NVIDIA on October 12, 2026 published a roundup of French companies running AI agents and open models on its hardware, and the most concrete commerce figure in it belongs to Mirakl: more than 70 million products onboarded with fine-tuned and open models running on NVIDIA GPUs and inference software, cutting seller catalog onboarding to under two hours. The same post disclosed planned NVIDIA investments in four French startups - Doctolib, Entalpic, Genesis AI and Pigment - without naming a single amount.

In Short

NVIDIA, the company that makes most of the chips used to run AI, wrote a blog post listing French companies that use its technology and four French startups it plans to invest in. For people who sell online or buy ads, the interesting part is that marketplace software firm Mirakl says it now gets a seller's product catalog live in under two hours, and image firm Omi says it puts real products into AI-made marketing pictures faster. None of the speed claims come with a starting point or independent check, so they are company-supplied numbers rather than proven results.

What NVIDIA published

The post appeared on the NVIDIA company blog on October 12, 2026, under the byline of Nat Ives, nine days before NVIDIA founder and CEO Jensen Huang's GTC Berlin keynote, scheduled for Wednesday, October 21, at 11 a.m. CEST. The page captured for this article carried the headline "NVIDIA Announces New Investments in France's Startups", while the two embedded videos on the same page were titled "France's Ecosystem Delivers Agentic Computing at Scale With NVIDIA". Both framings sit in one document. One is a capital story; the other is a customer list.

According to NVIDIA, companies in France are deploying agentic computing across space, financial services, the public sector and scientific research, "demonstrating how the country's ecosystem has moved from ambition to application." The post traces that progress to the AI Action Summit in February 2025 and the Choose France Summits in 2025 and 2026, which, according to NVIDIA, "anchored major compute commitments on French soil."

Across Europe, NVIDIA says it has committed nearly $10 billion to the technology ecosystem. The post does not break that figure down by country, by instrument or by year, and it does not say how much of it has already been deployed. That matters, because the number is the subhead of the piece and the only aggregate financial figure in it.

Four planned investments, no amounts

NVIDIA said it had previously invested in Alice & Bob, Gradium and Scintil Photonics. The new planned investments, according to the post, are in four companies:

  • Doctolib, a European healthcare technology company that, according to NVIDIA, builds agents to free up healthcare professionals and provides patients with a digital health companion.
  • Entalpic, a Paris-based startup working on engineering advanced materials at the atomic scale, with particular expertise in surfaces and interfaces for semiconductor manufacturing.
  • Genesis AI, described as a global robotics company developing general-purpose robots through foundation models, data collection systems, simulation infrastructure and robot hardware.
  • Pigment, a Paris-based enterprise software company whose planning platform helps businesses forecast performance and model scenarios across finance, sales, workforce and operations.

The word "planned" carries weight. The post does not state round sizes, valuations, whether NVIDIA leads any round, or when the deals are expected to close. Pigment is the name with the most direct line to marketing departments: planning software that models sales and workforce scenarios is where media budgets are often set before a single campaign launches. Doctolib, by contrast, sits in health data, one of the most heavily regulated categories in European privacy law.

The commerce and marketing claims

Most of the post concerns infrastructure, science and public services. Four entries, however, touch the work of retailers, marketplace operators and brand marketers directly, and they are the ones with the most immediate relevance to advertising.

Mirakl: 70 million products, under two hours

According to NVIDIA, Mirakl "has onboarded more than 70 million products with fine-tuned and open models running on NVIDIA GPUs and inference software, cutting seller catalog onboarding to under two hours."

Two things are absent. The post does not say what the onboarding time was before, so the scale of the reduction cannot be calculated. Nor does it say over what period the 70 million products were processed, or which open models Mirakl used. The figure is company-supplied, published by the hardware vendor whose products are named in the same sentence.

Mirakl's place in retail media gives the claim its relevance. The company's advertising arm, Mirakl Ads, signed a global integration with Criteo in July 2025 aimed at bringing smaller marketplace sellers into retail media buying, and in June 2026 Broadsign and Mirakl Ads announced an integration to plan and report online and in-store screen campaigns together. PPC Land reported at the time that Mirakl's platform supports more than 450 marketplaces and more than 100,000 third-party sellers.

Onboarding speed is not a cosmetic detail in that business. A seller's products cannot be sponsored on a marketplace until they exist in its catalog, with structured attributes the ad system can match against queries. Faster catalog ingestion therefore shortens the gap between a seller joining a marketplace and that seller becoming a retail media advertiser. Whether two hours is a median, a best case or a ceiling, the post does not say.

The direction mirrors moves elsewhere. Google added a beta to Merchant Center in May 2026 that uses AI to scan a website once and create product listings, so merchants can skip building a structured feed. The difference is the operator. Google's tool serves its own Shopping surfaces; Mirakl sells the capability to marketplace owners who run their own ad businesses.

