Nvidia GTC Berlin 2026: Merz Signs the AI Factory Deal

INDUSTRIAL AI: Nvidia GTC Berlin 2026: Merz Signs the AI Factory Deal

Two days before Jensen Huang's keynote, Germany handed its factories to Nvidia's chips — and the Chancellor asked its family firms to "open your doors for AI." Here's what today's 5am ET keynote means for your money.

Published October 21, 2026 · By Daniel Rao, TechPulse USA

Illustration of an AI-powered German factory with robot arms, a humanoid robot and an engineer, Berlin skyline behind

Nvidia is selling Europe "AI factories," not GPUs — and Germany just signed on. Image: TechPulse USA

The lecture came first. The deal came next. And at 5:00 a.m. ET today (11:00 CEST), Jensen Huang takes the stage at Berlin's Tempodrom to close the loop.

Two days before his GTC Berlin keynote, Nvidia's CEO sat beside German Chancellor Friedrich Merz at the Invest in Germany Summit and did something unusual for a guest: he told the host country it was losing. "It pained him to witness how far" economically advanced Germany had slipped behind on AI, he said, according to Euronews — and offered a single piece of advice: get to work.

Then they signed the paperwork anyway. The German government and Nvidia announced a strategic partnership to put AI directly into Germany's factories, formalized in a Joint Declaration of Intent. Nvidia supplies the chips, the software and its own engineers; Berlin coordinates the partners and writes the check — reportedly a nine-figure euro sum. The target: open, industrial-grade AI models built on German factory data, trained and run on German soil.

This is the keynote preview everyone actually needs: not the chip specs, but the money story behind them.

"It pained him": the tough love that opened Berlin's wallet

Huang arrived in Berlin swinging. Before the summit even started, he complained about the state of Germany's AI progress — and at the fireside chat with Merz, he landed a line that should worry every American boardroom listening to the livestream today:

"Regulation depends on industry growth. You need something worth regulating before the rulebook matters."

The subtext is unmistakable: Europe's instinct to write AI rules is running ahead of its ability to build AI things. Germany, the EU's largest economy, has the manufacturing data everyone else wants and almost none of the AI infrastructure to monetize it.

His other message was aimed at the people whose jobs are supposedly at stake:

"That knowledge belongs to you. It cannot be replaced. AI can only augment your capability, it will not replace your capability."

That framing matters. Europe's industrial workforce is nervous about automation, and a deal this size only works if the engineers buying in don't feel like they're training their replacements.

What they actually signed: open AI models, German factories, Telekom cloud

The declaration isn't a purchase order — it's a letter of intent, which means the numbers are still moving. But the structure is clear from the official announcements:

• The mission: jointly develop cutting-edge, open AI models for physical AI — robots, factory automation, manufacturing planning — rather than chatbots.

• The compute: Deutsche Telekom's Industrial AI Cloud, running on Nvidia's latest Blackwell GPUs. Germany keeps the data and applications inside Germany; the compute stays on a European operator's cloud instead of a US hyperscaler's.

• The money: Handelsblatt reported the government will commit hundreds of millions of euros, partly to rent compute on the Telekom cloud. Neither side has confirmed the final figure.

• The partners: roughly a dozen German companies are expected to participate, with Siemens and Audi named among them.

• The bureaucracy: the declaration formally plugs into Germany's Industrie 4.0 platform — the country's standing public-private manufacturing digitization effort, which was refocused toward industrial AI last year.

Siemens didn't wait for the ceremony to show receipts: the same day, it demonstrated AI planning agents built with Nvidia at its electronics factory in Erlangen. Nvidia's researchers simultaneously announced they had run a record 104-qubit simulation of a fault-tolerant quantum computer on the JUPITER supercomputer in Jülich — a reminder that this week's conference is about raw compute muscle, not just policy handshakes.

"Please open your doors for AI": Merz's bet on the Mittelstand

The most telling moment wasn't the signing — it was Merz's direct plea to the Mittelstand, Germany's family-owned small and mid-sized manufacturers, which hold decades of production data behind closed doors:

"The biggest advantage we have in Germany is our so-called Mittelstand. Please open your doors for AI to make use of your data."

