Caitlin Bolnick Rellas

CRV • 6K followers

Vertical AI feels like a horserace 🏇 but after stepping back and thinking about the current state of market, I think what's equally unique is you have three different models competing at once. I think about it through this lens: 1️⃣ AI-enabled SaaS: the familiar path Take a proven vertical category and supercharge it with AI. Clear ROI, legible GTM, easy for VCs to understand. But real questions around differentiation and long term moats. 2️⃣ AI Roll-Ups: PE gets a facelift Acquire businesses, deploy AI to improve unit economics, and own the data + operations. Powerful if executed well, but operationally and financially complex. Big open questions around end state valuation. 3️⃣ AI-native industry disruptors: Become the thing Don’t sell software. Don’t buy incumbents. Build a fundamentally new version of the industry from scratch. Highest upside, but probably the highest casualty rate. You have to choose the model where the physics match the market. Each model has different capital needs, metrics, risk profiles, and founder skill requirements. I did a deep dive on each model 🤿 : ▪️where each model wins (and fails) ▪️how data dynamics actually differ across them ▪️why valuation expectations may diverge more than people expect ▪️and where I think the biggest outcomes will come from... As always, if you are building, investing, or just trying to make sense of vertical AI landscape, I always love to jam. Full post here: https://lnkd.in/g9a_9rtj

Tomasz Tunguz

Theory Ventures • 408K followers

While OpenAI signed $1.15 trillion in compute contracts through 2035, DeepSeek trained a frontier model for $6 million. This was 2025’s central question : are we building on bedrock or quicksand? The top 10 posts of 2025 examined some of these topics : Are we in a bubble echoing the telecom crash, or building the next internet? Do traditional exit paths still work when secondaries dominate & IPOs vanish? How do you design tools when the user is AI, not human? 2025 forced a reckoning with reality. 1. How AI Tools Differ from Human Tools (https://lnkd.in/d9qcnbyz) : I consolidated my 100+ AI tools into unified, parameter-rich interfaces based on Anthropic’s research. The counterintuitive finding : AI systems need complex tools with complete context, while humans need simple, chunked interfaces. Claude’s success rate approached 100% after the redesign. 2. Back to Text (https://lnkd.in/dWcy53fv) : How AI Might Reverse Web Design : I watched an open-source agent book flights by navigating airline websites, extracting data from visual chaos. If AI thrives on pure text, the future of the web might look exactly like it started : simple text, but for robots instead of humans. The better AI performs, the fewer websites we’ll visit. 3. Circular Financing (https://lnkd.in/d8-KMGZf) : Does Nvidia’s $110B Bet Echo the Telecom Bubble? : Nvidia’s vendor financing totals $110B in direct investments plus $15B+ in GPU-backed debt, 2.8x larger relative to revenue than Lucent’s exposure in 2000. But unlike the telecom bubble, Nvidia’s top customers generated $451B in operating cash flow in 2024. The merry-go-round has paying riders. Read the full post here : https://lnkd.in/d8Xw6tJt

Kit Yu

33K followers

Waymo is Making Progress But It's Still a Small Component To GOOGL's ~$4 trillion Current Enterprise Value: Waymo's latest funding round in late 2024, per information in TechCrunch, brought the company's valuation to more than $45bn. Since then, Waymo has launched in Austin, Atlanta, and San Jose, while also announcing plans to expand into 14 new cities in 2026. The rapid pace of city launches along with the miles driven and safety data in existing cities likely drives Waymo's valuation beyond its previous $45bn valuation. However, Waymo would still represent a small fraction of GOOGL's overall valuation and is likely not a near-term driver of valuation upside for GOOGL. For example, in a hypothetical scenario, if we were to value Waymo at $200bn (approx. the same value as UBER's global rideshare and food delivery business), that would only amount to ~5% of GOOGL's SoTP value at an implied $345 share price.

