TLDR Founders 2026-03-02

Serving inference on modern LLMs often fails to scale when GPU memory is constrained. Larger shared memory capacity lets you serve more users at lower latency, with less wasted compute. But true scalability doesn't require more memory; it requires a distributed memory that behaves like it's local.

Crusoe's inference engine powered by MemoryAlloy™ technology delivers exactly that. The result? Breakthrough speed, throughput, and reliability for the most demanding AI workloads. Work with our team to optimize performance for your own fine-tuned model so you can scale without compromise.

Explore Crusoe Managed Inference

95th percentile seed valuations hit $80.5M last year, nearly 3x higher than six years ago. 42% of seed rounds were over $5M. One company raised a $2B "seed." The word seed is now applied to everything from $500K SAFEs to $15M equity rounds, making it functionally useless. The real shift is that institutional capital is now arriving at true formation stage for the first time - not through accelerators or angels, but dedicated inception funds writing $500K-$3M checks pre-product, pre-team, pre-traction. Before raising, founders should ask every prospective VC one question: What does success look like for your fund? The answer will reveal whether their incentives actually align with your stage.

Data companies without strong research DNA or volume advantages are collapsing or getting acquired because shifting RL research directions reward either frontier research leadership or rapid data scale, not generic RLaaS. Enterprise demand is moving toward subjective, self-serve post-training infrastructure as regulated environments, bespoke workflows, and reliability thresholds block direct model adoption. The winners combine top-tier research with field deployment engineers and full-stack post-training tooling, as sparse high-quality data, unverifiable tasks, and last-mile implementation become the real bottlenecks.

A $100M+ ARR AI company is crushing it - customers love the product, pipeline is strong. However, every deal closes on one-year terms. Not three. Not two. One. The reason is structural, not emotional. SaaStr recently switched AI sales agents by copy-pasting its best prompt into a competitor's product. Migration was 50-80% done in minutes. That's the new reality for AI agent companies. When gross retention drops from 92% to 82% because of prompt portability, you need an extra $10M in annual bookings just to stay even. The moat isn't your AI anymore. It's integration depth, vertical expertise, and infrastructure nobody wants to rebuild.

The way software is shipping is changing. Instead of shipping code to solve a narrow task for all users, developers are shipping general-purpose agents that modify themselves to adapt to the environment and task. This makes the context the product. The value capture in an AI-powered software industry will come from the context and runtime layer, along with hardware and software co-design. Many companies will likely regret the investment in chips they are currently making.

Agents will destroy aggregation moats. However, true network effects, where products improve with more participants, will become more viable in more categories and modalities. More businesses will start building for network effects because there are now new surface areas, modalities, and GTM approaches. Software isn't dead, but the business is transforming into something much different from what it is today.

Products that aren't lovable aren't viable in the market. The Minimal Lovable Product is the earliest version of a product that's genuinely lovable. Minimal Viable Products (MVPs) devolved into 'minimal functioning products' because dev work was expensive. However, costs are now collapsing. Building an MVP now just means building another tool people tolerate until they re-create it or a better one comes along. Lovability may be the last defensible moat.

Redact advertised in TLDR and cut their CPC in half compared to LinkedIn. 6M+ tech professionals, 40%+ open rates, max 3 sponsors per issue. See how TLDR compares.

Turn any prompt into a shareable URL.

Vision-enabled AI that fixes frontend bugs in Chrome.

For years, the winning strategy in software was consolidation. Own the workflow end-to-end. Make it painful to leave. That model is cracking. Agents need open protocols and shared standards to do their job - they don't care about ecosystem boundaries. Users now expect to swap parts freely. The result is that Anthropic, OpenAI, Google, Microsoft, and AWS are all sitting at the same table contributing to shared agent protocols. Your moat is no longer lock-in. It's being the thing everything else connects to. If your platform is where value gets exchanged between tools, removing you breaks the whole workflow. The defensible position is a bridge, not a wall.

Everyone assumes AI will slash R&D budgets. The math doesn't work that way. Token costs are dramatically cheaper per unit of code produced, but Jevons Paradox kicks in hard. When something gets cheaper, you use more of it. Companies aren't cutting engineers. They're shipping more features faster and hiring more people to steer AI output. R&D as a percentage of revenue is going to stay flat or rise because every competitor gets access to the same productivity gains. The savings don't hit your bottom line. They get absorbed by the market.

Software stocks are getting crushed but almost no insiders are buying. Out of 75 public software companies, only a handful of executives have purchased stock in the open market over the past two years. Asana's CEO has bought over $1B of his own stock and lost billions in the process. The more interesting signal is at ServiceNow, where the CEO, CFO, CPO, and AI Officer all canceled their scheduled selling plans on the same day. The CEO then waited exactly six months from his last sale because of the short-swing profit rule before buying $3M worth. Sometimes the absence of selling tells you more than any purchase would.

The story of how Dragonfly grew to become one of the largest VC platforms in an industry that few have managed to survive in.

The market for software will be ten times bigger in ten years, and AI-native startups will capture much of that growth.

Artificial intelligence, like past technology booms, is driving a new wave of speculative investment that could establish enduring infrastructure despite being viewed as a bubble.

David Teten has shifted his content strategy, focusing on proprietary, experiential, and reputationally costly insights instead of AI-generateable content.