"The Singularity doesn't arrive, it compounds" — Alex Wissner-Gross

Why this is in the vault

Innermost Loop is on the active reading list (parent investing thesis already filed at [[2026-05-12-innermost-loop-ai-infrastructure]]). This issue's opening frame — that the singularity is a compounding curve, not a discrete event — directly extends the compounding-intelligence cluster we've been assembling all week. Wissner-Gross marshals concrete capability data (4.5-month doubling time per UK AISI, Mythos clearing previously unbroken cyber ranges) as evidence for the compounding-not-arriving thesis, which is the same epistemic move our cluster is built on.

The core argument

The singularity is being mistaken for a future event when it is already underway as a compounding rate. Wissner-Gross stacks evidence across layers: (1) capability — UK AISI measures doubling time compressed to 4.5 months, Mythos Preview is first to clear both AISI cyber ranges, GPT-5.6 in internal testing; (2) methods — Nous Research's Token Superposition Training claims "2-3x wall-clock pretraining speedup" at matched FLOPs; (3) talent — Recursive Superintelligence emerges with $650M at $4.65B betting on AI-conducts-experiments-on-itself; (4) product — Anthropic's Claude for Small Business toggles into QuickBooks/PayPal/HubSpot/Canva, signaling enterprise as-you-go pricing replacing flat consumer plans; (5) substrate — compute is now currency (Huang foundation donating $108.3M of CoreWeave hours; Altman mulling a Stargate-redux compute co); (6) embodiment — Unitree's UniStore is a robot app store, Figure live-streamed 8-hour humanoid shifts, Tokyo opened a fully automated medicine lab; (7) frontier — Varda's Asparouhov predicts "195 of the next 200 products manufactured in space" will be pharma; (8) macro — Nvidia first to $5.5T cap, Anthropic overtakes OpenAI inside Ramp's customer base (34.4% vs 32.3%), 25% of DC lobbyists now work AI issues. The closing line ("Speak softly and carry a big GPU") reframes Teddy Roosevelt's doctrine for the compute-as-statecraft era.

Mapping against Ray Data Co

STRONG mapping to the compounding-intelligence cluster. The "compounds, doesn't arrive" frame is the same load-bearing premise underneath:

• [[2026-05-14-four-tier-buy-build-stack-soloproneur-tam-filter]] — the 4-tier BUY/BUILD model only makes sense if capability keeps compounding under the soloproneur; Wissner-Gross's 4.5-month doubling is the rate that makes the tier shifts inevitable rather than speculative.
• [[2026-05-13-dorsey-from-hierarchy-to-intelligence-block-mini-agi]] — Dorsey's hierarchy-to-intelligence reorg is what an org looks like when it accepts compounding rather than arrival.
• [[2026-05-13-amble-is-software-losing-its-head-defensibility-migration]] — defensibility migrates because the capability layer never stops moving. Wissner-Gross's evidence stack is the empirical underbelly of Amble's argument.
• [[2026-05-13-compounding-intelligence-parked]] — the parked Sanity Check angle. This issue strengthens it: rather than restating the thesis, the next iteration of the parked angle can quote the 4.5-month doubling data as the load-bearing number.
• [[2026-05-12-innermost-loop-ai-infrastructure]] — parent thesis filing; this issue is direct continuation of the same author's thinking.

Tactical reuse: the data points (4.5-month doubling, 25% of lobbyists, Ramp share flip) are quotable in any RDCO/Sanity Check piece that needs concrete numbers underneath the compounding frame.

Related

• [[2026-05-14-four-tier-buy-build-stack-soloproneur-tam-filter]]
• [[2026-05-13-dorsey-from-hierarchy-to-intelligence-block-mini-agi]]
• [[2026-05-13-amble-is-software-losing-its-head-defensibility-migration]]
• [[2026-05-13-compounding-intelligence-parked]]
• [[2026-05-12-innermost-loop-ai-infrastructure]]