Factory Joins a $5B+ AI Coding Club Now Worth $130B+ — The Real Money Is in the Picks and Shovels

Factory's valuation has compounded roughly 17x in12 months — from $300M at the September 2025 Series B to $5B in this week's round. The company told investors in April that revenue had doubled month-over-month for six consecutive months and that its Droids were deployed at hundreds of thousands of developers across NVIDIA, Adobe, Bayer, EY, Palo Alto Networks, Adyen and MongoDB. The Sept. 15 round was structured at the same $5B post-money but on a $200M primary check, meaning the marginal price is set by the new money rather than the April mark. With ARR estimated in the $60M range, the implied ~83x multiple is the cohort's richest — but it tracks the same 'early-stage, hyper-growth' profile Cursor carried at its $9.9B Series C in mid-2025.

Every AI coding agent in the cohort runs on roughly the same stack: a foundation model (often an OpenAI, Anthropic, or Poolside coding-tuned LLM), fine-tuning and routing over private codebases, and tens of thousands of NVIDIA GPUs sitting in someone else's data center. The cohort's revenue growth is bottlenecked by GPU availability long before it is bottlenecked by product. NVIDIA's own numbers show what that demand looks like at the source: Data Center revenue accelerated from $51.2B in Q3 FY26 to $89.0B in Q2 FY27, a 74% jump in nine months — and that is the gross figure before any of it gets attributed to a specific workload.

The acceleration in NVIDIA's data-center line tracks the same calendar window as the cohort's valuations: Q3 FY26 (Oct 2025) overlaps Cursor's $29.3B Series D; Q4 FY26 (Jan 2026) overlaps Replit's $9B Series D; Q1 FY27 (Apr 2026) overlaps Cognition's $26B Series D and Factory's $1.5B Series C; Q2 FY27 (Jul 2026) overlaps Lovable's $13.3B Series C and the Cursor/SpaceX deal closing. Causality runs in both directions — coding-agent adoption pulls GPU orders, and NVIDIA's capacity expansion makes the cohort's product economics viable at all.

The cohort's revenue passes through three publicly investable layers. First, the GPU layer — NVIDIA sells the silicon and the networking, with Q2 FY27 gross margin of 75.0% on $96.2B of revenue, confirming pricing power has not cracked. Second, the cloud layer — CoreWeave repackages those GPUs into inference and training capacity, posting $7.59B in trailing revenue (TTM through Jun 30, 2026), up 112.5% year-over-year, with capex running at 271% of revenue to lock in supply. Third, the IDE and platform layer — Microsoft's GitHub Copilot sits at 4.7M paid seats and drove Intelligent Cloud revenue up 32% in FY26 Q4 (quarter ended Jun 30, 2026).

The pattern is stark. The publicly traded picks-and-shovels — NVIDIA, Microsoft, CoreWeave, Meta, Alphabet — trade in a 6–17x revenue band. The private coding cohort trades at 17x to 80x on much smaller bases. That gap is the entire bet: either the cohort grows into its multiple (Replit at $9B needs roughly $2B of ARR to justify a 5x; Cognition at $40B needs $10B+) or the multiple compresses in the next down round. The cohort's defensibility is real — agentic coding is now embedded in the workflow at Citi, Mercedes-Benz, Goldman Sachs, the U.S. Army, and the U.S. Navy via Cognition — but the path from $900M to $10B of ARR is not yet visible in any disclosed pipeline.

Short-term (days to quarters), the catalysts are concrete. Cognition's $40B round is the next print: a confirmed close would reset the cohort's high-water mark and pressure every other name to defend its multiple. Microsoft's FY27 Q1 earnings, expected in late October 2026, will be the first read on whether GitHub Copilot's June 2026 move to usage-based pricing lifted or compressed segment margins after the FY26 Q4 disclosure flagged it as a drag. CoreWeave's Q3 2026 results will show whether the Meta $21B commitment is starting to convert into recognized revenue or sits in backlog. Watch the NVIDIA Q3 FY27 earnings in late November for any disclosure on coding-specific workloads — historically NVIDIA does not break this out, but management commentary on agentic inference has expanded.

Over a one-to-three-year horizon, the milestones to watch are: (1) Cognition closing (or failing to close) at $40B and what that signals about secondary-market discipline; (2) Replit hitting its publicly stated $1B ARR target by year-end 2026, which would compress its multiple to roughly 9x; (3) the first agent-coding IPO, which is more likely to come from Replit or Cognition than Factory given scale; and (4) whether NVIDIA maintains its 75%+ gross margin as Blackwell pricing normalizes — that single line item determines whether the picks-and-shovels trade remains attractive as the cohort matures.