Databricks Startup Forum 2026: VC and Founders Panel

Summary

The Databricks for Startups program expanded in 2026 to offer up to $200,000 in credits across Databricks and Neon, with the Lakebase team — which joined Databricks through the Neon acquisition — prioritizing speed of credit delivery so startups can build on a production-ready transactional database from day zero.
VCs from NEA, Essence VC, Hetz Ventures, and Databricks Ventures discuss how evaluation criteria for AI startups have shifted toward durable competitive moats, control points, and the distinction between product-led growth and enterprise sales strategies.
The 2026 Built on Databricks Startup Challenge featured pitches from Intello (AI agent teams for retail), LinkUp (web search APIs for AI models), and Visionite (agentic security platform), demonstrating diverse applications being built on Databricks.

Databricks Startup Forum 2026: VC and Founders Panel

Databricks Startup Forum 2026 brings together investors and founders to examine AI startup funding, durable competitive advantages, go-to-market strategy and infrastructure decisions. The forum also introduces expanded Databricks for Startups benefits, including up to $200,000 in credits across Databricks and Neon.
VCs from NEA, Essence VC, Hetz Ventures and Databricks Ventures explain how they evaluate AI companies, while founders discuss agentic workflows, data infrastructure, talent and build-versus-buy decisions. The 2026 Built on Databricks Startup Challenge features pitches from Intello, LinkUp and Visionite, demonstrating retail agent teams, web search APIs for AI models and an agentic security platform built with Databricks.
Databricks for Startups: https://www.databricks.com/product/startups
Startup Program announcement: https://www.databricks.com/blog/announcing-new-databricks-startup-program

Chapters

FAQs

What benefits does the Databricks for Startups program offer in 2026?

The program now offers up to $200,000 in credits across Databricks and Neon, a significant increase from prior offerings. The Lakebase team is focused on accelerating the speed at which credits reach startup accounts so teams can build immediately on a fully production-ready transactional database.

How are VCs evaluating AI startups differently in 2026?

Panelists from NEA, Essence VC, Hetz Ventures, and Databricks Ventures note that startup formation is happening at an unprecedented pace but building a company is harder than ever. VCs are increasingly focused on durable competitive moats and control points rather than simply the novelty of AI technology.

What companies competed in the 2026 Built on Databricks Startup Challenge?

The three competing startups were Intello, which builds AI agent teams for retail; LinkUp, which offers web search APIs for AI models; and Visionite, which is building an agentic security platform on Databricks. The challenge is part of the annual Databricks Startup Forum.

What infrastructure and go-to-market questions do founders face when building AI companies in 2026?

Founders on the panel discussed the build-versus-buy decision for data infrastructure and how to use models, context, and agentic differentiation to stand out in a crowded market. Hiring technical talent in the AI era and finding genuine product-market fit were also highlighted as major challenges.

