Ashwin Rao

India is not behind on talent. It is behind on risk appetite. Building LLMs or foundational AI infra is not about capability. It is about capital. Not just any capital, but capital that is comfortable with long timelines, unclear outcomes, and massive burn. Most of India’s funding still chases near-term traction. SaaS, fintech, consumer. Sectors where returns are easier to model. But AI infra does not play by those rules. It demands belief before proof. The Middle East launching a sovereign fund to accelerate AI through public private bets is not just money. That is intent. India will not build foundational AI by hoping startups will figure it out alone. It needs conviction at the system level. And a willingness to back uncertainty at scale.

56

10 Comments

I’m delighted to announce that AI Fund has closed $190M for our new fund, in an oversubscribed round. I look forward to working with many more builders to create new companies that serve humanity. AI Fund isn’t a traditional venture capital firm that invests in existing businesses. Instead, we are a venture builder (also called a venture studio): We co-found AI companies, so our team is directly involved in writing code, talking to customers to get feedback, iterating on product designs, preparing market analyses, and so on. We have a lot of fun building multiple AI products at a time, and thus live daily the emerging AI startup best practices. Many factors go into the success of a startup. But if I had to pick just one, it would be speed. Startups live or die based on their ability to make good decisions and execute fast, which has been a recurring theme of my posts here as well. If you are building an AI startup, here are some ideas to consider: - A startup with a small team that pursues one focused, concrete idea can move really fast. Rather than hedging, it is often better to pursue one hypothesis (for example, build one concrete product) but also be willing to switch quickly to a different hypothesis (say, change what features you decide to build) if the data that comes back indicates the original hypothesis is flawed. Concreteness gets you speed! - A subject matter expert’s gut is remarkably good at making quick decisions. Obviously, there’s a role for data and user studies as well. But if you’re deciding whether to build feature A or B, or to sell first to user persona X or Y, sometimes a domain expert’s gut will point to a quick decision that you can execute and validate or falsify. Trusting a domain expert’s gut gets you speed! - AI-assisted coding is making prototyping faster than ever before. Yes, AI assistance is speeding up building reliable, enterprise-grade applications and maintaining legacy codebases. But the acceleration it brings to building stand-alone prototypes is far greater. This is because stand-alone prototypes have low requirements for reliability, integration, or even security (if, say, you run them in a sandbox environment). This lets us prototype and test at a ferocious velocity. AI-assisted coding (including vibe coding, where you might barely look at the code) gets you speed! - Finally, with faster prototyping, the bottleneck shifts to getting feedback from users. A single learning cycle might consist of (i) building a prototype and (ii) getting user feedback to inform the next iteration. Since (i) is now much faster, accelerating (ii) is growing in importance. This means teams that are skilled at finding prospective customers and getting their feedback in hours/days rather than weeks can go faster. I’m grateful to AI Fund’s investors, team, and entrepreneur partners for working with us. There is much ahead to build! [Shortened due to length limit. Full text: https://lnkd.in/gBaqAAcF ]

6,016

290 Comments

I was talking to my friend Parul Singh about how too much VC money can be detrimental. I have been working with pre-seed/seed/A at and with companies for the last 20+ years and I have observed first hand that the saying “constraints breeds creativity” is true. In my experience, a company with 18 months of runway gets much more done in the second nine months than the first. 🪹 Pre-seed Companies: When you are in the “idea maze” (coined by Andreessen Horowitz) or “drunken walk” having too much money does not necessarily speed up progress. Of course you may need some money for founder living expenses but having a larger team wandering through the "idea maze" together doesn't necessarily make the journey go faster. There is a certain level of founder reflection that has to happen while navigating the “idea maze” and too many distractions (people, money) makes it hard. I teach an entrepreneurship class  (LeanLaunchpad) every year. Out of the eight teams we have navigating the “idea maze”, about half will test and land on a reasonable idea in just ten weeks (including a simple MVP and even some paid POCs). Total budget: zero. With the multitude of AI tools, time to PMF is accelerating. Check out https://lnkd.in/guM7Ymn9 (AI companies revenue >10x SaaS companies). 🌱 Seed Companies: When you are going from the “idea maze” to “product market fit” and are trying to first build a product that a few customers love and second build a way to acquire these customers efficiently, having too much money can actually be counterproductive. Why? Because too much money often translates to wasting time on things that don’t matter: you are not forced to focus on the features that are absolutely critical, you can spend in non-efficient marketing methods (skipping the crafting of the product-growth loop) and you can afford to over hire or to keep the wrong people onboard. 🅰️ Series A Companies: Post product market fit, when it is time to scale, money is absolutely critical. However there is one common failure mode: the “fake Series As”, the As that get overpriced without really having reached product-market fit. Take a seed in a hot space, a small team that has grown revenue to a few $100Ks or maybe even a million in just six months. There is little data on actual revenue stickiness or whether the founder-led sales motion can be replicated efficiently. There is a lot of competition to invest in the company and they end up raising $20M+ series A.  The founder feels confident, they are stretched-thin and they hire a bunch of folks. There is pressure to scale but the machine is not built yet (the product is not quite there, they can’t figure out how to get to customers efficiently) and things unravel. Venture money is like carbs: quality over quantity (not all money is equal); take in moderation (constraints bread creativity) and if you compete in endurance sports, you might need more carbs to perform well (if you are growing, you will need the cash).