Paul Chiusano

Structural.Chat • 2K followers

Sam Schillace has a nice optimistic piece on how "building reliable things out of unreliable parts is possible" https://lnkd.in/gx-hTV2e This got me wondering: are LLMs the same sort of unreliable humans have engineered around in the past? In some ways yes, but they are also unreliable in "interesting" ways we haven't really dealt with before. As a point of comparison (also mentioned in the post), distributed systems are also fashioned from unreliable components (nodes that fail). But these failures are quite rare and tend to be uncorrelated, especially if you consider nodes in different regions. You can thus design very robust systems using the general principle of "try several instances, use whichever's working". It's still hard and there's decades of study around how to do this, but people are smart and have sorted it out. The stakes are lower, too. Even a single-node system can get a few 9s of uptime. You can get extra nines by being multi-node, multi-AZ, and multi-region, but you still have a plenty-useful system with fewer 9s - the 9s are uptime, not correctness. When the system is online, it's giving correct results. LLMs are not the same. An LLM that works correctly on a task even 95% of the time may be doing quite well. Naively assembling a system from components that fail 5% or even 1% of the time will pretty quickly turn into a useless system that fails 100% of the time! LLM failures aren't usually uncorrelated either, so the analogue of tricks used for reliable distributed systems don't apply. What else do we have at our disposal? Well, if your task has a deterministic verifier, then you can apply the "give LLM the verifier feedback, repeat" strategy, and this can be a more reliable ensemble... to a point. A verifier's feedback doesn't necessarily provide a sufficient gradient to the LLM (there are many tasks which are easy to verify as being correct, but hard to say what should be changed if incorrect). Humans can be in the loop, too, though humans aren't the best verifiers for "AI slop". At a certain point, if the work product is large enough, we cannot go through it in exhaustive detail and catch mistakes. Nor do people enjoy doing this. Nor are there great ways of writing complex programs that incorporate human feedback in rich enough ways, with sufficient control. Karpathy has this nice quote re: coding assistance that I agree with. For coding, I want a collaborative tool that I interact with and which ends with me feeling confident in the result. I don't want "a pile of code I am told works". It would be a different story if these models knew their own limitations and would only attempt tasks they could do with 100% reliability. But that is not the reality today. I think LLMs can still be very useful even in their current form, but there is a lot to still figure out about how to build real systems with them.

TatianaSFcom 🌉🟧

HackathonSF • 16K followers

Unpopular opinion, but an important one. Autonomous driving can work technically and still fail economically. Waymo is a great example. The tech is impressive. It will likely save lives. It may reshape cities. But the hard question is not autonomy. The hard question is unit economics. Unlike Uber, there is no natural network effect. More demand means: - more very expensive cars - more maintenance - more idle capacity for peak hours To deliver Uber-like wait times, fleets must be sized for peak demand. Most of those cars sit unused off-peak. That is capital heavy by design. Google can afford this. Not because the model works yet, but because Search prints cash. This doesn’t mean autonomy fails. It means autonomy alone is not a business model. The real question: Will it take tens of billions or hundreds of billions before this becomes self-sustaining? And who can survive long enough to find out? If you’re interested in real offline discussions about AI, autonomy, and startup economics, follow the events I organize on Luma: https://lnkd.in/g2Td2N3t For frameworks, breakdowns, and visual thinking, follow my SlideShare: https://lnkd.in/gYrkdj2P #autonomousdriving #waymo #ai #artificialintelligence #startups #founders #venturecapital #mobility #futureofmobility #deeptech #siliconvalley #startupthinking #uniteconomics #scalability #techbusiness #aibusiness #innovation #transportation #robotaxi #platformeconomics

Ali Rohde

Outset Capital • 22K followers

Super exciting to see reports this morning that Waymo is in advanced discussions to raise $10–15B at a valuation near or above $100B, with Alphabet expected to anchor the round. That would be a major step up from Waymo’s prior ~$45B valuation in 2024 and a clear step change in investor confidence. Notably, Waymo has not raised this year, despite the broader AI funding frenzy. The size and structure of the round also make this feel like a step toward an eventual spinout or IPO. This matters because capital supercharges growth. Waymo is past the “does this work?” phase and firmly in “how fast can we scale?” mode. The company has already logged ~14M rides this year and expects to reach ~1M rides per week by late 2026 as it expands into markets like Miami, Dallas, and Philadelphia, including expanding onto freeways. At the same time, competition and consumer expectations are accelerating. In markets where Waymo operates, fully driverless rides are becoming normal rather than novel. Once people can reliably hail a car with no one in the driver’s seat, tolerance for “pilot programs” and distant timelines drops quickly. The race is no longer about whether robotaxis work. It is about who can scale them first, safely, and everywhere — a win for consumers who get cheaper, safer, and more reliable rides.