Full transcript

[00:08] Uh we have an incredible panel for you today. We have 5,000 startups here this week from the US, AMIA APJ. We're going to do a quick welcome, then dive into our new data bricks for startups program, then a VC panel, our founders panel, and finally our built-on data bricks challenge. Our startup community is really the lifeblood of data bricks.
[02:07] I'd like to bring up Brad from the Neon team to talk about how we're investing in startups and relaunching our program. I'm a product manager on the lakebase team. I came into data bricks through the Neon acquisition about a year ago. We are significantly increasing the number of credits available to startups: you can now get up to $200,000 in credits across data bricks and neon.
[03:26] Data bricks is getting really good at serving startups from day zero. If you look at what lakebase does, it's a fully operational, fully production-ready transactional database. Startups need to move fast, so we are doubling down on the speed at which we can get credits into your account and get out of your way and let you build as fast as possible.
[04:52] I'd like to bring on Omar from our AMIA team to MC the event. First up we have our VC panel talking about where the puck is going in terms of investment dollars, AI trends, and the general venture landscape. Please join me in welcoming Josh Lily, our director of startups, and our esteemed members of the VC panel.
[06:00] Fast forward 12 months, a lot's changed. Startup formation is happening at an incredible pace and companies are getting created that weren't even possible a few years ago. But building a company is more challenging than ever. Founders are facing questions like: is software dead? What's a durable moat? What are the growth expectations of me as a startup?
[07:05] My name is Guy. I'm focusing on data automation and AI infrastructure. As part of Hetz Ventures we are a seed-stage Israeli firm, mostly the first check. I'm Tim, I run a fund called Essence VC, an infra-focused fund. We were early investors in a lot of infra companies like Tabular, MotherDuck.
[08:28] I'm Shawn, a partner at NEA, one of the larger venture firms. We've been investing in AI and data for over a dozen years including early investments in data bricks. I'm Andrew Ferguson. I lead Data Bricks Ventures. We invest in series A or later companies alongside a lead financial investor to build an ecosystem of partner companies.
[10:24] Tim, how has the evaluation criteria that VCs look at shifted over the last 12 months? We back so early stage. We still care about GitHub stars, but this day and age it doesn't feel as special. We're trying to find exceptional founders with a differentiated way of doing something.
[11:44] What was amazing about the data bricks team early on wasn't just the Scala code. It's the changing-the-industry community efforts they did so early on. The metrics may not matter as much. What is the next wave they can change? Can they change behavior shift like Spark changed how everyone writes code?
[13:21] Guy: 100%. Every founder now needs to be focused on go-to-market and distribution more than ever because coding is not an issue anymore. It's all about the product, the architecture, and how fast you can distribute and reiterate with different distribution channels, which are also changing in the new era of AI.
[15:00] Andrew and Shawn, what part of the stack has the most value capture? SpaceX exercised their option to acquire Cursor this morning. Durable value has completely changed. Being able to build features and code that works well is no longer a durable moat. You should assume someone can one-shot your entire product. Value shifts to durable control points up and down the stack.
[16:20] At each layer there are control points with durable value. Maybe a data set no one else has that compounds. Maybe integrations or business relationships hard to displace, like fintech rails. Cursor is a great example of distribution as a durable control point. We look for control points that will be durable and compound in value over time.
[17:40] Andrew Ferguson, how does an investment fit the data bricks ecosystem? Being a strategic investor is different. We think like a financial investor, but there has to be a second lens: does it fit the broader ecosystem, a better-together story where the partner product and data bricks let the customer do something they can't on their own.
[18:44] We look for enough white space between what data bricks offers and the partner. Have you gone through the partner program, do you have a common customer identified? If you can land in a data bricks customer account and cause that customer to consume more data bricks by opening a new use case, that gives our sales team a real incentive.
[20:04] Early on, is this product-led growth? Do I need to hire a CRO and sales team early? Classic product-led things like GitHub stars matter less. It depends where the company sits. If you're an infra-layer company doing something deeply technical, start with hardcore enterprise deployments. If you're at the app layer, get lots of users and a sticky flywheel.
[21:24] Diffusion of AI into the economy is rate limited by human factors on the receiving end. If you have embedded deployments with sophisticated enterprises and you are their thought leader on how to implement AI, that's durable strategic value, because you're controlling the durable human relationship and trust.
[22:11] Tim, when a founder is raising their series A, what's your advice? I invest in infrastructure; a lot of our companies don't have that kind of revenue. The most important part from day one: can you be the best company in a category that may not exist fully yet but should? Series A is about how much of the story you built from day one is playing out.
[23:33] Look at Spark early days: what metrics for a series A? Not traditional revenue, but the behavior changes it was making. When Hadoop was making noise, Spark came in with a very different way of doing something, and it's how it landed, the virality. Series A is scaling capital, but infrastructure is about hitting an inflection point.
[25:08] The framing I use: are you the best company, can you say you're the best in this category and what category? If you can't, the series is tough. But if you're very unique, there will be people who take risky bets, because cursor and many companies early on are not obvious but became huge because the non-obvious and huge are typically disruptive.
