Silicon Valley spent much of the past year urging workers to use artificial intelligence as aggressively as possible. But as companies rack up usage-based bills and wrestle with internal adoption, the conversation is shifting from experimentation to accountability. The biggest question now is no longer whether employees can use AI everywhere, but whether the technology is paying for itself.
That tension is exactly where NEA partner Tiffany Luck says many enterprises find themselves today. In a recent conversation on TechCrunch’s Equity podcast, Luck argued that the market has moved beyond novelty and into a more sober phase, where organizations are measuring what AI actually delivers rather than simply celebrating how much of it they can deploy.
The shift is already visible. Reports of companies overshooting annual AI budgets, trimming access to premium model subscriptions, and quietly scaling back internal AI programs have become a sign of the times. The exuberance of “use it for everything” is colliding with finance teams, procurement departments, and executive pressure to prove return on investment.
Luck, whose investment career began during the era when venture capital was still trying to persuade businesses that e-commerce would fundamentally reshape commerce, believes AI is following a similar path. But she sees one important difference: this wave is not just about operational efficiency. In consumer products especially, AI is creating what she described as memorable “magic moments” that can change how people interact with software altogether.
Her comments come as the industry faces a broader inflection point. AI startups are pitching enterprises on better monitoring, better governance, and better ways to connect spending with business outcomes. Meanwhile, investors are trying to separate durable revenue from inflated usage figures, and corporate buyers are asking tougher questions about where AI belongs inside the workflow.
At the center of that debate is a simple but increasingly urgent problem: AI is easy to adopt, but hard to value.
From token-maxxing to budget discipline
The phrase “token-maxxing” became shorthand in Silicon Valley for companies pushing employees to generate as many model calls as possible. The logic was straightforward: if workers can be made more productive with AI, then more usage should mean more value.
In practice, the economics are proving more complicated. Large-model access can be expensive at scale, and the cost grows quickly when thousands of employees or customer-facing applications are involved. Companies that initially treated model spending as a flexible innovation budget are now discovering that broad access can become a serious line item.
That has led to a more selective approach. Some businesses have reined in access to premium tools. Others are limiting where models can be used, or requiring teams to justify spend against measurable business gains. The result is a new phase of AI adoption that looks less like a free-for-all and more like classic enterprise software procurement.
This is not necessarily a sign that companies are turning away from AI. Instead, it suggests the market is maturing. Early adopters often experiment first and justify later. But once usage scales, executives need evidence that the tools are increasing revenue, reducing costs, or improving retention enough to validate the expense.
Luck’s perspective reflects that reality. For investors, the challenge is not merely spotting enthusiasm; it is identifying the companies that can convert that enthusiasm into lasting economics.
Why NEA is watching the ROI layer closely
NEA has long been active in software and consumer technology, and Luck’s portfolio lens appears shaped by both enterprise pragmatism and consumer ambition. She has seen technology shifts unfold before: first as promise, then as infrastructure, and finally as an expectation baked into everyday products.
With AI, the same pattern is emerging in compressed form. Companies are no longer asking whether they should use AI at all. They are asking which workflows benefit most, which teams need oversight, and which tools actually justify the licensing and infrastructure costs attached to them.
That gives a new category of vendors room to grow. Startups are building products that help organizations monitor AI use, allocate spend, audit adoption, and connect deployments to performance outcomes. These firms are not selling frontier models; they are selling visibility, control, and financial clarity.
In other words, as model access becomes more commoditized, the economic value may shift upward into the layer that measures, manages, and optimizes the spend. For enterprise buyers, that kind of tooling can be the difference between a promising pilot and a scalable program.
For investors, it may also be a more reliable source of revenue than raw model usage alone. Businesses are often willing to pay for control systems when they are unwilling to keep funding open-ended experimentation without proof.
The consumer angle: AI as a source of “magic moments”
Luck’s enthusiasm for AI is not limited to back-office workflows. She is particularly interested in consumer applications, where the technology can create moments that feel less like automation and more like genuine delight.
