Technology releaseNovember 20, 2026Редакция 105.kz

Boston Dynamics to test the Atlas humanoid at a GlobalFoundries fab in 2027

GlobalFoundries will be the first non-automotive company to test the Atlas humanoid, starting in early 2027. What from this news applies to enterprises in Kazakhstan — and what is still a test.

Oil & GasEnergyUranium & MiningManufacturingPredictive MaintenanceComputer Vision for HSE

Boston Dynamics to test the Atlas humanoid at a GlobalFoundries fab in 2027

Key takeaway. GlobalFoundries will begin testing the Atlas humanoid in early 2027 — the first public testing outside the automotive industry. For enterprises in Kazakhstan, what transfers is not the humanoid but the proven loop: inspection routes, instrument data, an equipment condition model and a work order in the maintenance system. The decision to run such a pilot rests on a measurable baseline, not on news about a robot.

What happened

GlobalFoundries will become the first Boston Dynamics customer outside the automotive industry to begin testing the Atlas humanoid. The Robot Report reported this on 20 November 2026. Testing is due to start in early 2027: in the companies' wording, they will first identify potential applications for Atlas in semiconductor manufacturing, assess integration requirements and the customer's readiness, and then begin training the robot on specific operations. How many robots will be involved, at which sites and which tasks will come first is not disclosed.

The relationship did not start with a humanoid: GlobalFoundries has used the quadruped Spot for industrial inspections for several years, and the companies are now expanding the programme — additional Spot units will appear at other sites in the network. Since April 2026, two Spot robots at the Singapore site have covered almost 900 km in total and performed more than 90,000 inspections in sub-fab areas (technical floors beneath cleanrooms), with a 99% mission success rate. These are the companies' own figures; the counting methodology is not disclosed. GlobalFoundries' director of manufacturing Pradeep Singh said that Spot helped the company move from reacting to failures to predictive maintenance: the robots collect data at thousands of control points and prevent downtime.

For context: Boston Dynamics showed the product version of the electric Atlas at CES 2026 and said the first production run of the year is fully allocated — to Hyundai and Google DeepMind. Atlas is currently being trained to feed parts and move materials at the RMAC facility in Georgia; Hyundai expects to reach component assembly by 2030, states plans to deploy 25,000 Atlas robots and to build a plant in the USA with a capacity of 30,000 robots per year. In October the company showed a redesigned hand — four fingers and 13 degrees of freedom instead of seven on the three-fingered version — and an experimental wheeled version as a testbed for control development.

What this means for enterprises in Kazakhstan

Substantively, the news is not about humanoids at Kazakhstani sites. A semiconductor fab is a controlled environment: flat floors, stable climate, marked routes, dense IT infrastructure and many identical control points. What transfers here is not the robot itself but the loop: inspection route → instrument data → equipment condition model → work order in the maintenance system (ТОиР). That is exactly what already runs on Spot in Singapore, and it is closer to the tasks of Kazakhstani enterprises than a humanoid.

Oil and gas. Rounds over pump units, compressors, heat exchangers and flare system components are a classic object for mobile inspection: vibration, bearing temperature, thermal images of flanges and valves, leak acoustics. The data sits alongside process control system (АСУ ТП) telemetry and maintenance history. The constraint is explosion protection and hazardous gas zones: equipment must comply with TR CU 012 requirements and be admitted under a work permit.

Energy. Turbine and boiler shops, open switchyards, distribution feeder lines. Value comes from thermal monitoring of contact joints and partial discharge monitoring, and a link to SCADA allows routes to be built around telemetry deviations rather than around a schedule.

Uranium and mining. The crushing and grinding circuit, conveyor lines, heap leaching areas: dust, vibration, high running hours on bearing assemblies. The key task is estimating remaining useful life (RUL) from a combination of data rather than from a single threshold. In open-pit and underground operations, add the lack of stable communications and industrial safety requirements.

Plants. Feeding parts and moving materials are exactly the classes of tasks Atlas is being trained on at Hyundai. But for a shop floor today, an autonomous vehicle or a manipulator for a specific operation is more practical than a humanoid.

The constraints are common: on-premise as the norm (video and data do not leave the perimeter), personal data localisation, information security requirements of national companies, Russian and Kazakh in the interfaces, integration with legacy process control systems through standard gateways, and autonomous operation when communications are lost.