Omi: product twins inside generated visuals

According to NVIDIA, Omi "helps brands bring accurate products into AI-generated marketing visuals at scale, combining its approved 3D digital twins with open models optimized on NVIDIA for 4x faster inference."

The claim addresses a known weakness of generative imagery for commerce: models that invent or distort product details. Omi's approach, as described, starts from an approved 3D model of the actual product rather than from a text prompt alone. The 4x figure refers to inference speed, the cost of producing each output, not to any measured campaign result. The post does not state the comparison point for the 4x, the models involved, or any effect on click-through or conversion.

Adobe made a similar bet earlier this year. When Adobe and NVIDIA announced their collaboration in March 2026, it covered next-generation Firefly models and a cloud-native 3D digital twin product in public beta. Omi is a smaller, French-based entrant in the same category, and the NVIDIA post is, for now, the only source attached here for its performance claim.

Photoroom and Vestiaire Collective: impact without a number

The thinnest entry concerns two of the best-known French names on the list. According to NVIDIA, "Photoroom and Vestiaire Collective are demonstrating the measurable business impact of advanced visual modeling on marketplace performance."

That is the whole of it. The post describes the impact as measurable and then does not measure it. There is no conversion figure, no listing-quality metric, no time saved and no description of what visual modeling means in Vestiaire Collective's resale marketplace. Photoroom is a product-photo editing company; Vestiaire Collective is a secondhand fashion marketplace where image quality affects whether used goods sell. A reader looking for evidence of effect will not find it in this source.

L'Oréal: up to 100x, in the lab

According to NVIDIA, L'Oréal is integrating NVIDIA ALCHEMI into its research workflows so that scientists can explore formulation possibilities in silico. Combining computational predictions with hands-on experimentation "can make the discovery process up to 100x faster than traditional methods," the post states.

This is research and development, not marketing. The phrasing is conditional ("can make"), and "up to" marks a ceiling rather than an observed average. It is included here because L'Oréal is among the world's largest advertisers, and changes to how fast it can bring new formulations to market eventually feed product launch calendars and the media plans built around them.

Sovereign infrastructure and open models

The infrastructure half of the post is where most of the hard numbers sit.

Bull is extending its European AI server production to NVIDIA Vera Rubin Ultra NVL72 systems, planned for 2028, with trays produced in Europe and rack integration at its Angers factory, which was inaugurated on Friday, October 2, according to NVIDIA.

Further from the ground, Altair-1, the first satellite in a $1 billion, 50-satellite constellation with on-board autonomous capabilities, launched this month and reached orbit, according to the post. It was integrated by Loft Orbital and built in Abu Dhabi by Orbitworks, a joint venture between Loft Orbital and Marlan Space, and it flies NVIDIA Jetson AGX Orin edge modules running Mistral's models for real-time Earth observation and autonomous operations. The post does not give the launch date or name who is financing the $1 billion constellation.

On models, NVIDIA says its Nemotron open models are in production across French organizations "that require sovereign, trustworthy data." The largest deployment claim concerns LINAGORA and Docaposte, which secured the CANUT framework agreement, making Twake.ai, built on Nemotron, "the sole referenced sovereign digital workplace for more than 2,500 French public organizations," according to NVIDIA. A framework agreement makes a product available to buy; it does not by itself mean 2,500 organizations have deployed it.

In financial services, SBS is launching a sovereign platform for banks to deploy governed generative and agentic AI services within their own environments, in collaboration with NVIDIA. And Fondation Inria, with NVIDIA's support, is starting a working group on open-source advanced modeling and tool adoption, with findings expected in spring 2027.

The post also notes that "last week" Mistral introduced Mistral Large 4 under private preview, describing it as offering "state-of-the-art, open weight performance in cyber defense, manufacturing, finance and multimodal tasks." That characterization is NVIDIA's, and it is not uncontested. Two days after the preview, PPC Land reported that Ecosia was dropping Mistral for other open-weight models, with Ecosia's chief executive calling Mistral's models "a year behind." That article put the Large 4 preview at October 6, with 1 trillion parameters and 52 billion active.

Public sector, telecoms and quantum

The remaining entries describe operational deployments outside commerce.

In the Côte-d'Or department, according to NVIDIA, ThinkDeep and Akila built a digital-twin agent powered by NVIDIA Cosmos open world models, Nemotron and Omniverse libraries, managing nearly 5,800 kilometers of roads and buildings. The post says it cut total operating costs by 10% and targets 30% as deployment progresses. The 10% figure is the only cost saving in the entire document stated as achieved rather than possible, and it is company-supplied. Port Sud de France, the post adds, runs goods processing and port security on ThinkDeep's DeepBrain platform.