199

16 Comments

At this year’s Thoma Bravo annual LP meeting, the futures of both SaaS and AI — and their interconnection — were naturally on the mind of our investors. And while history doesn’t always repeat, it does often rhyme — so to help clarify, I felt it would be valuable to review some lessons from the pace and process of some of history’s transformational technologies. In our view, the first-generation digital transformation journey still has immense, untapped opportunity based on our experience navigating the industry’s evolution from on-premise to cloud software. I shared a few slides showing that: SaaS revenue is only approximately 25% of total expected software revenue in 2025, with projected 19% annual growth from 2025 - 2028. Many large enterprises still maintain significant on-premise infrastructure — particularly financial systems. Companies, including some of our portfolio companies, have been managing enterprise cloud and SaaS transitions for over a decade, and there is still plenty of runway left in this transition. So what accounts for that runway? Yes, technology evolves quickly. But products, organizations, and business processes that use technology effectively take multiples of that time to adapt and utilize. And the large enterprises that embark on transformations don’t do so at the pace of technology itself; they take each step with measured intention. As they should, given the stakes. The same dynamics are currently at play in AI. It's important to note that what AI foundation models can do right now is laps ahead of what software products can deliver, which is itself far ahead of what most enterprises are ready to implement. Because of that deliberate nature, proprietary data, regulatory compliance, and seamless user experiences will likely become crucial to proving the case for accelerating value — and that’s where data is a company’s competitive moat. We believe that — when combined with the data that sits within them — AI will benefit enterprise software customers in at least two key ways: 1) Democratize access so that more users can make use of deep business insights built from massive data sets; and 2) Enable customers to make real-time operating decisions at scale, with greater speed, precision and less labor input. For investors, this perspective is important. Creating economic value means using technology, not just deploying it. As basketball coach John Wooden famously said, "Be quick, but don't hurry." In enterprise software’s evolution, we believe that's the key to creating enduring value.

