AI Needs Customers More Than Chips - PYMNTS Analysis

Whoever controlled the compute, capital and foundational models would control the future. That story was always missing the harder half: demand. On Wednesday (May 13), Anthropic announced a small- to medium-sized business (SMB)-focused AI plugin for tools including PayPal, Intuit, Canva, Docusign and more, a move that signals AI's next stage may depend less on how much capability technology companies can supply and more on whether ordinary organizations can generate sustained demand inside everyday work.

The question facing the market is no longer simply whether frontier models can perform astonishing tasks. It is whether accountants, nurses, insurance adjusters, teachers, procurement managers, financial analysts and their peers across Main Street can integrate AI into the highly specialized workflows they understand better than any Silicon Valley engineer.

According to the PYMNTS analysis, the history of enterprise technology suggests that infrastructure breakthroughs alone rarely determine long-term economic impact. Productivity gains materialize typically when workers redesign workflows around new tools. Electricity, for example, did not transform manufacturing merely because generators became available; factories had to reorganize production around electrified systems. Cloud computing did not become indispensable because servers improved; businesses rewrote operational processes to take advantage of scalable software architectures.

In the case of AI, while organizations that have both the capital and technical resources to experiment aggressively are doing so, the broader global economy runs through the mid market. Regional banks, healthcare systems, manufacturing firms, logistics operators, school districts, accounting practices and professional services companies employ millions of knowledge workers. This segment represents a critical proving ground for AI because its constraints are different.

Mid market organizations rarely have the luxury of rebuilding systems from scratch. They operate on thinner margins, rely on legacy software and prioritize operational continuity over technological experimentation. For these businesses, AI succeeds only if it becomes practical. A nurse navigating patient intake procedures does not need a frontier model capable of generating poetry or software code. An accountant reviewing compliance documents may care less about raw model size than whether AI can reduce reconciliation time without introducing hallucinated data. Teachers evaluating student writing need systems that fit within curriculum structures and administrative realities.

PYMNTS covered how Anthropic is also launching a Claude SMB Tour that will take Claude for Small Business on the road and offer live, half-day AI fluency training for 100 local small business leaders at each stop. The Tour kicks off Thursday (May 14) in Chicago and will then travel to nine more cities this spring and additional cities in the fall. Findings in the "May 2026 Small Business Week" report by PYMNTS Intelligence reveal that digitally fluent small businesses are growing faster, adapting more easily and showing greater confidence about future expansion.

One of the defining assumptions of the early generative AI boom was that superior models would naturally create mass adoption. Better intelligence would automatically translate into widespread productivity gains. Reality has proven to be more complicated (a problem that has plagued users for years, frankly).

PYMNTS Intelligence found that advanced forms of AI, including large language models (LLMs) and agentic AI, are deeply embedded in only one enterprise area: data and technology. That gap underscores an important dynamic: AI's economic value may depend less on technological novelty than on institutional learning. Many organizations discovered that deploying AI tools is relatively easy while changing employee behavior is extraordinarily difficult.

But workers themselves may become the architects of the next AI phase because they possess the tacit knowledge necessary to identify where automation creates value and where it creates risk. A procurement officer understands the hidden friction points in vendor negotiations, while a nurse understands the practical inefficiencies of patient documentation and a financial analyst understands the contextual nuances behind reporting discrepancies.

The AI industry may therefore be approaching an important transition from capability maximization to utility optimization. This is not about whether the models can do more—it's about whether they can do what matters inside the actual physical workflow. Think about the clicks, the load times, the friction of switching between tabs, the way a nurse's fingers hover over a keyboard while simultaneously monitoring a patient's vitals. That's where the real work happens.

Meanwhile, the infrastructure spending continues unabated. Meta plans to spend between $115 billion and $135 billion this year as it races to construct data centers, chips and other AI infrastructure. The company recently announced an agreement with Amazon Web Services to bring tens of millions of Graviton cores into its compute portfolio. Amazon CEO Andy Jassy said in a 2025 Letter to Shareholders that Amazon's chips business will be much larger than most people think, with an annual revenue run rate over $20 billion and growing triple-digit percentages year over year.

Investors are betting big on AI as core hardware infrastructure. Ricursive Intelligence raised $300 million in a Series A at a $4 billion valuation to build AI systems that assist in the design of semiconductors themselves. Waabi announced $1 billion in new funding commitments to scale autonomous trucking and robotaxi deployments. Decagon announced a Series D round valuing the company at $4.5 billion for its AI concierge platform capable of handling complex customer interactions end-to-end.

The contrast is stark. While billions flow into chip design, data center construction and model training, the actual economic value will be determined by whether a regional bank's loan officer can close a file faster, whether a manufacturing plant's quality control technician can spot defects more reliably, or whether a small business owner can manage their books without hiring additional staff.

Whether users actually pay for it remains the real question. The infrastructure is being built regardless of whether anyone knows what to do with it once it's finished.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn