Three years after ChatGPT’s explosive debut, the so-called “AI revolution” in the business world is still more sizzle than steak. Despite billions poured into AI startups and endless headlines about the future of work, a jaw-dropping 95% of enterprises say they haven’t seen any real value from their AI investments, according to a recent MIT survey. The gap between AI’s hype and its actual impact in the workplace is wider than ever.

But don’t tell that to venture capitalists. In a recent survey of 24 VCs focused on enterprise tech, the message is the same as always: the AI tipping point is just around the corner. These investors are doubling down, insisting that the next few years will finally deliver the AI-powered business transformation we’ve all been promised. Sound familiar? It should,VCs have been making these predictions for years, with little to show for it so far.

So what’s the holdup? Kirby Winfield of Ascend says companies are realizing that large language models aren’t a magic bullet. “Just because Starbucks can use Claude to write their own CRM software doesn’t mean they should,” he quips. Now, the conversation is shifting to more technical topics like custom models, fine-tuning, and data sovereignty,areas that most businesses aren’t equipped to handle without serious help.

Some VCs, like Molly Alter of Northzone, are already moving the goalposts. She predicts that many AI startups will pivot from selling products to acting as consultants, building custom solutions for clients. In other words, the “AI product” dream is fading, and the new pitch is all about services and support.

Meanwhile, others are hyping the “next big thing.” Marcie Vu of Greycroft is betting on voice AI, claiming it’s the future of human-machine interaction. Alexa von Tobel of Inspired Capital is talking up AI’s potential to reshape infrastructure and manufacturing, promising a shift from reactive to predictive business models. The buzzwords just keep coming.

On the infrastructure front, Michael Stewart of M12 is focused on the nuts and bolts,data center cooling, memory, and networking. As AI workloads balloon, the need for more efficient and sustainable operations is becoming a favorite talking point for investors looking to justify their bets.

But let’s not ignore the chaos. Even VCs admit that the current glut of AI tools is creating confusion inside organizations. “Random experiments with dozens of solutions create chaos,” says Winfield. The new narrative? Enterprises will soon consolidate around a handful of “winners” that actually deliver value,if they can find them.

There’s also a new spin on what makes AI valuable. Rob Biederman of Asymmetric Capital Partners says the real advantage isn’t in the models themselves, but in how well they’re integrated into business processes and the proprietary data they use. The message: it’s not about the tech, it’s about the moat.

Vertical software is the latest darling. VCs are now touting industry-specific AI solutions as the best way to build defensible businesses. “It’s much easier today to build a moat in a vertical category rather than a horizontal one,” says Alter. Regulated industries, supply chain, and retail are the new gold rush.

But there’s a dark cloud on the horizon: energy consumption. Aaron Jacobson of NEA warns that the world is running out of power for all these GPUs. The hunt is on for more efficient chips and smarter software, but the clock is ticking.

Despite all these challenges, some VCs insist that enterprise AI is finally maturing. Scott Beechuk of Norwest Venture Partners claims the groundwork is set and the payoff is coming soon. As models improve and oversight increases, AI is supposedly becoming more reliable in daily business operations.

Still, don’t expect overnight change. Marell Evans of Exceptional Capital says there’s “still a lot of iteration” before AI can solve real business problems. Jennifer Li of Andreessen Horowitz, on the other hand, is already declaring victory, pointing to software engineers who are using AI coding tools as proof that value is here,and about to multiply.

Budgets are shifting, too. Instead of throwing more money at AI, companies are reallocating existing spend and demanding real ROI. “Budgets will increase for a narrow set of AI products that clearly deliver results, and will decline sharply for everything else,” says Biederman. Translation: the AI market is about to get a lot more cutthroat.

As companies get smarter about AI, they’re expected to cut back on experimentation and focus on proven winners. Andrew Ferguson of Databricks Ventures predicts CIOs will push back on “AI vendor sprawl,” consolidating around the few solutions that actually work.

For AI startups, the bar is higher than ever. Investors want both a compelling story and real traction. “Revenue without narrative is a feature; narrative without traction is vaporware. You need both,” says Jake Flomenberg of Wing Venture Capital. Only the most deeply embedded, mission-critical products will survive.

Retention is the new obsession. The highest retention rates go to companies whose software is so deeply embedded it’s nearly impossible to rip out. Jonathan Lehr of Work-Bench points to authorization and policy tools as prime examples. The more foundational the software, the stickier it is.

Looking ahead, the next big promise is AI agents,autonomous systems that work alongside humans. Some VCs predict that within a few years, most knowledge workers will have an AI “co-worker.” The winners, says Antonia Dean of Black Operator Ventures, will be those who balance autonomy with oversight. The rest? Just more noise.

Ultimately, the fastest-growing companies are those that spot new workflow or security gaps created by generative AI and move quickly to fill them. In cybersecurity, it’s abo