[25:40] Guy: I advise founders to get someone to pay for what you've built, hopefully more than one, and show you can scale a bit. There needs to be a flywheel starting in the go-to-market motion. Even more important is founders thinking about how it all ends together and the vision for how we win in the market.
[27:20] Rapid fire underhyped tech. Andrew: cost and token maxing; we'll see a backlash around spending for the sake of spending, and more small models, fine-tuning, custom-trained models. Shawn: intellectual property is really undervalued; with AI we might see a resurgence in the value of IP. Tim: I believe in code more with AI, especially formal verification.
[29:30] Guy: hardware, the borderline between hardware chips and software and optimizing for inference, especially GPU optimizations, inference, KV cache. We're just getting started. Give it up for our VC panel.
[30:25] Next, founders who are actually building, with insights on real world use cases, what it means to switch to agentic workflows, and the cost to value on scaling on data bricks. I'd like to introduce Dan Tobin and our founder leader panel.
[30:57] We'll start with you Mike. I'm Mike Poseman, CEO of Alpha Level, a security company; at the heart of what we do is identifying normal for our customers so you can start identifying novel attacks that rules miss. We're pre-seed with a little revenue, working with Fortune 100 companies on three continents with a five-person team.
[32:03] I'm Shinas, co-founder and CTO of Ourel, a go-to-market agentic operating system. We work with your legacy CRM or completely replace it, automating all aspects of your go-to-market life cycle. We are seed stage, raised 30 million, selling for about seven months on almost all continents, from AI native companies to the Fortune 500.
[32:52] I'm Callum Adamson, co-founder and CEO of Applied Computing. We build a foundation model called Orbital for energy operations focused on the hydrocarbon sector. We're post series A, getting towards 10 million ARR, about 50 headcount. I'm Phil, co-founder of Linkup. We build a web search API for AI models, for agents to find information to ground their answers. We raised a seed round and are based in Paris and New York.
[33:57] Colum, how do you think about your differentiation this year, the year of harnesses and context versus last year all about models? The model is the differentiator in our space. We're very narrow, a domain specific foundation model for a sector largely opaque to the SaaS industry.
[34:32] The industry is data literate but not good at harnessing that data because it's so vast and high velocity. Being able to access and use that data via data bricks is a point of differentiation. GTM is a differentiation; selling to these companies is super hard. Being domain specific and getting on the path towards world model capability is a massive differentiator for narrow AI.
[35:21] How have you thought about workflows as a differentiator? It's our focus towards the end of the year. Orbital is fabulously performant mathematically, but we maybe left workflows too late. We needed to be on the floor in the plants to adapt orbital to specific workflows. Making these workflows more dynamic for the sector is going to be huge.
[36:11] Shinyas, how have you communicated differentiation while raising capital? The go-to-market space has been fragmented for 25 years. Building in the AI native period comes down to putting all the context together, and to do that you really have to own the data, because getting it on demand is very expensive.
[37:14] Our differentiator was being the arbitrators of getting all this data, normalizing it, putting it in a context layer. That's how we raised 30 million, because this is how new age platforms are built: you have to have the data to power agentic flows.
[37:46] Phil, you're deepest in the agent space. There's one thing from the start with linkup: we built a web search index and crawl and scrape information from the internet at super large scale, and we do that thanks to data bricks.
[38:34] Our competition replicated what Google has done. We instead thought about what agents need, and decided to index not web pages but atoms of information found on web pages. Agents need the information so they can base their answers on it. That was a huge differentiation embedded in our stack from day one.
[39:22] We're based in Paris so we had to solve local data processing, GDPR, all that, and it became a strength. Now we can deploy in any region: Canada, Europe, US, Australia, Asia. Two years ago the idea that the main actor of the web would be AI agents was weird; now web searches by agents are 10x-100x bigger than by humans, so the vision is easier to sell.
[41:17] Mike, for early stage founders: build with open source plus cloud native, what's the trade-off? When we started, customer discovery made it clear it was beneficial to be cloud-native on everything. We don't want to own our customers' data, we just ingest it, run our algorithms and give it back, saving engineering and responsibility.
[42:22] You look at native AWS versus data bricks and you're paying a premium, and as a startup that's something to think about. But we get so much functionality that it accelerates our ability to deploy rather than spinning up open source piece by piece. We've introduced new customers to data bricks, and had access to resources within data bricks to become efficient.
[44:30] Callum: very similar to Mike, you're looking for an accelerant to value and willing to pay the premium. Several years ago we deployed Orbital first in India and thought everything had to be on-prem. As we moved into the states we heard: if you were on data bricks we wouldn't have to go through procurement, less governance risk. Every choice was made easy by paying the premium for speed to value.
[45:51] Shri: I'll second Mike and Callum. We're not at the optimization phase because building a startup you're always doing build versus buy. Should I build my own data infrastructure or buy it? For us it's important to focus on our users, so we needed a platform we could rely on and just build our product. Right now we're in build mode focused on delivering value.
