PM Portfolio Manager Reports

From Portfolio Manager reports

Training Data, Simulation, Teleoperation, And VLA Control

This node covers the software loop that records robot sensor, action, failure, recovery, and teleoperation data; tests policies in simulation; and uses vision-language-action control models to turn instructions and perception into robot motion. The investable question is whether that loop lowers human intervention, raises task success, improves the operator-to-robot ratio, and converts into compute revenue, robot software or services revenue, paid automation deployments, lower support cost, or better gross margin. The current basket is proxy-only: NVDA and TSLA have the broadest source-backed stack exposure, SERV has the clearest public fleet-teleop evidence, TER has paid robotics revenue, and AMZN has large internal robotics data but weaker separable revenue disclosure.

Current Setup

Software readiness gate The test is lower intervention, higher task success, and economics that show up outside demos.

Training data means synchronized video, tactile, proprioception, force, action, failure, recovery, and operator logs. Simulation tests the policy before deployment. Teleoperation records human rescue. VLA models connect perception and language to robot actions.

Toolchain anchor NVDA and TSLA have the broadest filed path into the robot learning stack. NVDA supplies compute, Isaac, Omniverse, GR00T, synthetic data, and deployment tooling. TSLA owns the Optimus, FSD, factory, and AI-infrastructure feedback loop.
Fleet evidence SERV, TER, and AMZN route the page back to operating KPIs. SERV discloses delivery-robot fleet metrics. TER reports Universal Robots and MiR revenue. AMZN has large warehouse robotics and AWS exposure, but robot economics are internal.
Attribution limit No ranked name gets direct humanoid software revenue credit. Most model logs, intervention rates, task success data, and control-stack economics remain private or embedded inside larger businesses.
  1. Experience logsRecord actions, failures, recovery, and operator help
  2. SimulationTest and scale policy training before physical deployment
  3. Policy trainingUse VLA and control models to turn perception into motion
  4. DeploymentMeasure task success, intervention rate, and incidents
  5. EconomicsConvert into revenue, margin, support cost, or cash

Basket

The ranking is the tournament merge result from ten xhigh primary research agents plus one xhigh merge selector. It ranks node economics and source-backed software exposure first, operating evidence second, and chart setup third. The right rail chart reads the local report API when it is running; the HTML does not embed static OHLC payloads or chart levels.

NVDA NVIDIA Corporation

Best public tooling route for simulation, synthetic data, VLA and robot foundation-model workflows, accelerated training, and edge robotics compute.

Market cap$4.73T
Next earningsNot confirmed
Latest qtr revenue$81.615B

Role in stack

NVDA supplies the accelerated compute and software toolchain around Isaac, Omniverse, GR00T, synthetic data, policy training, simulation, and robot deployment workflows. The page treats it as the platform route and requires separate evidence before assigning robotics revenue.

Revenue mix

Q1 FY2027 revenue was $81.615B. Data Center revenue was $75.246B, and the knowledge lane also routes robotics-relevant edge and platform exposure. Isaac and GR00T revenue is not separately disclosed.

Proof burden

Official customer adoption, robotics developer usage, simulation and control-stack traction, and evidence that robotics adds to compute or software economics. A data-center-only result keeps this as a proxy.

TSLA Tesla, Inc.

Direct public humanoid and fleet-control route through Optimus, FSD, Robotaxi data, internal manufacturing, autonomy software, and owned AI infrastructure.

Market cap$1.48T
Next earningsJul 22, 2026
Latest qtr revenue$22.387B

Role in stack

TSLA owns the Optimus development path, factories where early tasks can be tested, a large autonomy data loop, and the AI infrastructure needed to train and deploy control policies.

Revenue mix

Q1 2026 revenue was $22.387B. The lane also records Q2 deliveries of 480,126, Q2 storage deployment of 13.5 GWh, and 1.28M FSD subscriptions in the Q1 update. Humanoid robot revenue is not separately disclosed.

Proof burden

Factory task counts, task success, intervention rate, unit cost, safety, internal productivity, and a disclosed path from internal robot use to external customer economics.

SERV Serve Robotics Inc.

Closest public fleet-teleoperation KPI route through active robots, daily supply hours, fleet services, software services, and exception handling.

Market cap$488M
Next earningsNot confirmed
Latest qtr revenue$2.984M

Role in stack

SERV is the direct public fleet comparator for teleoperation, remote assistance, delivery-robot learning, incidents, and service economics. Its delivery robots differ from humanoids, and its fleet KPIs are close to the node's operating proof.

Revenue mix

Q1 2026 revenue was $2.984M, with fleet services at $1.958M and software services at $1.026M. The lane records 812 daily active robots and 10,295 daily supply hours.

Proof burden

Deliveries per robot, operator-to-robot ratio, remote intervention decline, incidents, service margin, support cost, cash burn, and dilution control. Q1 2026 gross loss was $9.001M and operating cash use was $41.422M.

TER Teradyne, Inc.

Paid robotics exposure through Universal Robots and MiR, with collaborative and mobile robots that can benefit from better control, fleet learning, and automation software.

Market cap$57.8B
Next earningsNot confirmed
Latest qtr revenue$1.2825B

Role in stack

TER brings Universal Robots and MiR as paid robotics routes, plus test and automation exposure around factories. It is a useful comparator for robot revenue that already appears in filings.

Revenue mix

Q1 2026 revenue was $1.2825B. Semiconductor Test was $1.1108B, or about 86.6% of revenue, while Robotics was $91.3M and slightly loss-making at the pre-tax line.

Proof burden

UR and MiR orders, robotics margin, AI-control adoption, backlog, and separation from the semicap test cycle. The stock can move on semiconductor test before robotics economics change.

AMZN Amazon.com, Inc.

Operator and cloud validation route through warehouse robotics, fleet logging, fulfillment automation, AWS AI capacity, and internal productivity.

Market cap$2.61T
Next earningsNot confirmed
Latest qtr revenue$181.519B

Role in stack

AMZN has one of the largest internal robotics estates, warehouse process data, fulfillment automation, and AWS infrastructure exposure. The node uses it as an operator/cloud validation route.

Revenue mix

Q1 2026 consolidated sales were $181.519B. AWS revenue was $37.587B and AWS operating income was $14.161B. Robotics economics are mainly internal productivity and are not separately reported.

Proof burden

Fulfillment cost per unit, throughput, safety, robot deployment counts, AWS robotics customers, cash conversion, and clearer disclosure that automation is improving margins rather than only adding capex.

Market caps are computed from 2026-07-02 local closes times local weighted shares outstanding. Next earnings dates were left as not confirmed unless the filed lane supplied a concrete date. Chart setup labels in metadata use weekly bars derived from local daily_ohlc; the latest rows were 2026-07-02 while representative ingestion lineage lagged at 2026-05-08.

Source-only monitors: SIEGY is the best local simulation and digital-twin monitor but lacks local chart coverage. ROK and SYM fit brownfield deployment more directly. GOOGL, MSFT, and META can matter for AI models and cloud tooling, but the filed lanes do not isolate humanoid-control economics. PDYN and ISRG stay in the monitor list until deployment and revenue evidence fit this node better.

What Confirms Or Weakens

Area What confirms What weakens or invalidates Watch next
01Node thesisLearning loop
What confirms

Task success rises while manual intervention falls across real deployments, with repeatable failure recovery and more tasks handled per robot.

What weakens or invalidates

Demos remain scripted, task coverage stays narrow, or support labor remains high for each deployed robot.

Watch next
  • Intervention rate
  • Task count
  • Failure recovery
  • Operator leverage
02EconomicsRevenue and margin
What confirms

Compute, robotics software, robot services, or automation revenue converts into gross margin, lower support cost, free cash flow, or higher utilization.

What weakens or invalidates

Robotics stays an R&D cost center, broad revenue lines swamp the signal, or capex rises without visible operating benefit.

Watch next
  • Segment revenue
  • Gross margin
  • Support cost
  • Cash conversion
03TeleoperationHuman assist load
What confirms

SERV and other fleets show more robots per operator, lower exception handling per delivery, fewer incidents, and better service margin.

What weakens or invalidates

Support labor scales with fleet size, incident rates rise, or deliveries require persistent remote intervention.

Watch next
  • Deliveries per robot
  • Remote assists
  • Incidents
  • Service margin
04Simulation and VLA stackTools leave research
What confirms

NVIDIA, SIEGY, Tesla, or cloud and toolchain sources show production adoption, developer usage, and customer references for robot simulation or control workflows.

What weakens or invalidates

Model releases remain research-only, customer references are absent, or deployments fail to show operating improvement.

Watch next
  • Customer references
  • Developer adoption
  • Product revenue clues
  • Latency
05Customer deploymentPaid use expands
What confirms

Tesla factory use, TER robot orders, Amazon fulfillment gains, SERV fleet activity, or paid pilots expand and keep operating metrics intact.

What weakens or invalidates

Pilots fail to become paid deployments, internal tests do not scale, or broad business cycles explain the reported result before robot economics matter.

Watch next
  • Earnings calls
  • Filings
  • Fleet counts
  • Backlog
06Data freshness and chart setupRefresh trigger
What confirms

Local daily_ohlc stays current, weekly chart labels match the ranked basket order, and the local report API serves the selected-security panel.

What weakens or invalidates

The API remains unavailable, lineage stays stale, SERV's shorter trading history limits chart confidence, or a new market session changes the setup.

Watch next
  • API checks
  • Daily OHLC
  • Lineage lag
  • SERV history

Source Trail

Routes used for this node
Route Use it for Primary links
Paper priorsHumanoid software learning loop, teleoperation reliance, simulation, robot foundation models, data flywheel claims, first deployment tasks, and investment-timing constraints.Citrini humanoid primer; Roland Berger 2026 extraction; Roland Berger PDF
Official NVIDIA checkGR00T N1, Isaac, Omniverse, robot simulation, synthetic data, and VLA architecture context. This supports NVDA's toolchain rank without assigning undisclosed robotics revenue.NVIDIA Isaac GR00T N1 release; GR00T N1 arXiv paper
Knowledge posteriorSecurity setup, proof burden, segment evidence, revenue mix, operating KPIs, source gaps, and sector-level routing.NVDA; TSLA; SERV; TER; AMZN; Information Technology; Industrials
Source-only monitorsSimulation, industrial software, warehouse automation, cloud model, and medical robotics references that matter to the node but are not ranked chart cards.SIEGY; ROK; SYM; GOOGL; MSFT; META; PDYN; ISRG
Discovery and DuckDB methodsRead-only local coverage, computed market caps, weekly setup labels, and selected-security chart provenance. NVDA, TSLA, SERV, TER, and AMZN have daily_ohlc rows through 2026-07-02.node metadata; read-only DuckDB query on daily_ohlc, instruments, and ingestion_runs; market caps computed as 2026-07-02 close times local weighted shares outstanding.
API chart routeRight-rail chart rendering. Each rank button and metrics strip calls the selected-security panel through data-chart-ticker. Weekly charts are expected from daily_ohlc, grouped by calendar week./api/securities/TICKER/chart?frequency=weekly&window=3y&as_of=latest; three-year visible horizon; no embedded OHLC chart payloads; local API check on 2026-07-05 returned connection refused at 127.0.0.1:8765.
Ranking processThe final order was selected by ten xhigh primary subagents and one xhigh merge selector. The merge kept NVDA and TSLA as broad stack anchors, moved SERV above TER and AMZN for direct fleet-teleop KPIs, retained TER for paid robotics revenue, and kept AMZN as operator/cloud validation.Final order: NVDA, TSLA, SERV, TER, AMZN.
Data limitationsDiscovery supports price, volume, chart routing, and computed market-cap context. It does not prove humanoid control revenue. Papers are priors; filed lanes, official NVIDIA sources, and read-only discovery checks are the posterior.Refresh if a new market session, filing, earnings release, guidance update, robot deployment disclosure, model release, or local API/chart behavior changes.

Categories