Enchanted Tools, a Paris robotics company, is deploying social robots in healthcare settings including pediatric cancer care. Bouygues Telecom and Prisme.ai have launched AI Studio V2, which, according to NVIDIA, lets all Bouygues Telecom employees build their own agents in a controlled environment. The post gives no employee count or usage figure.

In quantum computing, Alice & Bob has integrated the NVIDIA NVQLink platform for its superconducting cat qubits, and Quandela has used NVQLink to connect a DGX Spark system to its photonic qubits, according to the post.

Why it matters for marketers

Read as a whole, the post is a vendor showcase timed to a keynote. It names more than 20 companies, four planned investments with no disclosed values, and a handful of performance figures, every one of them supplied by NVIDIA or its customers. That is not unusual for a corporate blog, but it bears on how much weight the marketing-relevant claims can carry.

Three threads connect it to advertising.

The first is catalog and creative production. Mirakl's onboarding claim and Omi's product-twin claim both concern the raw material of commerce advertising: structured product data and images that accurately depict what is for sale. As shopping agents start choosing products on behalf of consumers, that raw material is under new pressure. PPC Land reported on October 11 that a Profound test found Meta's Muse agent drew four of its six typical product picks from two merchant websites it browsed live, not from Meta's own catalog. In September, Google's Heiko Hotz pitched the Shopping Graph's 60 billion listings as the base for AI shopping agents. Whoever controls clean, current product data is better placed in either model.

The second is the sovereignty framing. Nearly every infrastructure entry in the post leans on it: sovereign workplaces, sovereign banking platforms, European-built servers. The pressure behind that language is concrete. In June 2025, a Microsoft France executive told a Senate inquiry that he could not guarantee French citizen data would never be handed to US authorities. AWS made its European Sovereign Cloud generally available in January 2026, a sign that the large hyperscaler providers see the same demand. For advertisers and agencies handling customer data in France, the question is no longer abstract: where models run determines which law governs access to the data fed into them. The hardware in nearly every one of these sovereign stacks is supplied by a US company.

The third is concentration. In July 2026, France's competition authority found that OpenAI, Google and Anthropic held more than 84% of the global AI agent market in May, and flagged lock-in risk. The NVIDIA post presents open models such as Nemotron and Mistral's as the alternative to that concentration at the model layer. At the chip layer, the alternative is not discussed. The same pattern appeared in adtech this summer, when NVIDIA used Cannes Lions to showcase ad tech partners including Criteo, Taboola and AWS running on its GPUs, and in April, when LiveRamp moved its clean room compute to NVIDIA GPU infrastructure. Earlier this month, Zeta Global said it was building specialized models on top of NVIDIA's Nemotron for its AthenaOS platform.

There is also a sceptical reading of the economics. Cory Doctorow argued in September that AI generates about $50 billion in annual revenue against trillions in spending, and named NVIDIA as one of the seven companies carrying a large share of US stock value. A post that pairs customer success stories with planned equity stakes in those customers' peers invites the same question he raised: how much of the demand being showcased is independent of the vendor showcasing it? The October 12 post does not answer that, and was not written to.

What remains to watch is narrower and more measurable. Does Mirakl publish a before-and-after onboarding figure? Do Photoroom and Vestiaire Collective put a number on "measurable business impact"? And will the amounts behind the four planned investments, and behind the nearly $10 billion European commitment, appear at GTC Berlin on October 21?

Timeline

Summary

Who: NVIDIA, which published the post, and the French companies it names, including Mirakl, Omi, Photoroom, Vestiaire Collective, L'Oréal, Bull, Loft Orbital, Mistral, LINAGORA, Docaposte, SBS, ThinkDeep, Akila, Bouygues Telecom, Prisme.ai, Alice & Bob and Quandela, plus the four startups NVIDIA plans to invest in: Doctolib, Entalpic, Genesis AI and Pigment.

What: A roundup of French deployments of agentic AI and open models on NVIDIA hardware, paired with planned investments of undisclosed size. The commerce claims most relevant to advertising are Mirakl's 70 million products onboarded with seller catalog onboarding under two hours, and Omi's 4x faster inference for product-accurate marketing visuals, both company-supplied and without published baselines.

When: The post was published on October 12, 2026, ahead of Jensen Huang's GTC Berlin keynote on October 21, 2026. Bull's Angers factory opened on October 2, and Mistral Large 4 entered private preview the week before publication.

Where: France, with the satellite built in Abu Dhabi and NVIDIA's nearly $10 billion commitment described as covering all of Europe.

Why: NVIDIA is positioning its chips and Nemotron models as the base for European "sovereign" AI at a moment when French regulators have flagged concentration among US AI agent providers. For marketers, the relevance lies in catalog speed and product-accurate creative, the inputs that retail media and shopping agents depend on, though the evidence in the post rests entirely on vendor-supplied figures.