GPU Solutions

Solutions · Industry

AI for manufacturing, engineering and R&D without taking your trade secrets out of Spain

GPU Solutions runs open AI models on its own NVIDIA B200 GPUs in a Tier III datacenter in Madrid, so manufacturers, engineering firms and R&D teams can use AI on drawings, manuals, source code and plant data without handing them to a foreign API. Data is processed and stored in Spain and is never used to train models.

In short
  • Drawings, code and know-how processed in Madrid, with no subprocessors outside the EU.
  • Isolated environments: MIG fraction, dedicated pod of 1 to 4 GPUs or a full 8×B200 node.
  • Platform certified to ISO/IEC 27001:2023 and ENS Medium category by EQA.
  • Full NVIDIA B200 GPU at €8.00/h; standard dedicated environment 72 h after signature.

Know-how is only a trade secret if you protect it as one

In an industrial company, valuable knowledge is spread across thousands of documents: maintenance manuals, customer specifications, quality records, drawings, machine code and years of production data. AI lets people query it in plain language, extract data from scanned forms, review code or spot patterns in sensor series. It is a real productivity gain for engineering, quality and maintenance, and for R&D teams that need powerful GPUs without building a datacenter of their own.

But that knowledge is usually your most protected asset. Spain's Trade Secrets Act (Law 1/2019) only protects information if its holder has taken reasonable steps to keep it secret; sending drawings or code to a third-party AI service outside the EU, with no contract limiting its use, weakens that position. On top of this, the EU Data Act, applicable since 12 September 2025, governs access to data from connected products with safeguards for trade secrets, and the Cyber Resilience Act requires security across the whole life cycle of the products with digital elements you make.

Running AI on Spanish infrastructure, isolated and under contract, is a reasonable measure you can document. Our cluster uses NVIDIA B200 GPUs with 180 GB of HBM3e memory each, InfiniBand NDR networking and Exascaler parallel storage: details on infrastructure and public rates on pricing. We track Data Act, CRA and NIS2 deadlines in the regulatory radar.

Use cases in manufacturing, engineering and R&D

Four uses that already work with open models on our infrastructure. You can try the models on GPU Flow with non-sensitive material and then bring the real case into a private environment.

Assistant over technical documentation and manuals

Maintenance technicians and engineers ask questions in plain language (what torque a part needs, what the customer specification says about tolerances, when a procedure last changed) and the assistant answers by citing the exact manual or document. Documents are indexed in your own storage space in Madrid and the model never learns from them: it only looks them up when someone asks, and every answer can be checked against the source.

Semantic search with sources (RAG) · Qwen3 14B or Gemma 4 26B · Exascaler storage with quotas

OCR of drawings, forms and quality records

Many quality records, inspection sheets, delivery notes and older drawings still live on paper or in scanned PDFs. A vision-language model extracts fields, tables and annotations and turns them into structured data you can load into your ERP or quality system. Because drawings are sensitive, processing happens on our GPUs in Spain, without going through external OCR services. For complex drawings it is worth validating first on a sample of your own.

Qwen3-VL 32B OCR · Nemotron Nano 12B VL · batch jobs on a MIG fraction or dedicated pod

Private coding assistant over proprietary source code

The firmware in your machines, your control software and your internal engineering tools are trade secrets. A coding assistant running in a dedicated pod can read the repository, suggest changes, write tests and help review vulnerabilities without the code leaving your environment. Sandboxes include SSH, Python, PyTorch, Jupyter and coding agents, with an API compatible with the OpenAI and Anthropic SDKs, so existing tools plug in with minimal changes.

GLM 5.3 or Qwen3.8 · sandboxes from €0.15/active hour · dedicated pod with custom image

Analysis of sensor and production data

Vibration, temperature, energy or part quality series can be analysed with models you train or fine-tune yourselves on your own data, on dedicated B200 GPUs with PyTorch and Jupyter. You can also pair that analysis with a language model that explains in plain words what is happening on a line or during a shift. Plant data stays in Spain and the resulting model is yours, to deploy wherever you choose.

PyTorch and Jupyter · pod of 1 to 4 B200 · bare-metal 8×B200 node for large training runs

How a drawing or a repository travels

From your company's systems to the model's answer, without leaving Spain at any point.

  1. 01

    Encrypted link from plant or office

    Data travels over TLS or over a dedicated encrypted point-to-point link, with post-quantum cryptography available for those who need it.

  2. 02

    Isolated environment in Madrid

    Each customer works in its own isolated MIG fraction, a dedicated pod with GPU passthrough, or a full 8×B200 node outside the shared cluster with its own access.

  3. 03

    Storage with quotas

    Drawings, manuals, repositories and datasets are kept in your Exascaler storage space, with quotas and permissions you define.

  4. 04

    Results and a clean exit

    You receive answers, extracted data or fine-tuned models. Nothing is used to train models and, if you ever leave, you take your data and weights with you in standard formats.

Regulatory framework and how we help

The rules that weigh on AI in manufacturing and R&D, and the part the infrastructure covers. Classifying your company and your products is for you and your advisers; we provide an environment you can document.