1,507

69 Comments

Chowdeck’s rise is not accidental, it’s a deliberate execution of frameworks that balance unit economics, logistics optimization, and vertical integration. Here's my technical breakdown: Unit Economics & Contribution Margin Discipline Take Rate Engineering: Chowdeck sustained >25% take rate per order by setting commission structures that covered rider payouts, platform operations, and marketing without relying on subsidies. Contribution Margin Monitoring: They calculated CM per order (revenue – variable costs) daily, enforcing a minimum CM threshold before deploying promotions. This prevented negative-margin growth cycles. Cohort LTV:CAC Ratio: Each city expansion was evaluated against projected lifetime value of a user (frequency × AOV × retention × take rate) versus acquisition costs. This kept payback periods within 3–6 months. Logistics Optimization through Network Design Dynamic Rider Routing: Using demand-forecasting models, Chowdeck geofenced cities into high-density clusters and dynamically routed riders to minimize idle time. Fleet Mix Optimization: Motorcycles for high-distance orders, bicycles for dense urban cores, lowering last-mile delivery costs while maintaining SLA compliance. Expansion Methodology: Breakeven-First Framework Market Entry Playbook: Each new city was assessed on three axes: merchant density, rider supply elasticity, and projected CAC vs. retention. Only when breakeven was achievable within 8–12 weeks did Chowdeck greenlight expansion. Zero-Paid Ads Growth Loop: In Ghana, Chowdeck leveraged merchant-side virality (restaurants encouraging orders through Chowdeck-exclusive offers), creating a pull model that lowered blended CAC significantly. Operational KPIs for Expansion: Daily order velocity, active merchant ratio, and delivery SLA compliance formed the go/no-go framework for scaling headcount and marketing spend. Quick Commerce & Dark Store Rollout Demand Heatmapping: Chowdeck used transaction-level data to identify high-frequency SKUs (e.g., groceries, OTC meds) and positioned dark stores within a 2–3 km radius of dense demand clusters. 15-Minute SLA Design: Inventory planning and SKU assortment were algorithmically optimized per neighborhood, reducing stockouts and enabling 15-minute delivery promises. Vertical SaaS Integration via POS Acquisition Closed-Loop Merchant Data: By acquiring a POS provider, Chowdeck integrated real-time merchant inventory into its platform. This reduced “phantom inventory” and increased order completion rates. Demand-Supply Synchronization: Merchant-side data fed into Chowdeck’s logistics algorithms, enabling predictive dispatch (rider is en route before food is fully prepped). Chowdeck’s success is a masterclass in unit economics discipline, algorithmic logistics, and vertical SaaS integration. This methodology is not limited to food delivery, it’s a replicable framework for scaling in fragmented, high-volume, low-margin sectors across Africa and beyond.

16

The $100K H1-B policy could reshape Silicon Valley completely. Let's be clear: there WAS a problem. IT consulting firms have been gaming H-1B visas for years. But we solved the consulting arbitrage problem with a sledgehammer when we needed a scalpel. 55% of America's billion-dollar startups have immigrant founders. But most of them didn't arrive as founders. You can't actually start a company on an H-1B. The actual path looks like this: Student visa → H-1B job → build networks → get green card → start company. They became founders or tech leaders after years of working here, meeting co-founders at hackathons, understanding what Americans would buy, building credibility with VCs. The magic of Silicon Valley wasn't importing proven talent. It was attracting unproven talent and giving them room to become exceptional. This is what’s made it different from every other innovation economy in the world. At $100k per bet, those chances don't get taken. Those builders go elsewhere. Startup teams will become distributed by default. Research labs and core operations might move offshore, while still headquartered in US (for bragging rights). While the U.S. still gets capital gravity, the intellectual center of gravity (the weird ideas, the late-night hacking, the "what if we tried this" conversations) disperses. We'll keep the venture capital. The IPO market. The fancy headquarters. But the garages where magic happens will be somewhere else. Link in comments below.

25

2 Comments

This section on how dbt Labs transitioned from a purely PLG motion to layering on enterprise is a fascinating one. Two instructive passages that I have bolded (below from First Round Review's path to PMF series featuring dbt Labs) TLDR: GTM comprises your ICP, channel, and message. When you transition from PLG / bottom up to Enterprise / top down motion, naturally your ICP and channel also changes, but your messaging / proposition needs to change too, e.g., the enterprise may be multiple personas and are buying assurance as much as solution. --- "Handy found product-market fit organically for dbt as an open-source tool mostly used by data practitioners and developers. But a few years into running a commercial business, he realized he had to build a growth curve all over again with C-suite data leaders. “Even though we had an unbelievable amount of market pull, as we initially commercialized, it wasn’t easy for us to transform this open-source command line tool into a product that enterprises would pay a million dollars for,” says Handy. “When you have enough product-market fit, sometimes it allows you to get away with not being super tight on product marketing or sales motions. So around 2022, we went from this gigantic acceleration curve and overnight we realized, we have to sit at the adults’ table and figure things out real fast,” says Handy. After the PLG flame started to fizzle, Handy turned his attention to layering on a sales-led motion for the cloud platform. “We had to focus our efforts on telling cohesive stories to senior data leaders. We had to have a very clear, explainable answer to the question, ‘Why should I use the commercial product and not the open-source product? And it had to be digestible by someone with a C in their title,” he says. Handy’s answer: dbt Cloud can handle complex data for companies of every size. “The longer people used dbt, the more complex their code became,” he says. “It was a problem for the most sophisticated dbt users, who were often at the largest companies. So there was a real opportunity for us to step in and solve that for them with dbt Cloud.” To tell that story to enterprise customers, Handy relied on data, naturally. “At a user conference we presented a chart that showed the number of dbt projects that had a certain number of models in them — over 100, over 1,000, et cetera,” he says. “We watched that number climb and we knew as users ourselves, ‘Oh my God, trying to work in a dbt project with 5,000 models in it is challenging.’ So we started with that quantitative data point and asked folks in our community about their experiences with these very large, complex dbt projects, and validated that this was a pain in the ass without a cloud platform.”

