DECTA incremental efficiency gains. These applications are not
without merit, but they tend to prioritize near-term re-
t is increasingly evident that the AI boom has not yet turns.
burst. That said, it would be complacent to assume
the path ahead will be smooth. A dramatic collapse
I may not be inevitable, but a pause, slowdown or
uncomfortable recalibration still feels plausible. Markets
rarely move in straight lines, and AI will be no exception. AI in payments: Where it already works
The key question is not when a correction might arrive,
but what kind of value will endure if it does. While generative AI dominates headlines,
payments and fintech have relied on artificial
The dot-com era remains a useful reference point, not be- intelligence for years in quiet, highly effective
cause today's conditions are identical, but because it illus- ways. AI models help detect suspicious activity
trates how periods of rapid technological optimism tend and improve risk scoring across payment
to resolve. Companies like Pets.com gambled millions on networks. These systems continuously analyze
pet-food ecommerce before consumers had fully changed vast volumes of transaction data, learning
habits – and disappeared quickly (see https://www.red- patterns that help reduce fraud losses while
dit.com/r/ValueInvesting/comments/1o4qgrg/2000_bubble_ minimizing false positives that frustrate
vs_2025_bubble/?utm_source=chatgpt.com). legitimate customers.

Those solving real problems, even imperfectly at first, en- Unlike many experimental AI applications,
dured and eventually defined the next economic cycle. these tools are embedded directly in payments
Amazon is a standout success story coming from humble infrastructure and evaluated through measurable
beginnings, surviving and building a trillion-dollar giant outcomes such as lower fraud rates and
that it is today. The same Darwinian process is at work in improved authorization accuracy. Their success
the AI world now. highlights an important lesson: the most durable
AI applications often solve practical operational
There is, however, a notable structural difference. The AI problems rather than chasing headline-grabbing
landscape is far more concentrated than the early internet innovation.
economy ever was. A small group of dominant players —
such as OpenAI and Nvidia — sit at the centre. Nvidia

28