· The Capital Lens
Most of it ignores the workforce.
Capital is voting. Listen to it.
Frontier labs have crossed the trillion dollar combined valuation line. The capex commitments behind them dwarf the entire workforce tech category by an order of magnitude.
Frontier labs, May 2026.
OpenAI, Anthropic, xAI, Google DeepMind, Mistral combined paper valuation. Up 8x since Q1 2023. The strategic premium is no longer theoretical.
US AI capex, 2026.
Committed or contracted spend on compute, energy, and infrastructure. Larger than the entire global enterprise SaaS revenue base.
The talent slice.
Of the $1.6T projected 2027 capex, only 14 percent touches workforce design. The underallocation is the bottleneck no balance sheet shows.
Four frontier labs, four years.
From combined $40B in 2022 to combined $1.5T+ in 2026. The trajectory is steeper than any prior tech cycle, including the dot com era.
Frontier lab valuations, quarterly
The mega deals shaping 2026.
Six transactions that reset the cost of capability. Each rewrites either the compute curve, the energy curve, or the build vs buy calculus.
| Q | Counterparty | Asset | Size | What it sets |
|---|---|---|---|---|
| Q4 24 | Microsoft · OpenAI | Compute commit (10 year) | $80B | Sets the compute floor for frontier training. |
| Q1 25 | Anthropic · AWS / Google | Compute and equity | $40B+ | Locks Anthropic to two hyperscalers in parallel. |
| Q3 25 | Tesla · TerraFab | Compute manufacturing | $119B | Vertical integration of compute supply. |
| Q4 25 | Anthropic · SpaceX | Inference platform deal | undisclosed | Frontier model meets sovereign comms. |
| Q1 26 | G42 · OpenAI · Oracle | Stargate UAE cluster | $72B | First trillion watt cluster outside the US. |
| Q2 26 | Meta · Scale AI | Data labelling acquisition | $14.3B | Reprices proprietary data at strategic premium. |
Where the dollars actually go.
From 2022 to 2026, AI infrastructure jumped from 3 percent of enterprise IT capex to 42 percent. Workforce design barely moved off the floor.
The old allocation
The new allocation
Capital says yes. Workforce still says wait.
Across 250+ episodes, three findings recur: capital and compute are repricing fast, while workforce investment stays close to flat.
If you owned the enterprise, where would you spend?
The board question that exposes the 100:1 misallocation. Three numbers reveal the gap, and the fix.
We allocated $200M to AI compute this year. What did we allocate to redesigning the work the compute serves?
AI compute and platform
A mid cap enterprise's typical AI infrastructure spend in 2026.
If the line item exists at all
1 percent of compute spend. Often untracked. Usually buried in HR opex.
Workforce budget at compute scale
What a workforce line proportional to compute investment would imply. The gap the capital evidence makes visible.
What the capital evidence points to.
Three takeaways the money supports, drawn together from the deals, the capex shift, and the valuation curve.
Three patterns the numbers agree on.
Scale and pace of capital
Frontier valuations and committed capex have grown by roughly an order of magnitude since 2022. The build is among the largest infrastructure programmes on record, and the trajectory is steeper than prior tech cycles.
Workforce underinvestment
Workforce design held near 6 to 8 percent of enterprise IT capex while AI and compute climbed to 42 percent. The people who run the stack are funded out of proportion to the systems themselves.
Energy as the binding constraint
Cluster and power deals recur as the limiting factor on the podcast. Forecasts cited there treat energy availability, not capital or chips, as the constraint most likely to govern the pace from here.
The workforce data has not caught up.
The build is outpacing the workforce line.
AI and compute climbed to 42 percent of enterprise IT capex while workforce design held near 6 to 8 percent over the same period.
Cost per task is falling faster than headcount.
Inference costs cited on the podcast fell roughly threefold in twelve months, while skills half life shortened from 15 years to 2.5.
Energy reads as the binding constraint.
Cluster and power deals recur as the limiting factor. Forecasts cited on the podcast treat energy, not capital or chips, as the governing constraint.
One of five briefings distilled from the Moonshots record.
Drawn from over 250 episodes of Moonshots with Peter Diamandis and the Moonshot Mates. Figures reflect statements and reported transactions on record; forecasts are marked as forecasts.