Three years after ChatGPT sparked an AI boom, the enterprise software world is still trying to turn excitement into measurable outcomes. AI tooling has proliferated across customer support, coding, go-to-market, security, and data operations—but many large organizations remain stuck in pilot mode, unsure which deployments will truly move the needle.
One widely cited data point underscores the frustration: an MIT survey in August found that 95% of enterprises weren’t getting a meaningful return on their investments in AI. That gap between spending and results has fueled a growing debate across CIO offices, boards, and venture firms: when will enterprise AI shift from experimentation to durable value?
TechCrunch surveyed 24 enterprise-focused venture capitalists about what they expect in 2026. The consensus: next year will be the point when enterprises begin adopting AI in a more meaningful way, seeing clearer value, and increasing budgets—though several investors added important caveats about consolidation, governance, and vendor sprawl.
Why enterprise AI adoption has been slower than expected
The AI wave that followed ChatGPT’s release produced a surge in new products and an equally dramatic rush of funding. Yet large companies rarely transform overnight. Enterprise adoption is constrained by security reviews, procurement cycles, data governance, compliance obligations, and the complexity of integrating new systems into legacy workflows.
Another issue is expectations. Many early buyers treated large language models (LLMs) as general-purpose problem solvers. As investors now emphasize, LLMs can be powerful, but they are not a universal “silver bullet.” Enterprises have had to learn—in production, not just in demos—that reliability, evaluation, observability, orchestration, and data sovereignty matter as much as clever prompting.
The result is a messy middle: dozens of tools being tested, overlapping capabilities, uncertain ROI, and teams struggling to standardize what “working” even means for AI in a corporate setting. Several VCs believe 2026 is when that mess begins to resolve into a smaller set of scaled deployments.
Enterprise trends VCs expect to take off in 2026
When asked which enterprise-related trends will accelerate in 2026, investors pointed to a shift away from generic deployments and toward focused, governed systems that integrate with real work.
LLMs won’t be treated as magic—enterprises will prioritize the plumbing
Kirby Winfield, founding general partner at Ascend, argued that enterprises are learning that LLMs don’t automatically solve most business problems. In his view, companies will increasingly concentrate on foundational capabilities such as custom models, fine tuning, evaluations, observability, orchestration, and data sovereignty rather than assuming a model alone is the product.
Some AI product companies will morph into AI consulting-style implementers
Molly Alter, partner at Northzone, expects a subset of enterprise AI startups to evolve from selling a narrow product into acting more like AI consultancies. The dynamic she described: companies may begin with a defined wedge (like AI customer support or AI coding agents), then expand by deploying teams that build additional workflows on top of their platform—replicating a “forward-deployed engineer” model. Over time, specialized product vendors could become broader AI implementers once enough customer processes run through their systems.
Voice AI as a new primary interface
Marcie Vu, partner at Greycroft, highlighted voice AI as a major opportunity. She described speech as a more natural and efficient interface than screens and keyboards, and said she is watching for builders who redesign products, experiences, and interfaces around voice as the main way users interact with “intelligence.”
AI pushing into the physical world
Alexa von Tobel, founder and managing partner at Inspired Capital, expects 2026 to be a milestone year for AI’s impact on physical systems—particularly in infrastructure, manufacturing, and climate monitoring. Her thesis centers on a transition from reactive operations to predictive ones, where systems detect issues before they become failures.
Emily Zhao, principal at Salesforce Ventures, echoed interest in “AI entering the physical world,” pairing it with another frontier: the next evolution of model research.
Frontier labs may ship more turnkey enterprise apps than expected
Lonne Jaffe, managing director at Insight Partners, said she is watching how frontier AI labs approach the application layer. While many assumed these labs would focus on training models and leave product-building to others, she sees a possibility that labs will ship more turnkey applications directly into production—especially in finance, law, healthcare, and education—than the market expects.
Quantum: momentum, not a software breakthrough (yet)
Tom Henriksson, general partner at OpenOcean, characterized quantum’s 2026 outlook as “momentum.” He pointed to growing trust in quantum advantage and the increasing publication of roadmaps that clarify the technology. At the same time, he cautioned against expecting major software breakthroughs before hardware performance improves enough to cross a key threshold.
Where enterprise VCs say they want to invest
In 2026, several investors expect capital to flow not only to application-layer startups, but also to the infrastructure required to scale AI responsibly and efficiently.
