What YC Demo Day Actually Feels Like

I have been to startup events before. I have met founders before. I have listened to pitches in rooms that were supposed to feel important.

YC Demo Day is different.

We got there around 8:30 a.m. Coffee first, pastries next, and then straight into the main stage where company after company would pitch. This Winter 2026 batch was massive. YC’s public company directory shows 199 W26 companies, and the day moved with the kind of compressed intensity that makes time feel both fast and heavy at once. The pitches happened in rapid succession, with breaks every roughly fifty companies, and there were overflow viewing options on other floors via live broadcast. But the main room was where you wanted to be. That was the center of gravity.

The atmosphere was electric, cramped, polished, slightly performative, and unmistakably San Francisco. In other words, it felt exactly like venture capital when venture capital is most alive. Founders had clearly thought hard about how to present themselves, and it showed. Different styles, different personalities, different technical domains, but the same underlying pattern: precision, signaling, and conviction. They were not trying to explain every layer of the business. They were trying to make you understand, in under a minute, what mattered, why it mattered, and why they were the team to do it. That compression is part of what makes Demo Day so effective.

And that is also what people miss if they have only ever seen startups online.

Reading companies on the internet is one thing. Watching nearly two hundred of them presented in sequence, in person, in a room full of capital and urgency, is something else entirely. You see the product, the framing, the ambition, and the founder all at once. You feel the speed at which conviction forms. You also feel the emotional mechanics of the room. There is real FOMO. There is real social proof. There is real pattern recognition happening in real time. The desire to write checks is not abstract there. It is physical.

That, to me, was one of the most striking truths of the day. Investors were not just politely curious. They were moving. By the time Demo Day rolled into happy hour, it turned into a frenzy. Everyone was trying to talk to everyone. Conversations stacked on top of each other. People were chasing obvious winners, quietly circling stranger and more ambitious ideas, and learning through side conversations as much as through the formal pitches themselves. Some investors were underwriting on the spot. That willingness to move fast was one of the biggest surprises of the day, along with the sheer level of the batch itself. Public commentary around the event pointed to 14 companies already at $1 million or more in ARR, about 90% of the batch working with AI in some form, and acceptance below 1% from roughly 23,000 applications. That changes the temperature in the room.

It also sharpened something I already believed– real-time inference is going to consume the internet economy.

That sounds dramatic until you spend a day watching what founders are actually building. AI is no longer a narrow application layer. It is becoming a system-level technology. Garry Tan has been talking publicly about this from multiple angles. At Demo Day, the batch reflected it. Outside the event itself, Tan’s open-source gstack project also made the same underlying point in a different form: software can now be built much faster when AI is embedded into the workflow itself, not treated like a bolt-on assistant. Gstack’s GitHub repository describes it as a stack of opinionated tools for product, engineering, release, docs, and QA, which is basically another signal that the software production function is changing fast.

The implication is bigger than software productivity.

The internet economy as we know it was not built for inference. It was built for search, social, SaaS, and transactions. Real-time inference has different physical and economic requirements (lower latency, tighter scheduling, more dynamic compute, and infrastructure that can support interactive, stateful, always-on applications). That is part of why we spend so much time thinking about infrastructure, and why Simply Silicon exists. Simply Silicon is building AI token factories in urban areas to enable real-time inference applications. After YC Demo Day, that thesis felt less like a contrarian bet and more like an obvious direction of travel.

Another thing that stood out was the composition of the batch. There was a visible seriousness to it. More hard technical companies. More infrastructure. More companies aimed at changing how the physical economy or core compute stack works. The batch was very strong (in my opinion the strongest batch yet) and especially tilted toward hard technical problems. That lined up with what we felt in the room. It also lines up with the broader market turn toward what Apollo recently called HALO: Hard Assets, Low Obsolescence. That framing matters because in an AI era, some of the most durable opportunities sit where technical progress meets hard constraints in the real world.

That is exactly why Augur was super excited to underwrite: Beyond Reach Labs and Cumulus.

Augur invests across the watt-bit value chain in foundational technologies that deflate the levelized cost of intelligence. Those two companies fit that thesis.

Beyond Reach Labs is building deployable space solar arrays that grow from the size of a dining table to the size of a football field in orbit. The company can deliver 10 times more usable power without increasing launch mass or volume, and frames the demand pull clearly: power in orbit is becoming a binding constraint for orbital datacenters, commercial space stations, and lunar outposts. That is why we backed them. This is not a marginal improvement story. It is a system-one technology for the space economy. If you can deliver orders of magnitude more power per launch, you do not just improve existing missions. You expand what becomes possible.

The founders are exactly the kind of team you hope to find before the rest of the market fully catches up. Mitchell Fogelson, co-founder and CEO, earned his PhD at Carnegie Mellon, where he worked with NASA on kilometre-scale deployable space structures, and has published in top robotics and aerospace venues. Pele Collins, co-founder and CTO, spent years at SpaceX leading Dragon parachute engineering and also held a technical leadership role at Commonwealth Fusion Systems. That combination matters. This is a company attacking a brutally hard problem with founders who have already lived inside high-consequence engineering environments. We are incredibly excited to have Beyond Reach Labs in the Augur portfolio because they are building enabling infrastructure for a future space economy that will require far more power than today’s architecture can support.

Cumulus hits a different part of the same macro shift.

Cumulus Labs is building a fast multimodal inference platform centered on its proprietary Ion engine, with a value proposition built around faster performance, lower cost, and near-zero cold-start latency for teams using fine-tuned and open-source models. Its YC materials frame the problem well: AI teams waste money on idle GPUs, spend too much time debugging infrastructure, and tolerate inference spin-up times that are bad enough to break user experience. They can acheieve 50 to 70% savings by charging based on actual compute resources used and by optimizing workloads across their infrastructure. The unlock here is straightforward: if you can make GPU access feel closer to serverless compute while materially reducing cold starts and infrastructure burden, you make real-time inference cheaper, faster, and more usable.

Again, the founders are cracked. Veer Shah studied computer science at the University of Wisconsin-Madison and worked on Space Force and NASA-related programs at an aerospace startup. Suryaa Rajinikanth studied computer science at Georgia Tech, worked as a lead engineer at TensorDock building a distributed GPU marketplace, and later deployed critical AI systems and infrastructure at Palantir. Augur is extremely convicted in underwriting Cumulus because they are going straight at one of the core constraints in the inference stack: cost-effective, responsive compute that does not force teams to choose between latency, utilization, and operational pain.

That was the real feeling of the day.

Yes, it was exciting. Yes, it was cinematic. Yes, it was a venture spectacle. But underneath that, it was also clarifying. Demo Day did not just showcase companies. It compressed a view of where the economy is going. More AI, obviously. But also more infrastructure, more hard tech, more foundational systems, and more urgency around the layers beneath the application surface.

That is part of why it felt so different from the Canadian venture norm. The scale of conviction was different. The speed was different. The willingness to underwrite ambition was different.

We left more energized than when we arrived. Grateful to have been in the room. Grateful, too, to back founders like Mitchell, Pele, Veer, and Suryaa, who are building at the layer that actually matters when a technological shift becomes real. The next economy will not be built by surface-level software alone. It will be built by the people rebuilding the underlying systems. And we are very proud that Augur’s portfolio already includes two of them.

• Michael Brown

Co-Founder, Ventures Edge | Investor, Augur VC | FinOps, Simply Silicon