A quick look at an IPO S-1 filing shows how artificial intelligence euphoria has spun out of control. Jersey Mike’s, a U.S. sandwich chain with over 2,000 locations, slipped AI among its growth drivers in the paperwork submitted to the SEC. Not a model provider, not a cloud company: a business that slices cold cuts and bakes bread.

The anecdote is telling not because it suggests Jersey Mike’s has secret fine-tuning projects, but because it reveals how the phrase “artificial intelligence” has become a must-have ticket for any public listing. A perceived obligation, almost a stylistic clause, reminiscent of the era when slapping “.com” onto a name earned absurd valuations.

When even sandwiches have to wink at LLMs, the risk of misunderstanding for those making real infrastructure decisions skyrockets. AI washing – cloaking a business in a veneer of intelligence to appear trendier – is not harmless. It feeds misguided expectations in boardrooms, pushes budgets without a clear grasp of what is actually needed, and can lead to rushed architectural choices.

For organizations evaluating on-prem stacks, the phenomenon has a concrete cost. The pressure to “do something with AI” can trap them into off-the-shelf cloud solutions that are expensive and lack the control a data-sovereignty-conscious company would want. Or it prompts hardware purchases – GPUs with ample VRAM, inference-optimized servers – without mapping workloads, total cost of ownership (TCO), and actual latency and throughput requirements.

The sandwich-IPO mindset, in short, arrives in data centers: the label is wanted, not the substance. Yet the choice between a self-hosted deployment and a cloud service cannot rely on a knee-jerk reflex. It must be grounded in data: how many tokens per second are needed, what quantization level is acceptable, which regulatory framework (GDPR, data residency) must be respected.

Jersey Mike’s lesson is not that AI is useless for retail. It is that the hype demands an extra dose of clarity from technical decision-makers. Separating substance from what merely impresses investors is an exercise that, downstream, determines a project’s success or failure. For teams sizing up on-prem infrastructure, now is the time to confirm that VRAM and service-level agreements are not filled in with buzzwords.