It has been three years since OpenAI released ChatGPT, igniting one of the fastest technology adoption cycles in modern history. Since then, artificial intelligence has dominated boardrooms, earnings calls, and venture capital pitch decks. Optimists have long argued that AI would become foundational to enterprise software, and startups rushed in accordingly, buoyed by tens of billions of dollars in investment.
Yet for all the experimentation, most enterprises are still struggling to show results.
An MIT survey published in August found that 95% of enterprises reported no meaningful return on their AI investments. Pilot programs abound, but few scale. AI is everywhere , except on the balance sheet.
So when will enterprise AI actually pay off?
According to 24 enterprise-focused venture capitalists surveyed by TechCrunch, the answer is familiar but firm: 2026.
The problem is that VCs have been pointing to “next year” for three consecutive years. The question now is whether 2026 marks a real inflection point, or just another postponement.
From Silver Bullets to Systems Thinking
One clear shift is already underway: enterprises are abandoning the idea that large language models alone can solve complex business problems.
“LLMs are not a silver bullet,” said Kirby Winfield, founding general partner at Ascend. “Just because Starbucks can use Claude to write its own CRM software doesn’t mean it should.”
Instead, investors expect companies to focus on the unglamorous but necessary layers of AI adoption: custom models, fine-tuning, evaluation, observability, orchestration, and data sovereignty. In other words, infrastructure and governance, not demos.
That reality is pushing many AI startups to rethink their business models altogether.
Molly Alter, a partner at Northzone, predicts that many AI product companies will morph into AI consultancies in disguise. After deploying a narrow solution like customer support or coding agents, these firms increasingly find themselves embedding engineers with customers, building bespoke workflows across departments.
“Specialized AI product companies will become generalist AI implementers,” Alter said.
Voice, Physical Systems, and the Real World
Beyond enterprise software, VCs see AI expanding into more tangible domains.
Marcie Vu, a partner at Greycroft, is bullish on voice AI, arguing that speech is a more natural interface than screens or keyboards. Meanwhile, Alexa von Tobel, founder of Inspired Capital, expects 2026 to be the year AI reshapes the physical world , from manufacturing and infrastructure to climate monitoring.
“We are moving from a reactive world to a predictive one,” she said, where systems identify problems before failures occur.
Others are watching how frontier AI labs behave.
Lonne Jaffe, managing director at Insight Partners, notes that model developers are increasingly shipping turnkey applications directly into regulated industries like finance, healthcare, law, and education, rather than leaving implementation to startups.
Where Investors Are Placing Their Bets
Capital, unsurprisingly, is flowing toward bottlenecks.
Salesforce Ventures’ Emily Zhao is focused on AI in physical systems and next-generation model research. M12’s Michael Stewart is investing in the future of data centers, cooling, networking, memory, and energy efficiency, as AI workloads strain global infrastructure.
Energy constraints loom large.
“We’re at the limit of humanity’s ability to generate enough energy to feed GPUs,” said Aaron Jacobson of NEA. Performance-per-watt, not raw compute, is becoming the new competitive frontier.
At the application layer, investors favor vertical enterprise software where proprietary workflows and regulated data create defensibility, according to Jonathan Lehr of Work-Bench.
The New Definition of an AI Moat
Few investors believe model performance alone constitutes a moat.
“If OpenAI launches a model tomorrow that’s 10x better, does this company still need to exist?” asked Jake Flomenberg of Wing Venture Capital.
The strongest defenses, VCs argue, come from deep workflow integration, proprietary or continuously improving data, and high switching costs.
“AI moats are about economics and integration, not prompts,” said Rob Biederman of Asymmetric Capital Partners.
Data moats and workflow moats, particularly in vertical industries like manufacturing, healthcare, and legal services, are increasingly favored over horizontal AI platforms.
Will 2026 Finally Deliver ROI?
Opinions diverge, but most agree the enterprise AI experiment is entering a more disciplined phase.
“Random experiments with dozens of tools create chaos,” said Ascend’s Winfield. “Enterprises will focus on fewer solutions with more thoughtful engagement.”
Others are more skeptical.
Antonia Dean of Black Operator Ventures warned that AI risks becoming a scapegoat, used to justify layoffs or cost cuts rather than genuine productivity gains.
Still, there are signs of progress. Scott Beechuk of Norwest Venture Partners sees 2026 as the year the application layer finally proves whether AI infrastructure investments were worth it.
And some argue value is already visible, just unevenly distributed.
“Ask any software engineer if they want to give up AI coding tools,” said Jennifer Li of Andreessen Horowitz. “Enterprises are already gaining value. It will multiply.”
Budgets Will Rise But Concentrate
Most investors expect AI spending to increase in 2026, but not evenly.
Budgets will concentrate around a small number of vendors that demonstrably deliver results, while weaker tools are cut. CIOs, according to Databricks Ventures’ Andrew Ferguson, will aggressively push back against vendor sprawl.
The era of endless pilots is ending. AI spending is shifting from experimental line items to core operational budgets.
What It Takes to Raise a Series A in 2026
For AI startups, the bar is rising.
Revenue alone isn’t enough. Neither is vision.
“The best companies combine a compelling ‘why now’ with proof of enterprise adoption,” said Wing’s Flomenberg. Investors want evidence that customers view the product as mission-critical, not optional.
Twelve-month contracts, referenceable customers, and real usage in production environments matter more than flashy demos.
AI Agents: Promise, Not Panacea
Despite hype, most VCs expect AI agents to remain in early adoption by the end of 2026.
Compliance, oversight, and interoperability standards remain unresolved. Some foresee consolidation into universal agents with shared memory; others emphasize human-agent collaboration rather than replacement.
Still, optimism abounds.
“The majority of knowledge workers will have at least one agentic co-worker they know by name,” predicted NEA’s Jacobson.
The Bottom Line
Enterprise AI is not failing, but it is sobering.
The industry is moving away from spectacle toward systems, from pilots toward production, and from horizontal tools toward deeply embedded infrastructure. Whether 2026 is the long-promised breakthrough or merely another waypoint depends less on model advances than on organizational discipline.
After three years of hype, enterprise AI’s future may finally be getting realistic.
And that, investors argue, is what makes 2026 different, at least this time.
