AI in the enterprise: what Meryem Arik predicts for 2030

Meryem Arik, cofounder of Doubleword, laid out a framework for how artificial intelligence is likely to transform enterprise computing by 2030—a prediction she feels qualified to make, having correctly identified the inference bottleneck three years ago when NVIDIA's stock went on to climb tenfold.

Speaking at QConAI, Arik emphasized that her predictions target early adopters and the early majority in enterprise organizations, not laggard segments. She structured her talk around several interconnected trends, each resting on a coherent set of assumptions. If you disagree with her conclusions, she suggested examining which assumptions you question rather than the logic itself.

The Token Spend Crisis Will Define Enterprise AI Operations#

One of Arik's central predictions concerns how organizations will manage the exploding cost of AI inference. Although token prices for a fixed level of intelligence have fallen roughly tenfold annually over the past three years—driven by open-source model competition—overall spending continues to skyrocket, a classic case of Jevons paradox. Every new generation of AI use cases, from simple chatbots to reasoning models to agents, consumes exponentially more tokens than the last.

Arik noted that the problem has become acute only recently. An unnamed company burned through $500 million in Claude credits in a single month, Uber exhausted its entire token budget by late April, and J.P. Morgan is spending $2 billion more on IT this year than the previous one. She debunked the myth that AI labs are subsidizing tokens to lock in market share—in fact, their inference business is highly profitable, and frontier models will only grow more expensive as labs build larger, more capable systems.

By 2030, she predicts organizations will treat token spend as a critical pain point comparable to headcount budgets, requiring frameworks for tracking, attribution, and optimization across teams.

Everyone Will Become a Code Builder#

Arik's second prediction addresses a fundamental shift in who writes software. Today, enterprises funnel hundreds of thousands of employees with diverse needs into a handful of centralized applications. This model made sense when software was expensive to build and maintain. That assumption no longer holds.

With tools like Lovable and Claude now enabling non-technical users to write functioning applications in minutes, the barrier to entry has collapsed. In her own organization, a business development representative fresh out of college built a fully functional lead-tracking app connecting multiple systems in about 25 minutes—something that would once have required weeks of engineering work. When software is essentially free and anyone can build it, Arik expects a "Cambrian explosion" of bespoke internal applications within organizations.