23

4 Comments

Meta just hit Command + Zuck on its AI strategy - shredding the open-source playbook and replacing it with one that reads: Compute. Talent. Secrecy. The vibe is no longer “open source for all.” It’s “closed doors, infinite compute, elite team, existential stakes.” Let's break it down: (1) Compute: Zuck’s Manhattan Project Meta is building gigascale AI clusters. Prometheus comes online with 1 GW in 2026; Hyperion scales to 5 GW soon after. For context, Iceland’s total electricity consumption is ~2.4 GW, Cambodia is at ~4 GW. Meta’s Hyperion cluster alone could out-consume entire nations. These clusters are for training frontier models - GPT-4-class and beyond. In this new regime, FLOPS per researcher is the KPI, and Meta is going from GPU-starved to GPU-dripping. Each researcher now has more compute to play with than entire labs elsewhere. That’s not just good for performance, it's a hell of a recruiting pitch. (2) Secrecy: From Open Arms to Closed Labs Meta won developer love by open-sourcing its LLaMA models. But it also accidentally became the free R&D department for its own competitors. DeepSeek AI, for example, built on Meta's models and vaulted ahead. Now Meta is reportedly shelving its most powerful open model, Behemoth, due to both internal underperformance and external regret and shifting toward a closed frontier model, aligning more with OpenAI and Google. This is a massive philosophical reversal from “open wins” (as Yann LeCun would say) to “closed dominates.” (3) Talent: Just Buy Everyone Comp packages reportedly range from $200 million to $1 billion for AI leads. All AI efforts are now housed under a new unit, Superintelligence Labs, run by Alexandr Wang (ex-Scale AI). This elite team is small, only ~12 engineers, working in a separate, high-security building next to Zuckerberg himself. Forget beanbags and 10xers. This is a DARPA-style moonshot with a trillion-dollar company behind it. Zuckerberg has said, basically, “Look, we make a lot of money. We don’t need to ask anyone’s permission to spend it.” He’s not wrong. While OpenAI, Anthropic, and xAI rely on outside capital to fund their ambitions, Meta runs on a $165B/year ad engine. And unlike Google and Microsoft - who have boards, activist investors, and share classes that allow for dissent - Zuckerberg controls Meta, structurally and operationally. Meta’s unique dual-class share structure gives Zuckerberg over 50% of the voting power, even though he owns less than 15% of the company. He doesn’t need anyone’s approval, he can build whatever he wants. This makes Meta less like a public company and more like a founder-led sovereign AI lab - with Big Tech cash and startup flexibility. That governance structure is a strategic weapon, letting them place bold, long-term bets at breathtaking speed. Meta’s open-source era is over. This is the closed, compute-soaked, capital-fueled empire play. Less GitHub, more Los Alamos.

1,119

68 Comments

Factories Are Ready for a Reset YC’s modern metal mills thesis: ▫️ AI-driven planning and scheduling ▫️ Real-time execution ▫️ Energy and automation designed in from day one Software-defined factories. I mapped 10 ways to attack this, where bottlenecks actually are: https://lnkd.in/dNMnt7Xt

