Field guide · evidence map
Robotics evidence is stronger than a demo-only cycle, but the sector is still S3/S4-heavy. S5 requires a chain: paid deployment, repeatability, runtime/intervention, customer ROI, production economics and financial materiality. This is a public evidence map, not a winner ranking or investment recommendation.
Updated 2026-06-15. Evidence map only; no trade recommendation.
NVIDIA Isaac GR00T, Physical Intelligence, Skild AI, Figure Helix and Tesla Digital Optimus make robot AI a trackable infrastructure layer. S5 still needs paid model/software contracts, deployed robot counts, uptime, intervention metrics and revenue economics.
Unitree R1 AIR from US$4,900, G1 from US$13.5K and H2 at US$29,900 lower experiment cost, but do not prove shipments, industrial reliability, gross margin, support cost or customer ROI.
Figure BMW shows task-count and runtime evidence. The S5 gap is repeat orders, customer-confirmed ROI/payback, uptime, intervention and margin.
Tesla capacity-intent language, Figure BotQ production metrics and Leaderdrive supplier filings are meaningful, but each still needs actual output, yield, robot-specific economics or customer confirmation before S5.
The upgrade chain is paid deployment, repeatability, runtime/intervention, customer ROI, production economics and financial materiality. A single demo, logo, price point or capacity headline is not enough.
Each layer now has better public anchors than old-cycle demo videos, but each still needs its own proof-quality bridge. Designed capacity, customer-site KPI, low-cost bodies, filing-backed supplier revenue and model releases should not be collapsed into S5 economics.
Current anchor: NVIDIA Isaac GR00T plus Physical Intelligence π0 / openpi / π0.5 / π0.7 and Skild AI official posts make robot AI a trackable infrastructure layer.
What it proves: Robot AI is becoming observable through model, tooling, simulation and data-stack releases rather than only isolated demos.
What it does not prove: Software revenue, attach rate, customer ROI, autonomy economics, gross margin or paid multi-OEM adoption.
S5 upgrade evidence: Paid model/software contracts, deployed robot counts using the stack, uptime/intervention metrics, ARR or licensing economics and gross margin.
Current anchor: Unitree official pages list R1 AIR from US$4,900, R1 from US$5,900, G1 from US$13.5K and H2 at US$29,900.
What it proves: Cheaper humanoid bodies can lower developer, education, research and pilot experimentation cost.
What it does not prove: Industrial reliability, gross margin, customer mix, support cost, shipment scale or customer payback.
S5 upgrade evidence: Product-line shipments, customer mix, uptime, intervention rate, repeat orders, gross margin, support cost and customer ROI/payback.
Current anchor: Tesla's Q1 2025-Q1 2026 Form 8-K Exhibit 99.1 sequence moves from pilot-line language to Fremont 1M/year and Texas long-term 10M/year designed capacity.
What it proves: Tesla has disclosed increasingly concrete Optimus manufacturing infrastructure and designed-capacity ambition.
What it does not prove: Actual production rate, accepted units, yield, runtime, external paid customers, robot revenue or unit economics.
S5 upgrade evidence: Actual Optimus output, production yield, internal deployment count, uptime/intervention, internal factory ROI, external paid customers, pricing, revenue and gross margin.
Current anchor: Figure reports BMW deployment KPI of 90,000+ parts and 1,250+ runtime hours, plus BotQ manufacturing KPI of 350+ Figure 03 robots delivered and >80% EOL first-pass yield.
What it proves: Figure has one of the clearest public deployment/manufacturing KPI curves among private humanoid OEMs.
What it does not prove: Contract value, customer-confirmed ROI/payback, robot count by customer site, repeat-order economics, intervention rate, gross margin or economically utilized fleet.
S5 upgrade evidence: Customer-confirmed ROI/payback, contract value, repeat order, deployed robots by site, uptime, intervention rate, service cost, production yield over time and gross margin.
Current anchor: UBTECH's 2025 annual report disclosed RMB 820.557m full-size embodied humanoid revenue, 1,079 units sold and more than 6,000 annualized full-size humanoid capacity at end-2025.
What it proves: A listed company has disclosed humanoid revenue, unit sales and capacity in filings.
What it does not prove: Humanoid-specific gross margin, customer ROI, uptime/intervention, repeat-order economics, warranty/service cost or cash-flow durability.
S5 upgrade evidence: Humanoid-segment margin, named utilization data, repeat orders, positive unit economics, service/warranty cost, cash-flow path and customer ROI/payback.
Current anchor: Leaderdrive's 2025 annual report disclosed RMB 570.714m revenue, RMB 422.528m industrial / embodied-intelligence robot component revenue and 425,158 harmonic reducers sold; its 2026-06-09 abnormal-trading announcement also warned there was no undisclosed major matter behind a +36.00% share-price move.
What it proves: Robot-relevant revenue/category growth and shipment growth are real filing-backed supplier signals.
What it does not prove: Named humanoid OEM exposure, humanoid-only revenue, single-robot value content, contract value, customer ROI or S5 humanoid commercialization.
S5 upgrade evidence: Named OEM relationship, humanoid-specific revenue/order value, margin durability, single-robot value content, repeat orders and capacity tied to humanoid demand.
The minimum bar is not a single headline. A stronger S5 candidate should show several of these signals in the same company, customer program or value-chain layer.
A real S5 candidate should disclose contract value, pricing, paid scope or enough customer-confirmed economics to connect deployment to value capture.
A first logo can be S4; repeat orders and multi-site expansion are the bridge toward repeatable commercialization.
Runtime hours are useful, but S5 requires the quality of that runtime: autonomy level, intervention rate, failures, safety events and task success in customer environments.
The decisive proof is not robot motion; it is whether the customer can justify labor displacement, throughput, quality or safety gains after integration and service costs.
Capacity, cadence and first-pass yield matter only when paired with accepted units, cost curve, warranty/service burden, revenue and gross margin.
If quantified deployment, production or filing data disappears for multiple quarters, keep the evidence at S3/S4 instead of upgrading on narrative momentum.
One pilot can be a strong signal, but no repeat order, no multi-site expansion and no customer economics is a downgrade against the S5 path.
Revenue growth can still be weak evidence if receivables, cash burn, warranty/service burden or gross margin deteriorate materially.
Evidence map only. No winner ranking. No trade recommendation. S4-heavy evidence does not equal S5 scaled-commercial economics.