Datacenters as “token factories”
Michael Stewart, managing partner at M12, said the firm is interested in future datacenter technology—what he framed as “token factory” infrastructure. His focus includes improving how efficiently and cleanly AI workloads run, with attention to cooling, compute, memory, and networking both within a site and between sites.
Performance per watt—and the limits of power-hungry GPUs
Aaron Jacobson, partner at NEA, argued that the industry is approaching the limits of humanity’s ability to generate enough energy for power-hungry GPUs. He said he’s looking for software and hardware that can unlock breakthroughs in performance per watt, including better GPU management, more efficient AI chips, next-generation networking approaches like optical, and rethinking thermal load in AI systems and data centers.
Vertical enterprise software with defensible workflows and data
Jonathan Lehr, co-founder and general partner at Work-Bench, said he’s focused on vertical enterprise software where proprietary workflows and data create defensibility—especially in regulated industries, supply chain, retail, and other operationally complex environments.
What “moats” look like for AI startups in enterprise
As AI models improve quickly and become more commoditized, investors increasingly argue that defensibility won’t come from having a slightly better model. Instead, it will come from integration, economics, and unique access to workflows and data.
Rob Biederman, managing partner at Asymmetric Capital Partners, said an AI moat is less about the model and more about economics and integration: being deeply embedded in enterprise workflows, having access to proprietary or continuously improving data, and building switching costs, cost advantages, or outcomes that are hard to replicate.
Jake Flomenberg, partner at Wing Venture Capital, expressed skepticism toward moats based only on model performance or prompting, arguing those advantages can disappear in months. His litmus test: if OpenAI or Anthropic released a model tomorrow that was 10x better, would the company still have a compelling reason to exist?
Alter also emphasized the difference between vertical and horizontal defensibility, arguing moats are often easier to build in vertical categories. She highlighted “data moats,” where each customer or interaction improves the product, and “workflow moats,” where defensibility comes from deep understanding of how work moves from point A to point B in a specific industry.
Harsha Kapre, director at Snowflake Ventures, framed the strongest moat as a startup’s ability to transform enterprise data into better decisions, workflows, and customer experiences—while working within governed data environments and without creating new silos. He said Snowflake Ventures looks for teams that combine technical strength with deep domain knowledge and can deliver targeted, trustworthy reasoning over existing enterprise data.
Will 2026 finally deliver real AI ROI for enterprises?
Investors surveyed by TechCrunch generally believe enterprise value will become clearer in 2026, but they disagree on how fast and how broadly it will show up.
Winfield expects enterprises to move away from chaotic experimentation with dozens of tools and toward fewer solutions with more deliberate engagement.
Scott Beechuk, partner at Norwest Venture Partners, said 2026 will test whether the application layer can convert infrastructure investments into value. As specialized models mature and oversight improves, he believes AI systems will become more reliable in everyday workflows.
Marell Evans, founder and managing partner at Exceptional Capital, expects progress but described it as incremental. She said there is still significant iteration ahead, and suggested that solving simulation-to-reality training could open opportunities across both existing and emerging industries.
Jennifer Li, general partner at Andreessen Horowitz, pushed back on headlines about enterprises not seeing returns. She argued that value is already present, pointing to how unlikely software engineers would be to give up AI coding tools—and said enterprise gains will multiply across organizations next year.
Antonia Dean, partner at Black Operator Ventures, offered a more critical lens: she warned that some enterprises may claim to be increasing AI investment as a way to justify cuts elsewhere or workforce reductions, and predicted AI could become a scapegoat for executives seeking cover for past mistakes.
AI budgets in 2026: bigger, but more concentrated
On spending, the dominant theme is not simply “more,” but “more concentrated.” Multiple investors expect enterprises to cut back on scattered pilots and consolidate budgets around proven vendors.
Rajeev Dham, managing director at Sapphire, said budgets will increase, but in a nuanced way. He expects organizations to shift portions of labor spend toward AI technologies or to see top-line ROI strong enough that the investment “effectively pays for itself three to five times over.”
Biederman predicted a bifurcation: budgets will rise for a narrow set of AI products that clearly produce results, while spending will fall sharply for everything else. Overall enterprise AI spend may grow, he said, but it will be concentrated among a small number of vendors while many others see revenue flatten or decline.