44

2 Comments

I had dinner the other night with a fascinating group of folks building AI-enabled products. PMs from companies like Perplexity and Harvey, plus engineers and founders. One question kept coming up: Why haven't incumbent software providers been able to leverage their valuable data assets yet? Both giants like Salesforce and growth-stage companies with 10+ years of collected data seem to be struggling to deliver truly differentiated AI experiences. Most of us would assume these companies are sitting on data gold mines that should give them a massive advantage. But is that actually true? One perspective I found compelling: The historical data these companies have collected simply isn't the right kind to make AI models truly effective. The workflow data, outcomes data, or database info they've aggregated over the years might not be as useful as we'd think. What's most valuable in building agentic AI isn't just workflow data or outcomes data—it's a granular understanding of how humans actually perform tasks. This includes all the context switching between applications and even offline activities that traditional software workflows don't capture. The real value is understanding the detailed ontology of how tasks get done so you can replicate that in an agent. AI-native companies are building these task-level datasets from scratch and at the same time incumbents are trying to figure this out—potentially leveling the playing field. The counter perspective: Incumbents aren't failing; they're just slower. Over the next 12-18 months, we'll see them release AI that delivers differentiated insights thanks to this accumulated data. Their data is valuable; they're just taking longer to leverage it effectively. I don't know which view is right, but both offer interesting frames for thinking about the AI landscape and who might win in the long run. Curious what you all think—will incumbents' data moats prove valuable, or are AI-native companies starting on more equal footing than we might assume?

604

150 Comments

In 2010, Larry Ellison owned 23% of Oracle. Today, he owns 40%. How is that possible?     Capital Allocation. Oracle's resurgence in AI is getting alot of news but what we think is even more compelling is Oracle's understanding of capital allocation. The tech industry is often at the leading edge of innovation and business model transformation, but there's one area where the tech industry has under-optimized. #CapitalAllocation. In corporate America, capital structure is often at the core of optimizing shareholder value while in tech, it's the opposite. How often do tech founders and Boards think about the cost of equity dilution, ROIC/ROE, share buybacks and equity vs. alternative financing mix? In tech, the playbook for growth has been for decades to just raise equity throughout one's lifecycle. The result is additional dilution to founders, investors and employees. Oracle raised $32MM in their IPO and has not raised any equity since. They have raised over $150Bn+ in total debt for M&A, buybacks and general corporate uses, and repurchased ~$130Bn of their shares, offsetting equity dilution to increase equity returns.    At General Catalyst, we view the Oracle story as illustrative of a different approach tech companies can now take to build their customer base and fuel their growth. For decades, tech companies have relied solely on equity financing to fund their growth. The lack of alternative growth financing led us to create the Customer Value Fund (CVF), where we help companies of all sizes fund their sales & marketing, R&D investments and share buybacks, while minimizing shareholder dilution and risk. CVF allows companies to stop using their own cash to fund sales and marketing and instead, unlock that cash for other corporate uses, making the company more capital efficient without adding any risk to the business. To learn more, see the comments. General Catalyst #CustomerValueFund #CapitalAllocation

473

29 Comments

The Nvidia–Groq deal is not about eliminating a competitor. It is an architectural admission. I broke down this specific bottleneck on CNBC (clip below). As AI systems scale, inference economics start to dominate. Inference already costs an order of magnitude more than training over a model’s lifetime, and that reality puts pressure on architectures built for generality. Nvidia has optimized GPUs for inference extremely well. But this move signals something deeper: large-scale inference may not converge on a single, general-purpose architecture. Retrofitting can only go so far. The CNBC segment is necessarily short. It does not allow room to walk through the unit economics, architectural trade-offs or what this means as systems move toward more agentic workloads. I have linked my full analysis in the comments for those who want the deeper dive. 👇🏾 #Nvidia #Groq #InferenceEconomics #AIInfrastructure #TechStrategy #CNBC

165

6 Comments

For the past few years, AI headlines have centered on training - bigger models, larger clusters, record-setting compute. But a structural shift underway is toward inference. As AI moves from experimentation to deployment at scale, inference will define responsiveness, energy efficiency, cost economics and real-world utility. This is no longer just a scaling problem. It’s an architectural one. The next chapter of AI will be written by those who can deliver intelligence efficiently, reliably, and sustainably at global scale.

