brandon suede

Application layer companies lack moats because frontier models (Claude, GPT) will natively absorb vertical capabilities (legal, medical, finance); infrastructure companies (data, compute, evals) build real moats via network effects, data compounding, and R&D cycles, enabling sustainable high margins.

Model performance improvements drive 10x YoY capability gains, causing total token consumption to explode (Jevons paradox); Mercor already spends more on agent tokens than salaries; average enterprise will spend more on compute than headcount in 5 years as ROI on model inference compounds exponentially vs linear human productivity.

Attackers now use swarms of coding agents to exhaustively probe codebases at machine speed, making traditional defenses obsolete; this drives enormous boom in AI security engineering tools and defensive capabilities, with frontier labs actively seeking best AI security engineers.

Pre-sales GTM moats erode as models replicate SaaS products; durable defensibility comes from post-sales forward-deployed motion where agents are trained on organization-specific tacit knowledge (Slack, messages, workflows) to go the last mile; this is why labs invest heavily in forward-deployed teams and Sequoia's 'services are the new software' thesis resonates.

Enterprises will build eval systems of record for each workflow, benchmarking every new model to hot-swap and distill; majority of inference in 5 years shifts to open-source/distilled models, not frontier APIs; stickiness only exists in workflow layers (Claude Code routines, ChatGPT customizations), not pure API consumption.

Academic benchmarks (GPQA, IMO) are disconnected from enterprise outcomes; companies must build evals for real workflows (multi-week financial modeling, end-to-end SaaS cloning) to distill models, achieve 10x price-performance via open-source alternatives, and commoditize the model layer.

Every major lab builds custom silicon (TPU, MTIA, Trainium, etc.); Cerebras executing well; in 5 years Nvidia holds 30-40% share but in a market orders of magnitude larger, still making it the world's most valuable company; investment thesis shifts from monopoly to dominant share in hypergrowth market.

Extreme demand from frontier labs (Meta's superintelligence group offering $20M/yr) creates 10:1 demand-supply ratio; compensation will escalate for elite researchers but broaden as more people acquire frontier training skills, gradually normalizing the 99th percentile market.