Merz said Germany wants to be "one of the most advanced places in the world" for integrating AI into industry — "one of the leading, if not the leading, country in the world" for the spread of AI across manufacturers, machinery makers and automakers — and suggested meeting Huang at least once a year to check on progress.

He had reason to be confident about the room. Merz claimed the summit generated €800 billion in investment commitments through 2028. And here's the number most AI coverage missed: Amazon announced a roughly €16 billion German investment at the same summit, with about €4.8 billion of it going to expand AWS data centers around Frankfurt. The Germans aren't just renting AI — the American cloud giants are pouring concrete there too.

Huang's shopping list: four startups, one thesis

The Germany deal wasn't Huang's only announcement. He unveiled planned investments in four German startups — part of what Nvidia calls a ~$10 billion (≈€8.6 billion) commitment to European companies, universities and research institutes:

• Agile Robots (Munich): robot arms and the Agile ONE humanoid for factories, logistics and healthcare

• Avelios Medical (Munich): an AI-native hospital information system

• Mirelo: AI models that generate sound, speech and music for video

• Reflex Aerospace (Berlin): satellites it says can go from design to launch in about a year

Nvidia disclosed neither the size of each stake nor closing dates. But the pattern is the thesis Huang has been pushing all year: "The next wave of AI will be physical." Factories, robots, cars, heavy industry — the domains where AI's biggest gains come next. For Nvidia, these bets extend its reach from selling chips into owning a stake in the software that runs actual production lines.

That's also the number to put beside Monday's market bounce: the PHLX Semiconductor Index rose 1.6% on Monday, snapping an eight-day losing streak, and Nvidia reports against a tape that is finally rewarding the capex story again.

Illustration of a European sovereign AI data center with server racks and digital-twin factory visualization

Compute runs through Deutsche Telekom's Industrial AI Cloud on Nvidia Blackwell GPUs. Image: TechPulse USA

Everything else that dropped before the keynote

Nvidia spent Monday warming up Berlin like a product launch:

• RTX Spark PCs went on sale: Windows laptops with Blackwell RTX graphics and up to 128GB of unified memory, starting at $2,399, from ASUS, Dell, HP, Lenovo, Microsoft and MSI — with a surprise Huang appearance at the Berlin celebration.

• The Build-a-Claw developer tour closed in Berlin with 500+ builders competing on autonomous agents built with Nvidia's new OpenShell zero-trust agent framework.

• Nvidia says it has invested nearly $1 billion in Germany over the past year through startup and company stakes.

The keynote delivered: DSX POD, OpenShell, and the $50B-per-gigawatt math

The keynote landed, and it was a product event, not a policy speech. The biggest hardware drop was DSX POD: factory-built AI data centers that Nvidia says can go from a prepared site to producing tokens in as little as 14 days, versus roughly six months today — each unit a complete Vera Rubin system (compute, networking, cabling, power, liquid cooling) assembled and tested inside modular containers before delivery, with Dell, HPE, Foxconn, Schneider Electric and Vertiv signed on for worldwide deliveries from early 2027 (aistockwire). The software headline was the OpenShell Reference Architecture for safe AI agents — Huang's line: agents "think on Rubin, act on Vera and be governed on BlueField," with BlueField-4 DPUs running NVIDIA Sentry for out-of-band monitoring and in-silicon policy enforcement — and a partner list that reads like the industry's phone book: Anthropic, AWS, Cisco, CrowdStrike, Dell, Google Cloud, HPE, IBM, Meta, Microsoft, OpenAI, Palantir, Perplexity, SAP, ServiceNow, Siemens and SpaceX Super Intelligence (Morningstar). Huang also put numbers on the buildout: a one-gigawatt AI data center costs about $50 billion and should generate about $40 billion in annual revenue — so a three-month deployment delay postpones roughly $10 billion in potential revenue. And Nvidia, Nokia and nine EMEA telecom operators announced they are turning networks into distributed intelligence platforms, with T-Mobile completing the world's first AI-RAN call and Deutsche Telekom advancing physical AI for industries (Morningstar). NVDA shares were up 1.9% at $220.53 in premarket trading (aistockwire). The earnings context that framed the keynote still applies: see our earlier reporting on Meta's Q3 earnings preview and the capex question and Tesla's Q3 setup, where the Roadster's spectacle meets a margins test.