Regulatory framework and how we help
RuleWhat it requiresHow we help
Spanish Trade Secrets Act (Law 1/2019)Information is only protected as a trade secret if it has value because it is secret and its holder has taken reasonable steps to keep it secret.Processing in Spain in isolated environments, no use of data for training, DPA with optional zero retention and a dedicated encrypted link: measures you can document.
Data Act (Regulation (EU) 2023/2854)Since 12 September 2025 it governs access to and sharing of connected-product data, with safeguards for trade secrets, and makes switching between data processing providers easier.Open models, an API compatible with market standards and data in portable formats: you can change provider without rebuilding the integration.
Cyber Resilience Act (Regulation (EU) 2024/2847)Cybersecurity requirements for products with digital elements across their life cycle. Reporting of actively exploited vulnerabilities applies from 11 September 2026 and the remaining obligations from 11 December 2027.A private coding assistant to review firmware and product software without exposing the code, on a platform certified to ISO/IEC 27001:2023.
NIS2 (Directive (EU) 2022/2555)Risk management and supply chain security for essential and important entities, including manufacturers in certain sectors. In Spain the transposing law is still going through the legislative process.Platform security documentation (ISO/IEC 27001:2023, ENS Medium category) for your supplier assessment.
GDPRA legal basis, data minimisation and a processing agreement when AI handles personal data, for example staff data or operator data linked to machine data.Standard DPA, data in Spain, no subprocessors outside the EU and no model training on your data.

Regulatory radar: industry

Deadlines and updates from the Data Act, Cyber Resilience Act, NIS2 and AI Act that affect industry, with official sources. Full view in the regulatory radar.

  1. AI Act high-risk obligations for Annex III systems start to apply

    Covers credit scoring, life and health insurance pricing, justice, essential public services and employment. Requires risk management, data quality, logging, human oversight and technical documentation.

    Official source: EUR-Lex · Reglamento (UE) 2026/1744 ↗

    DeadlineEU

  2. EuroHPC AI Gigafactories call closes (up to 7 consortia)

    Awards are expected in early 2027. Spain is bidding with a public-private candidacy. It signals where European compute capacity for training and running models will be located.

    Official source: EuroHPC ↗

    DeadlineEU

  3. Data Act Art. 3(1): connected products must make their data accessible by design

    Applies to connected products placed on the market from this date. Manufacturers must design them so users can access the data they generate, which shapes architecture, contracts and industrial analytics.

    Official source: EUR-Lex · Reglamento (UE) 2023/2854 ↗

    ApplicableEU

  4. Cyber Resilience Act: manufacturers notify ENISA and CSIRT within 24 h of exploited vulnerabilities

    Manufacturers of products with digital elements must report actively exploited vulnerabilities and severe incidents. This includes software with AI components placed on the EU market.

    Official source: EUR-Lex · Reglamento (UE) 2024/2847 ↗

    ApplicableEU

  5. AI Omnibus Regulation (EU) 2026/1744 published: new timetable for high-risk AI

    In force since 27 July 2026. Annex III obligations move to 2 December 2027 and Annex I obligations (AI in regulated products such as medical devices) to 2 August 2028.

    Official source: EUR-Lex · Reglamento (UE) 2026/1744 ↗

    In forceEU

  6. Commission refers Spain to the CJEU for failing to transpose NIS2

    A lump sum and daily penalties are requested. Until the Spanish law arrives, essential and important entities are well advised to prepare risk management, incident reporting and supplier control now.

    Official source: Comisión Europea ↗

    OngoingEU

Recent laws, guidance and decisions →

Frequently asked questions

Does using an external AI on our drawings put trade secret protection at risk?

It can, because Spain's Law 1/2019 requires reasonable steps to keep information secret. If drawings are processed by a third-party service outside the EU with no contractual limits on use, that protection becomes harder to defend. On our platform they are processed in Madrid, in isolated environments, without training models and under a signed DPA.

Can we fine-tune or train models on our own data?

Yes. In a dedicated pod of 1 to 4 NVIDIA B200 GPUs or a full 8×B200 node you get PyTorch, Jupyter and SSH access to fine-tune open models or train your own. The resulting weights are yours and you can take them whenever you want. We help you size the environment by measuring your workload on our cluster before you reserve capacity.

Which models can read drawings and scanned documents?

For OCR and image-heavy documents we use open vision-language models such as Qwen3-VL 32B OCR and Nemotron Nano 12B VL. They extract text, tables and form fields; for complex drawings it is best to validate first on a sample of your own documents. You can try them on GPU Flow with non-sensitive material before moving to a private environment.

How much does an environment for an engineering team cost?

It depends on usage: a 1/4 fraction of an NVIDIA B200 GPU (45 GB) costs €3.00/h, a full GPU €8.00/h (€5,840 per month) and an 8×B200 node €64.00/h, excluding VAT. Sandboxes start at €0.15 per active hour. On demand there is no commitment; reserved capacity is guaranteed with a six-month minimum. More on pricing.

Does the Cyber Resilience Act apply if we make equipment with software?

Most likely yes, if you place products with digital elements on the EU market. The CRA requires security across the whole life cycle, with reporting of exploited vulnerabilities from 11 September 2026 and the other obligations from 11 December 2027. A private coding assistant helps review firmware without exposing it; the conformity assessment remains yours.

Can we connect our plant to the datacenter securely?

Yes. Besides TLS, we offer dedicated encrypted point-to-point connectivity, with post-quantum cryptography available, to link your sites or plants to your environment in Madrid. It is an add-on without a public price that we size according to locations and volume. Support comes from our own team in Spain, with a response in under 4 hours on business days.

Let's talk

Tell us which knowledge you want to put to work and we will size it on our cluster

Talk to the team →