NOTUS reported on July 6, 2026 that a draft Treasury Department report warns about risks in the artificial intelligence market, including comparisons to parts of the dotcom bubble. The article says Treasury analysts are looking at what could happen if AI growth assumptions, productivity gains, financial conditions, or infrastructure bottlenecks do not line up with investor and business expectations.
That does not mean every AI project is reckless. It does mean AI spending deserves the same discipline owners already expect for major software, cloud, security, and infrastructure decisions. When a technology category becomes this hot, the budget conversation can move faster than the evidence.
The business risk is not only the market
For a New Jersey business owner, the immediate issue is not whether public AI stocks are fairly priced. The practical issue is whether AI-related spending inside the business is being approved with enough proof.
AI can show up as a separate tool, a feature inside existing software, a new vendor proposal, a cloud upgrade, a data project, a chatbot, an automation platform, or a managed service add-on. Each one may be small enough to approve quickly. Together, they can become a sizable operating commitment before anyone has documented what changed, what improved, and what still requires human review.
The safer conversation is not anti-AI. It is pro-evidence. Owners should be able to see what the tool is expected to improve, what data it touches, who owns the workflow, what the vendor is promising, and how the business will measure whether the spend is working.
Where optimism can turn into exposure
The NOTUS report is useful because it moves the AI conversation from hype to resilience. If a vendor prices a product around aggressive AI adoption, if a business builds a workflow around an AI feature that is still changing, or if a department signs up for tools without clear ownership, the risk eventually becomes operational.
Common weak spots include unclear ROI, duplicate subscriptions, sensitive data going into tools without approval, vendors changing features or prices, staff relying on AI output without review, and no fallback process if the service is unavailable or the result is wrong. None of those problems require a market crash. They happen when normal technology governance does not keep up with excitement.
There is also a vendor-accountability angle. If a provider recommends AI-related work, the owner should not have to accept a vague answer such as productivity, innovation, or modernization. Those words are useful as goals. They are not evidence by themselves.
Questions to ask before approving AI spend
- What specific workflow changes? Ask which task, department, or customer process will improve, and how that improvement will be measured.
- What data does the tool access? Confirm whether customer records, financial information, employee data, email, files, or regulated information are involved.
- Who reviews the output? Decide where human approval is required, especially for money movement, customer communication, legal language, medical information, HR decisions, or security changes.
- What is the exit plan? Ask how data can be exported, what happens if pricing changes, and whether the workflow can continue if the vendor changes direction.
- What proof will be reviewed later? Put a review date on the calendar and define what would count as success, adjustment, or cancellation.
A practical next step
Before adding another AI line item, owners can ask their IT provider, software vendor, or internal lead for a one-page AI spend review. It should list the tools in use, the business purpose, data exposure, monthly or annual cost, owner, measurable benefit, and fallback process.
That review does not have to slow useful work. It keeps the decision visible. If the tool is valuable, the evidence will help it survive budget scrutiny. If it is riding on excitement alone, better to find that out before renewal season.
AI may still become a durable advantage for many small and mid-sized businesses. The reality check is simpler: optimism is not a control, and a forecast is not a business case.
Sources and further reading