That matters because consumer software has historically rewarded products that change behavior, not just streamline tasks. The best consumer experiences often feel intuitive, personalized, and slightly surprising. AI, when it works well, can produce that effect by making software more adaptive and responsive to a user’s intent.
These “magic moments” may include tools that anticipate needs, generate high-quality output with minimal prompting, or reduce friction in ways users immediately feel. That can make AI more than a productivity feature. It can become a reason to choose one product over another.
Still, consumer AI is also full of risks. Delight can quickly turn to frustration when outputs are unreliable, overly generic, or intrusive. A good consumer AI product needs not only strong technical performance but also strong product design, clear expectations, and careful guardrails.
That is one reason investors are paying attention to startups that can pair model capabilities with thoughtful user experience. The winning consumer companies may not be the ones with the most powerful models, but the ones that make AI feel naturally embedded in a product people already want to use.
What enterprises are learning about AI value
Companies that rushed to deploy AI across departments are now learning that adoption does not automatically equal impact. A large percentage of internal usage can be difficult to translate into concrete financial results, especially when the benefits are spread across many employees rather than concentrated in one measurable process.
That makes AI ROI tougher to track than, say, a sales tool with a direct pipeline effect or a logistics system with clear cost savings. Many AI deployments improve speed, convenience, or employee satisfaction in ways that matter but are hard to quantify. Others may reduce small amounts of friction across dozens of workflows, creating value that is real but diffuse.
The challenge for companies is to define success before the budget disappears. Without a framework, AI can become a perpetual pilot: widely discussed, widely used, and difficult to defend during budget reviews.
Enterprise leaders are increasingly building scorecards that assess:
- time saved per task or workflow,
- reduction in support or operations workload,
- quality improvements in content or code,
- customer conversion or retention gains, and
- cost per seat or per model request.
Those metrics help executives understand where AI is moving the needle. They also reveal where AI is merely adding a layer of novelty.
Startups rush in to solve the measurement problem
Where there is confusion, startups tend to follow. The enterprise AI stack is now expanding beyond model providers and application builders into a market for observability, governance, optimization, and cost management.
These companies aim to answer questions that model vendors often leave unresolved: How much are teams using AI? Which use cases are worth keeping? Which employees are getting the most benefit? Where is the spend bloating without improving outcomes?
That category matters for several reasons.
- It helps chief financial officers control exposure to unpredictable usage-based pricing.
- It gives IT and security teams more oversight of internal deployments.
- It gives product and operations leaders a clearer view of performance.
- It provides investors with a more credible story about repeatable enterprise spend.
As AI matures, the market may increasingly favor companies that make adoption accountable. In many enterprise software cycles, the tools that survive are not the flashiest ones; they are the ones that become indispensable to how businesses manage the technology they already bought.
That dynamic could also shape deal flow. Venture investors are likely to look harder at whether a startup is creating direct workflow value, or merely riding the wave of hype surrounding model access. The distinction may determine which businesses win funding and which ones get compressed as pricing becomes more competitive.
AI IPOs: promise, pressure, and the public-market test
Luck also discussed the year’s AI initial public offerings, a reminder that public markets are beginning to test the sector in a different way. Private investors can tolerate long timelines and messy margins if the growth story is compelling. Public investors, by contrast, tend to care much more quickly about efficiency, retention, and profitability.
That means AI companies going public must do more than demonstrate excitement. They need to show that their revenue is durable, that demand is not purely cyclical, and that the business can withstand scrutiny around margins and operating costs.
The public-market phase often exposes a gap between narrative and operating reality. A company can have strong adoption and still face hard questions about customer concentration, inference costs, and future pricing pressure. It can also win headlines while losing confidence if investors believe the economics are too dependent on continued hype.
For the AI sector, IPOs may serve as a discipline mechanism. They force companies to explain where their margins come from, how they plan to scale, and why customers will keep paying once the novelty fades.