[47:08] Talent is an incredible challenge, key to your success and how you're evaluated early on. Phil, our technical team is in Paris with a mathematically strong pool of talent who studied math more than CS, approaching technical discussions scientifically. One team member with no technical background studied coding when he joined and now builds product deployed in production.
[49:03] Those products are a series of endpoints easier to prototype, now built by our business team because they know what users need. We think about very strong technical talent or builders who think about business, go-to-market, product and aren't terrified of coding tools. We've built a research team training embeddings models to retrieve data efficiently at scale.
[49:53] Mike, you've accomplished a ton with a small team. Part of it is specialization: my co-founder is a statistician, which differentiates us from frontier models because statistics is not something LLMs do well. Cloud infrastructure and LLMs to build and code have allowed us to be lean; we built our own CRM with Claude in a couple weeks and an integration manager in a few days.
[52:04] As small as we are, largely because of my co-founder's background as a cyber security thought leader for decades, we have people reaching out constantly and get great resumes through our network. People say this is where security has been needing to go.
[52:52] Last topic, go-to-market. Shinas, you're the experts. Think of three things: first, product market fit, solving a compelling problem for a group of people at the right time. Second, distribution: if you're a niche product distribution is more important, and if you're a bigger platform build deeper, wider products to differentiate. Third, do you have the tools to stand out against competition using cutting edge stuff.
[54:44] Rapid fire lessons partnering with data bricks. Phil: we had three months to 10x the size of our index, 10xing pipeline throughput. We chose data bricks over building it ourselves and the speed of iteration is a big learning; we went faster than competitors. Callum: focus on what makes you special. We're not a data company, we're an AI company, that's why we partnered.
[56:07] Shri: to be a great AI company you have to be a great data company, with the right infrastructure to ingest data in every form. Getting data bricks in was an important differentiator so we could focus on agentic use cases driven by context. Mike: we have to get access to data quickly and work with it, and data bricks enables that easily. Give it up for our panelists.
[57:12] I'm Andrew Ferguson bringing it home. I'm excited to announce the three finalists for this year's Built on Data Bricks startup challenge, a global competition for early stage startups building B2B applications on data bricks. The grand prize winner is eligible for $400,000 in data bricks product credits and a $600,000 investment from Data Bricks Ventures.
[57:45] This is our third startup challenge, over 80 applications judged on team caliber, market potential, and innovative use of the platform, with a panel of two data bricks executives and two VC judges from Madrona and NEA. The three finalists are Intello (AI agent teams for retail merchandising), Linkup (production web search APIs for AI), and Visionite (unified pre-attack intelligence platform).
[58:50] Jeff Fish from Intello: we are the agentic workforce for retail merchandising and planning, building agent teams across retail, fully on data bricks. Merchandising and planning is the nerve center of retail; there's almost $2 trillion in lost sales annually to stockouts and over $140 billion in wasted inventory in luxury.
[01:00:10] We deliver capabilities across five agent teams from merchant analytics through pricing and promotion, creating personas across the merchandising team from executives to planners and buyers, thinking three seasons out about weather, economic status, GLP1s. Our agents deliver a system of action, not a system of record, powered by data bricks lakehouse architecture and compute.
[01:02:17] We work with brands like Dolce Gabbana, Balenciaga, Versace, Children's Place, Express, delivering inventory plans in minutes versus days, all on data bricks. Next, Phil Misrahi from LinkUp: AI models know a lot but not everything; without internet access they're like a capable smartphone in airplane mode. Linkup brings your AI models out of airplane mode.
[01:03:54] We built endpoints fitting the needs of AI agents across legal, go-to-market, finance, procurement, consulting. The one thing we're extremely good at is accuracy: the most accurate web search APIs on public benchmarks, both fast search and deep search. We do it through data bricks, crawling billions of pages every day.
[01:05:44] The thing nobody else does is identifying the atoms of information on pages, one embedding per atom instead of per page, making us more precise and granular. Our team includes the person who built the search engine at Algolia and the inventor of Splade. Data bricks processes information from the web page all the way to our vector store, and generates ML ranking signals like page rank.
[01:07:56] We use data bricks across the entire team, including science A/B tests and product/ops looking at logs. The final presentation, Nitsan Danielle from Visionite: we're building a live map of the attacker side of the internet. Your security teams see one half of the moon, your environment; the other half is where attackers build infrastructure before reaching you.
[01:09:52] We're proactive using our own unique data, tracking attackers as they set up infrastructure and stopping them before they reach your environment, reducing SOC noise. Everything is built for the agentic SIEM, with proprietary passive telemetry from every data center combined with threat intelligence into one source of truth. We rely entirely natively on data bricks, even an MCP server deployed on data bricks.
[01:11:43] Now the winners in reverse order. Third place is Intello, a $25,000 data bricks product credit. Second place is LinkUp, $75,000 data bricks product credit. The grand prize winner of the 2026 Built on Data Bricks Startup Challenge is Visionite. Thank you everyone for attending this year's startup forum.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.