Condition at the GlobalFoundries fab Typical condition at a site in Kazakhstan
Controlled environment, flat floors Dust, moisture, frost, uneven surfaces
Standardised routes Routes change, some zones are hazardous gas areas
Dense IT infrastructure Legacy process control systems, local gateways

How to apply: an 8–12 week pilot

  1. Select one area and one route. The criterion is a measurable cost of downtime or the regularity of failures. Not robotisation, but a specific processing stage.
  2. Audit the data. What already exists: process control system and SCADA tags, maintenance logs, running hours, failure history, inspection records, thermal image archives. What is missing — as a separate list.
  3. Fix the baseline and acceptance criteria. Technical availability factor (КТГ), number of failures per asset, hours of manual rounds, share of missed deviations. Without a baseline, the effect cannot be measured.
  4. Check infrastructure constraints. Explosion protection, access, communications, data storage inside the perimeter, interface languages.
  5. Run the pilot and measure the effect. Comparison with the baseline on historical failures and over the pilot period; the scale-up decision is made on pre-agreed metrics.

More on the classes of tasks — predictive maintenance and computer vision for HSE; industry specifics are in the oil and gas section.

Limitations and risks

The source describes a testing agreement, not a deployment: the number of robots, the sites and the first operations are not disclosed, and the start date is early 2027. The Spot figures (900 km, 90,000 inspections, 99% mission success) are the companies' claims without a disclosed methodology: it is unclear what counts as an inspection and how failed missions are accounted for. Distance travelled is not in itself an indicator of value.

Hyundai's plans — 25,000 Atlas robots and a plant for 30,000 robots per year — are stated intentions, not completed deliveries. Testing a humanoid in semiconductor manufacturing does not prove its suitability for a refinery or a mine: different environment, explosion protection, temperature, dust and occupational safety requirements. There is no public data on the use of humanoids at enterprises in Kazakhstan (estimate); the status of admitting such machines to work and occupational safety questions remain undefined.

Finally, there are few facts in the source: it provides neither a list of Atlas tasks nor measurement results, so conclusions about the maturity of the technology outside the automotive and semiconductor industries remain an assumption (estimate).

Sources

  1. Boston Dynamics Atlas humanoid lands first non-automotive trial · 2026-11-20
    Since April 2026, two Spot robots at its Singapore facility have collectively traveled nearly 900 km (559 mi.) and completed more than 90,000 inspections in sub-fab areas, achieving a 99% mission success rate, according to the companies.
  2. Фото: Uploaded by Duk 08:45, 16 Feb 2005 (UTC) — Clean room.jpg (Public domain, Wikimedia Commons)

This material was prepared with AI from the sources listed above. Spotted an inaccuracy? Write to info@105.kz.

FAQ

Frequently asked questions

Short, straight answers to the questions we hear most often.

Should we expect humanoids at a Kazakhstani refinery or mine?

Within the pilot horizon — no. The source describes testing Atlas at a semiconductor fab from early 2027, and the list of tasks, the number of robots and the sites are not disclosed. For a refinery or a mine, explosion protection, autonomy and admission to work are additionally required.

What from this news is actually transferable today?

The loop: inspection routes → instrument data → equipment condition model → work order in the maintenance system. The data are vibration, bearing temperature, thermal images and leak acoustics, plus process control system telemetry and maintenance history.

How do we find out in 8–12 weeks whether there is an effect?

Fix the baseline before the start: technical availability factor (КТГ), number of failures per asset, hours of manual rounds, share of missed deviations. Then compare with the pilot results on the same area and on historical failures. Without a baseline the comparison will be subjective.

Where will the video and inspection data be stored?

Inside the enterprise perimeter, on-premise: for national companies this is the norm under information security and personal data localisation requirements. Interfaces and reporting must support Russian and Kazakh.

Next step

Discuss a pilot project

Tell us about your task — we'll come back with a data audit plan and an effect estimate within two business days.

  1. 1A 30-minute call: the task, your current systems, who makes the decision
  2. 2A review of data and processes, a staged plan with time and cost ranges
  3. 3A 2–4 week prototype or an 8–12 week pilot with success criteria agreed upfront

NDA from the first contact

Reply within one business day