Gordon Ritter, founder and general partner at Emergence Capital, also expects concentration. He said enterprises will increase budgets where AI strengthens institutional advantages, but pull back from tools that merely automate workflows without capturing—and securing—proprietary intelligence.
Andrew Ferguson, vice president at Databricks Ventures, expects CIOs to push back on “AI vendor sprawl” in 2026. He described today’s environment as one where enterprises test multiple tools per use case because monthly spend and switching costs can be low, encouraging experimentation. As real proof points emerge, he expects companies to rationalize overlapping tools and redirect savings toward technologies that have demonstrated results.
Ryan Isono, managing director at Maverick Ventures, expects budget shifts from pilot and experimental spending into line items. He also anticipates a tailwind for startups as some enterprises that attempted in-house builds confront the difficulty of running production-grade systems at scale.
What it will take to raise a Series A in enterprise AI in 2026
In a crowded market, investors described a higher bar for enterprise AI startups looking to raise a Series A—especially on proof of adoption, referenceability, and staying power.
Flomenberg said the strongest companies combine a compelling “why now” narrative (often tied to GenAI creating new attack surfaces, infrastructure needs, or workflow opportunities) with concrete proof of enterprise adoption. He cited $1 million to $2 million in annual recurring revenue as a baseline, but emphasized that mission-critical status matters more than the number. In his framing, “revenue without narrative is a feature,” while “narrative without traction is vaporware”—and companies need both.
Jaffe advised founders to show they’re building in a market where total addressable market expands instead of evaporating as AI pushes down costs. She contrasted high elasticity markets—where a 90% price drop could lead to 10x market growth—with low elasticity markets where lower pricing could “vaporize the market,” leaving customers to capture all the created value.
Lehr said startups should demonstrate that customers use the product in real day-to-day operations and are willing to take reference calls and speak candidly about impact, reliability, and the buying process. He also expects companies to clearly show how they save time, reduce costs, or increase output in ways that survive security, legal, and procurement scrutiny.
Stewart noted a shift in how investors view pilot revenue. He said investors were recently skeptical of estimated annual recurring revenue or pilot revenue, but now pay more attention to customer willingness to evaluate solutions amid a flood of options. Still, he added that investors want to see conversions become the core of the story after six months of pilot usage—and argued that winning evaluations in 2026 will require quality and strong marketing messages, not only forward-deployed engineering.
Evans emphasized execution and traction, looking for genuine user delight and technical sophistication. She also cited “real contractual agreements” of 12+ months as a major signal, and asked whether founders can attract top-tier talent away from competitors or traditional hyperscalers.
AI agents in the enterprise by the end of 2026
Investors also weighed the likely role of AI agents—a fast-moving category that blends autonomy, workflow execution, and multi-step reasoning.
Nnamdi Okike, managing partner and co-founder at 645 Ventures, expects agents to remain in an early adoption phase through the end of 2026. He pointed to technical and compliance hurdles, and said standards for agent-to-agent communication will also need to emerge before enterprises can fully benefit.
Dham predicted convergence: rather than separate agents for inbound SDR, outbound SDR, customer support, product discovery, and other roles, he expects a single “universal agent” to begin emerging by late next year—one that shares context and memory across tasks, potentially breaking down organizational silos and enabling more unified conversations with users.
Dean stressed that winners will be the organizations that quickly find the right balance between autonomy and oversight. She argued that agent deployment will look more like collaborative augmentation than a clean split, with humans and agents working together on complex tasks and the boundary between roles continually evolving.
Conclusion
For enterprise AI, 2026 is shaping up—at least in the eyes of many enterprise VCs—as a year of consolidation and operational realism: fewer experiments, more governed deployments, and spending that concentrates around vendors that can prove reliability and ROI. Whether that delivers the broad payoff the market has been promising for years may depend less on headline-grabbing model advances and more on the unglamorous work of integration, evaluation, security, and change management.
<<>>
Related Articles
- VCs Say Enterprise AI Budgets Will Rise in 2026 — but Spending Will Consolidate Around Fewer Vendors
- Samsung Will Add Google Photos to Its TVs in 2026, Starting With Memories
- TechCrunch Disrupt Startup Battlefield: 32 Enterprise Tech Startups to Watch
Based on reporting originally published by TechCrunch. See the sources section below.
Sources
- TechCrunch
- https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
- https://url.usb.m.mimecastprotect.com/s/KiOXC93z2zCxkYp1u1ujFqkGSr?domain=openocean.vc