Why it matters: what it means for your money

Here's the honest read. The letter of intent is not a contract — the euros aren't wired, and "hundreds of millions" is a rounding error next to the €800 billion the summit claims in total commitments. Europe's sovereign-AI money flowing to a Telekom-hosted cloud instead of a US hyperscaler is a small but real competitive sting for Azure, GCP and AWS — and Nvidia's critics will note that much of its "investment" comes back as GPU orders.

But the direction is what matters for US investors and workers:

• Capex is the trade. Every sovereign AI factory — German, French, British — is a GPU order. That's why semis bounced Monday. When Meta answers its capex skeptics tonight, remember: the demand they're doubting is the same demand Merz is paying for.

• Physical AI is the jobs angle. Factory-floor AI, robotics, planning agents — this is automation designed to augment skilled workers, not replace call centers. If it works, the Mittelstand model (family firms, apprenticeships, regional jobs) gets a productivity upgrade instead of a disruption shock.

• The risk is delivery. No budget, no timetable, no first model has been named. Past Nvidia Europe pledges have occasionally been more headline than hardware. Hold the applause until the keynote puts megawatts and site names on the slide.

The honest verdict: Huang's Berlin trip is the strongest version of the AI-infrastructure thesis — factories instead of chatbots, open models instead of walled gardens, sovereign compute instead of hyperscaler dependence. The keynote at 5 a.m. ET will show whether there's hardware behind the handshake.

FAQ

When and where is Jensen Huang's GTC Berlin keynote, and how can I watch? The keynote is Wednesday, October 21, 11:00–13:00 CEST (5:00–7:00 a.m. ET, 2:30–4:30 p.m. IST) at the Tempodrom in Berlin. It is livestreamed free, with no registration required, at nvidia.com/en-eu/gtc. A pre-keynote show begins at 9:00 a.m. CEST.

What is "physical AI"? AI that controls or interacts with physical systems — robots, factory equipment, vehicles — rather than producing text or images. It is the centerpiece of the Germany-Nvidia deal: open AI models built for industrial use cases like manufacturing planning and factory automation.

How much money is Germany putting into the Nvidia deal? The exact figure is unconfirmed. Handelsblatt reported a sum in the hundreds of millions of euros, partly to rent computing power on Deutsche Telekom's Industrial AI Cloud. The agreement was announced as a Joint Declaration of Intent, not a finalized contract.

Why should a US reader care about a German AI deal? Three reasons: (1) sovereign AI factories are GPU orders — this is the demand side of the capex debate driving NVDA and the chip sector; (2) it sets the template other countries may copy, competing with US hyperscalers abroad; (3) the physical-AI wave targets manufacturing jobs and supply chains that touch US exports.

What else is Nvidia expected to announce at GTC Berlin? Watch for the itemization of Huang's $2 billion Europe investment pledge, new European AI-factory site announcements with power figures, hardware updates (GB300/Rubin, Vera CPU), RTX Spark availability, and sovereign open-model releases.

What did Jensen Huang announce at the GTC Berlin keynote? (1) DSX POD — factory-built AI data centers deployable in ~14 days at a prepared site (vs ~6 months), each a complete Vera Rubin system, Dell/HPE/Foxconn/Schneider Electric/Vertiv partners, deliveries from early 2027; (2) OpenShell Reference Architecture for safe AI agents — "think on Rubin, act on Vera, be governed on BlueField" — with Anthropic, AWS, Cisco, CrowdStrike, Dell, Google Cloud, HPE, IBM, Meta, Microsoft, OpenAI, Palantir, Perplexity, SAP, ServiceNow, Siemens and SpaceX Super Intelligence on board; (3) the $50B-per-gigawatt math — a 1GW AI data center costs ~$50B and generates ~$40B annual revenue, so a three-month delay postpones ~$10B; (4) Nvidia + Nokia + 9 EMEA telecom operators on AI-RAN, with T-Mobile completing the world's first AI-RAN call and Deutsche Telekom advancing physical AI for industries.