That kind of scrutiny mirrors the enterprise conversation. Whether in public markets or inside the corporate budget process, AI vendors are being asked to justify themselves with numbers, not slogans.
Why the market is shifting from enthusiasm to evidence
Every major technology cycle begins with experimentation, but lasting winners are usually defined by evidence. The same is happening with artificial intelligence.
In the early phase, organizations were eager to show they were participating in the AI boom. They tested tools, opened up access, and encouraged staff to integrate models into everyday work. That urgency was fueled by fear of missing out as much as by genuine productivity gains.
Now the decision-making environment is more demanding. AI is no longer a side project or a press-release feature. It is becoming part of procurement, budgeting, compliance, and strategy. That transition forces companies to ask difficult questions about value creation and organizational fit.
For some businesses, the answer will be to keep spending. For others, it will be to narrow use cases, switch vendors, or build internal systems that lower cost. Either way, the indiscriminate phase is ending.
The next stage is likely to reward discipline. The companies that can prove ROI, whether through direct savings or new revenue, will have a much easier time defending their AI roadmaps. Those that cannot may find themselves cutting back even as the broader market continues to grow.
What Tiffany Luck’s view signals about the AI cycle
Luck’s comments offer a useful snapshot of where the AI industry stands today. Her outlook is not anti-AI; if anything, it is a sign of deep belief in the technology’s potential. But it is also grounded in a recognition that transformative technology still has to earn its keep.
That makes her perspective especially relevant to both founders and buyers. Founders need to understand that selling AI is no longer enough. They need to sell outcomes. Buyers need to understand that broad usage alone is not a strategy unless it leads to meaningful business results.
The conversation also underscores a broader truth about enterprise adoption: technology waves do not end when the excitement fades. They end when the business models become clear. AI is entering that phase now.
The companies that succeed will likely be the ones that help customers answer three questions:
- What problem does AI solve better than existing tools?
- How do we measure whether it worked?
- What does success cost us, and is that cost worth paying?
Those are not glamorous questions, but they are the ones that determine which technologies become standard and which remain experiments.
Timeline: how the AI ROI conversation evolved
| Phase | What companies were doing | What changed |
|---|---|---|
| Early enthusiasm | Rolling out tools widely and encouraging heavy usage | AI was seen as a competitive necessity |
| Usage surge | Employees experimented freely with premium models and copilots | Spending increased faster than many budgets expected |
| Budget backlash | Companies began restricting licenses and revisiting policies | Finance teams demanded evidence of value |
| Measurement phase | Businesses started tracking savings, productivity, and outcomes | ROI became the central question |
The road ahead for enterprise AI
AI is not losing momentum. If anything, it is becoming more embedded in how companies work. But the rules are changing. The market is moving from a growth-at-any-cost mindset toward a more demanding environment in which utility, efficiency, and proof matter more than hype.
That may ultimately be healthy for the sector. Technologies often become more durable once users learn to separate usefulness from excitement. Companies that can demonstrate repeatable value tend to outlast those that depend on speculative adoption.
For enterprise buyers, the message is clear: AI should be treated like any other serious investment, with expectations, benchmarks, and accountability. For investors, the most attractive businesses may be those that help organizations control, measure, and improve the way AI is used.
And for consumer products, Luck’s “magic moments” may be the clearest reminder that AI still has room to surprise people in ways that matter. The challenge is turning that surprise into something customers will pay for, keep using, and recommend.
That is the real AI test now. Not how much companies can spend, but how much value they can prove.
Key takeaways
- Enterprises are moving from AI experimentation to tighter budget scrutiny.
- Usage growth has exposed the cost of broad model deployment.
- Startups are emerging to help companies track AI spend and ROI.
- Consumer AI remains attractive when it creates intuitive, memorable experiences.
- The next stage of the AI market will likely reward measurable outcomes over hype.
Luck’s core message, paraphrased from her Equity appearance, is that enterprises are no longer asking whether AI is exciting; they are asking whether it is worth the money.
