​​Nobody can accuse 2025 of being dull. Between explosive growth in AI spending and an unpredictable economy, it was a year where nothing seemed to stay the same. Three of our Bain Capital Ventures partners look back at the highlights, the surprises and what they believe is coming next.

Abby Meyers, Partner, New York invests in growth stage AI and application software companies.

Rak Garg, Partner, Bay Area focuses on infrastructure software, cybersecurity and AI incubations, supporting technical founders creating the future of intelligent computing.

Saanya Ojha, Partner, Bay Area works with growth-stage founders building the future of AI, software infrastructure and cybersecurity.

What was a highlight of your year?

Saanya: Co-chairing our CTO and CISO advisory boards at BCV was one of the most rewarding parts of my year. It gave me a rare, bifocal vantage point: one lens trained on the bleeding edge of technical innovation, and the other on the slow, risk-calibrated march of enterprise adoption.

So when the MIT and McKinsey reports surfaced showing real friction in enterprise AI adoption, I wasn’t surprised. I’d been hearing those tensions echoed in every board conversation. Understanding where enterprises actually get stuck helps us spot founders solving real bottlenecks, not academic ones.

Abby: My highlight has been getting to know the incredible Pylon team as we co-led their Series B round. The energy in their office is palpable. You can feel progress and momentum happening in real time.

What surprised you about 2025?

Rak: 2025 was the year San Francisco staged a comeback. The city’s energy was unmistakable — cleaner streets, a wave of new cultural weirdness (including the now-legendary underground robot tournaments), and a tech community united by a shared mission: using AI to revitalize the real economy. The whole city is now operating with a renewed sense of purpose.

This year also reset expectations for what breakout startup performance looks like. I’ve seen multiple early-stage companies hit their first million in ARR within weeks, and several later-stage companies doubling or tripling revenue off already meaningful bases. AI has become a bellwether of the economy, with every company in every industry asking themselves whether they should and could be doing more. We’ve seen that momentum firsthand in our own portfolio, with companies like Decagon and Cognition posting extraordinary growth as AI adoption deepens across industries.

Saanya: Two things surprised me: Google got good again and open source caught up.

After being largely written off in 2024, Google came roaring back with state-of-the-art models and an impressive push to embed AI across its entire product suite. It was a reminder that scale and distribution still matter.

At the same time, open-source AI reached parity faster than anyone predicted. The gap between closed and open models has effectively collapsed. Kimi’s K2 Thinking model and DeepSeek V3.2, for example, now rival the top commercial systems on agentic and coding benchmarks.

The net result is that the one-model-to-rule-them-all era is over. Teams are increasingly building hybrid stacks, mixing and matching models to optimize for latency, cost and performance by use case. That’s a very healthy place for innovation to be.

Abby: For years, if not decades, we’ve been investing in the idea that tech can revolutionize physical industries and the real world economy. Until now, this has never felt like a consensus idea. It’s been a surprise that in 2025, legacy, unsexy industries (including logistics, lending, car washes, home healthcare) have seen surges in both investment and innovation. Founders and investors are realizing how AI can revolutionize legacy markets, with improvements in voice AI helping to open the floodgates.

What was the hardest call you had to make this year?

Rak: We’d known this founding team for years. They were sharp, well-referenced and already working with design partners in our own portfolio. But I couldn’t shake the sense that the product would hit friction as they scaled — some of it driven by buyer behavior, some by the short-term nature of the problem they were addressing. In the end, we passed on leading the seed. Ask me again next year, though; we may end up paying a premium for the A round.

Saanya: Good companies aren’t always good deals, especially in a hot market. Passing on businesses you genuinely like because pricing has detached from reality is always hard. But discipline compounds. Saying no when everyone else is leaning in reinforces your long-term edge.

Abby: The hardest moments for me have been knowing when to tune out the hype and trust my own instincts and due diligence. It’s never easy to hold your ground when the market is stampeding toward a herd of hot AI deals or a buzzy category you don’t believe can produce a generational company.

What’s a bold prediction for 2026?

Abby: Incumbents are finally poised to make real moves in AI in 2026. Better foundation models and orchestration tools have lowered the barrier to shipping meaningful AI products, while AI-native challengers have pointed to where demand is strongest and what great tools actually look like. Incumbents will use both the new capabilities and some hard-earned lessons to fight back, whether by building or buying. The result should be a wave of innovation on both sides.

Rak: ​​In 2025, Anthropic uncovered what appears to be the first large-scale cyberattack predominantly engineered and distributed through AI, allegedly carried out by Chinese state-sponsored actors. In 2026, I expect this to escalate: Adversary groups will increasingly use AI to steal data, intellectual property and critical resources from large organizations in sectors like semiconductors, energy and rare-earths.

This will become a clarion call for a new generation of security products — offensive and defensive — as companies adapt to the reality of an emerging, always-on era of AI-driven cyberwar.

Saanya: I agree with Andrej Karpathy: 2025 was not the “year of AI agents.” What we’re seeing is the opening chapter of a decade-long evolution. Models are becoming natively multimodal, with longer-horizon reasoning and the ability to execute multistep workflows. The next wave of value won’t come from raw model quality alone, but from the scaffolding that enables these systems to operate safely, predictably and productively inside organizations.

The biggest opportunities will be in the frameworks and infrastructure that give agents structure: boundaries, context, monitoring and feedback loops. This invisible layer is what will transform raw capability into enterprise-grade reliability.

Describe the founder you want to meet in 2026

Abby: My answer is the same every year: I’m drawn to founders who aren’t just interested in building something, but who feel compelled to build this thing. The ones who are so deeply motivated by the problem that they’ll run through walls to solve it.

Saanya: The founder I’m most excited to meet is obsessed with the problem, not just the technology. I’ve seen brilliant technical founders over-engineer platforms that solve everything in theory and nothing in practice. The best founders do the opposite. They fixate on a narrow, repeatable use case where they can deliver unmistakable value, and only then earn the right to expand. You can’t launch as a platform. You earn your way to becoming one.

Rak: I want to work with founders who have experience working with and hiring world-class talent, because that shapes how they recruit. Whether it’s early team members or even their first customers (who effectively “hire” themselves into shaping the product roadmap), these founders are ruthless about qualifying people out. Every startup I’ve seen with a “butts in seats” mentality has failed. The best founders know exactly what they’re looking for, and they relentlessly network their way into bringing those people into the company.