168

22 Comments

Gartner's stock crashed 53% this year. Their federal business has halved. Yet Q2 revenue grew 5.7% to $1.7B, expanded EBITDA 6.6%. The market saw something else: → Contract Value contraction: 7.9% → 6.7% → 4.9% (Q2) → Federal business: 47% dollar retention → Decision-making: From functional leaders to CFOs and CEOs → Stock: –53% YTD CEO Eugene Hall blamed DOGE and policy uncertainty. Maintaining calls for $6.455B revenue and double-digit growth. The market sees dual threats converging: procurement freezes while AI replicates, fracturing relationship-based sales models beyond repair. Threat One: Procurement Freezes 47% dollar retention. DOGE reframed knowledge work as waste. Government buyers never switch vendors. When your "safest" customers cut you in half, every other client is watching. Hall revealed this spreading: "Purchase decisions...are now being escalated to the CFO or even the CEO." DOGE gave every CFO permission to question advisory spending. Threat Two: AI Replication But procurement freezes are just the beginning. Anyone can paste Gartner content into ChatGPT for instant summaries. AI generates research reports that feel equivalent to basic human analysis. Management points to 500,000 executive conversations as their edge. But when CFOs control spend, those calls get harder to secure - and easier to question when AI delivers something similar in seconds. The Self-Disruption To counter this they launched AskGartner: instant AI answers from their research archive. Hall calls it "augmentation, not replacement." The market sees it differently. If clients can get instant Gartner insights through AI, why pay $25,000 for analyst access? Why Defense Is Impossible Gartner operates with governance built for gradual change. When your core metric collapses 50% in six months while your product becomes replicable, committee decisions become a luxury you can't afford. And it gets worse. Top analysts are realizing they can leave, use the same AI tools, and compete independently. The Escape Route The only escape requires destroying what made them profitable. Embedded workflows could work - becoming infrastructure, not content. But this demands dismantling the analyst pyramid while the current model still prints money. The very changes that could save them require cannibalizing their high-margin business model. The Reckoning The debate isn't whether Gartner's human expertise matters - it's whether their business model can capture that value when procurement freezes and AI replicates. DOGE proved "essential" knowledge work can be cut overnight. AI is proving knowledge work can be replicated instantly. For every knowledge business: your most stable revenue streams just became your most vulnerable. The enterprise sales model that built billions is breaking in real time. The question is whether they'll cannibalize themselves before the market does it for them. (Summary - full analysis sent to subscribers)

181

64 Comments

On-Prem is the New Cloud: Why Enterprises are Reversing Stance in the Age of AI Remember when cloud was the battleground for innovation, agility, and scale? Everyone was tracking the % of on-prem compute moving to the cloud and it was growing every year? Well now there’s a reversal. The reality today? On-prem is the new cloud. Here’s why. AI’s Dark Side: Legal Risk and Data Governance Headaches Enterprises are excited about AI, but also uneasy as vendors like OpenAI face intensifying legal pressure. In a high-stakes lawsuit, a judge ordered OpenAI to preserve all ChatGPT chats (even deleted ones), overriding its 30-day deletion policy. Imagine that: you delete sensitive conversations, but they’re still sucked into legal limbo. The precedent is clear: AI prompts and outputs are discoverable records, demanding integration into enterprise ESI (Electronic Stored Information) policies. For large organizations handling sensitive data, the risk is simple: you can’t afford your AI tools to turn into legal liabilities. Especially when global regulations (GDPR, etc.) and internal governance collide with unpredictably broad discovery orders. Sovereignty, Control, Trust Many enterprises still ban ChatGPT or impose strict rules on usage. Even when allowed, token limits are throttled. Enterprises are cautiously letting AI in (as they should). Enterprises need to have control and trust. SAP is leaning into this with its Sovereign Cloud On-Site, deployable on customer premises with SAP-managed infrastructure. It is built to deliver data, operational, technical, and legal sovereignty in one package. Doubt it will be the last. Scaling Pains Talking to founders, three reasons stand out why cloud isn’t scaling with AI: 1/ The rate at which AWS/GCP lets you provision compute and the compute it wants you to provision both falter as AI and AI driven developers deploy a lot more code more frequently than before. So sometimes it’s not about data governance, but actually about performance and functionality. 2/ Shadow AI is rampant—employees paste sensitive info into models despite restrictions—creating demand for better permissioning tools. 3/ Data leakage remains a live risk, like the August 2025 Grok incident where hundreds of thousands of private chats became publicly accessible and indexed. One thing is for sure - founders selling into the enterprise are scaling faster if they offer the ability to run on-prem. The Future is Hybrid Don’t get me wrong - cloud is not going anywhere. But neither is on prem. A few years agoI thought we were headed for a full cloud shift, but now I think the future is truly hybrid.

34

4 Comments