Back to perspectives
Our outlook for 2025 is measured yet optimistic. Despite increasing complexity and noise in the startup landscape, large industries are being refactored with AI. The question remains: which teams and products will create enduring and differentiated value for customers? Here are our predictions for the year across our key investment sectors.
A few high quality companies going public will mark an end to the IPO drought, but still won’t lead to a flood of new offerings. The valuation divide will deepen between the fastest growing companies and everyone else. As a result, a wave of good, but not torrid, growers will become prime M&A targets. We’ll witness bigger SaaS platforms, including new category leaders, seizing the moment to build comprehensive suites through strategic acquisitions.
The magnitude of efficiency improvement, capacity to ingest multi-modal data, and ability to publish to other systems in a closed loop will cause customers to jump and initiate projects at a fast clip. Smaller markets that were considered ‘TAM constrained’ will become big enough to attract investment capital. A cohort of new market winners will be crowned. Some of these companies will be able to expand verticals and become multi-sector, large scale companies.
While the SaaS revolution has created incredible agility for enterprises, it has also created a level of complexity that is the enemy of security. SaaS platforms house massive amounts of data: customer information, proprietary algorithms, and operational data. However, these systems are riddled with vulnerabilities like over-permissioned accounts, unsecured APIs, and poorly understood integrations. A single misconfigured API can mean a breach affecting thousands. 2025 will be a year we’ll see major attacks targeting large enterprise SaaS implementations – the Snowflake attack was just the beginning.
After two disastrous years, Fintech stocks started to recover in 2024. This upswing will persist in 2025. Payments, fintech infrastructure, and enterprise SaaS selling to financial services companies will continue to be the best areas of focus in fintech generally. Still, a number of factors will drive pressure for AI companies to prove out clear ROI:
- Continued regulatory overhang even with changes in government administrations.
- Scrutiny over sponsor banks and their relationships with fintechs will add difficulty to landing and maintaining these relationships.
- False early signals from banks as they scramble to figure out their AI strategy and “shop around.”
Data infrastructure in its current form is mature. Tools that can manage data and ML workloads at scale are abundant. On the other hand, AI adoption is in its experimental phase. Despite being nascent, it is a top priority for most enterprises. Enterprises are pouring resources into AI before they have even nailed the basics of data that are necessary to deploy AI at scale.
As AI proves increasing value through experimentation and small features, teams will begin to build more complex workflows and push more AI apps into production. The requirements for AI tooling will then closely resemble those of data and ML infrastructure. The need for good hygiene around data quality, governance, right-sizing workloads, orchestrating complex data pipelines, performance, and scalability – themes that have been prevalent in data infrastructure – become even more critical. There will be meaningful opportunities for consolidation.
Winners in this market will be specialized AI that:
- Takes a close-loop workflow where ground-truth can be timely & automatically collected,
- Collects the ground truth as the workflow runs, and
- Systematically optimizes that workflow using reinforcement learning-like methods based on historical runs
One example is AI agents for Data Scientists = AI-generated scripts are automatically validated by executing the scripts to get results → the AI system learns and automatically optimizes based on these signals, potentially through a generative-adversarial network. Similar logic applies to troubleshooting AI agents for DevOps and to threat detection AI agents for Security Engineers.
AI agents applied to coding, search, legal, and customer support have radically enabled new levels of productivity and never-before capabilities. However, agentic solutions tackling highly manual, expensive, resource-intensive applications in the physical world have yet to take off at the same clip.
2025 will be a turning point. Agentic models paired with vertical workflows applied to applications such as AI-powered chip design, drug discovery, clinical workflow automation, industrial robotics, etc will accelerate time to market for new products, advance scientific discovery, and significantly enhance physical labor productivity. Ultimately, these problem spaces will attract missionary founders eager to tackle transformative challenges worth solving, and we are excited to meet them.
While investing in defense applications is increasing in popularity, it demands an entirely different perspective compared to investing in enterprise software. Multiple complex external factors in the wider risk landscape can drive a company’s success or failure in selling to the government. For example, Skydio, an American drone manufacturer, has supplied drones to Taiwan; as a form of retaliation, China announced sanctions against the company, which the company has said will negatively impact their battery supply chain. Other near-term factors also add volatility. Trump’s Deputy Secretary of Defense pick could mean good news for defense tech startups, or bad news to the broader market if it leads to self-dealing. Future Department of Defense spending remains uncertain, with some government advisory groups advocating for broad reductions, while others in the incoming administration are advocating for “rebuilding military capability”. These actions could hurt defense tech startups if the DoD clamps down, or could be a blessing in disguise if the DoD implements lower-cost alternatives or new contract structures.
Buyers will need to purchase both structured and unstructured data for different use cases, from building in-house healthcare foundation models to analyzing molecules in drug discovery. Training general healthcare models requires great volumes of data. Specialized needs, like those of pharmaceutical companies, require highly specific types and linkages of data. Buyers across pharma, med devices, and AI companies will start to make calculated, rational trade-offs between volume and detail, displaying increased price sensitivity. We’ll see startups in the data space move from being pure data broker models selling one-off data deals, to becoming full-stack partners – tacking on data cleaning and specialized analysis tooling to grow share of wallet and generate recurring revenue.
Overhyped categories like automated sales reps and emailing will falter. More valuable tooling such as Ideal Customer Profile research, CRM data entry, and identifying early markers of customer churn will rise. The winners in large sales organizations will be those who get better at market segmentation and targeting within those markets. They will build out different sales teams for large enterprise sales, mid-market, and SMB sales. “Referral sales” will be refactored as new sales tech emerges to help companies leverage relationships instead of just LinkedIn trolling.
Some of you might remember the days of “Intel Inside.” Foundation models will bring brand power to products building on them. This goes beyond tone of voice, which looks like this today (according to Anthropic’s Claude):
- GPT-4: The premium, measured consultant
- Claude: The careful, ethical advisor
- Llama: The scrappy, flexible generalist
Tone is tuneable. What really sets products apart is how foundation models apply understanding and recall to user needs to feel more intuitive. Most people won’t know it’s actually strong data science separating great product experiences from average ones.
There will be more clear breakouts in the AI-native app world, driven by carefully crafted product experiences based on the models that power them.
We’ll see more AI recruiting tools come out next year to help automate certain parts of the hiring process, and consolidation into established applicant tracking tools. While AI boosts productivity, it struggles with strategic tasks like talent gap analysis or reading non-verbal cues that bring an important nuance to hiring managers. AI can enhance the candidate journey and support, but can’t replace human recruiters.
Also, while hiring has been slow over the last year, I think we’ll start to see more roles open across functions, not just in tech positions. Companies will have to contend with multiple offers again – not at the fevered pace of 2022-2023, but definitely a tighter market than we saw this year.
Author
Written by
Share VIA
