Europe Isn't Building a Better ChatGPT Either. Should It Be?

The Vertical AI Thesis

The United States and, to a large extent, Europe have built AI ecosystems centred on general-purpose foundation models: large language models, multimodal systems, and reasoning engines designed to do everything for everyone. The competing approach, practised with deliberate intensity by Chinese industrial conglomerates, prioritises what policymakers there call "AI plus": the systematic integration of artificial intelligence into specific vertical industries, from electric vehicles to battery manufacturing, from drone logistics to telecommunications infrastructure.

The result is not one AI champion. It is dozens of world-leading companies that happen to be extraordinary at AI because their industries demanded it. European policymakers and industrialists would do well to study this model carefully, because it represents a direct challenge to the competitiveness of the EU's manufacturing base, which accounts for roughly 17 percent of European GDP.

Margrethe Vestager, the former European Commission Executive Vice-President for digital policy, repeatedly warned during her tenure that Europe risked becoming a consumer of AI rather than a producer of it. The vertical AI gap suggests the concern was warranted, though the category she had in mind may have been too narrow.

BYD: The AI Car Company Hiding in Plain Sight

The case study that best illustrates the vertical AI thesis is BYD. Most European observers still categorise it as an electric vehicle manufacturer, a Tesla rival undercutting Volkswagen and Stellantis on price. That framing misses what BYD has actually become: an AI-integrated vehicle platform company that happens to sell cars.

In early 2025, BYD rolled out its "God's Eye" advanced driver-assistance system as standard equipment across its entire lineup, from a roughly 9,000-euro Seagull to the premium Yangwang U8. This was not a luxury add-on. It was a strategic declaration: every BYD vehicle would become an AI platform.

The architecture operates on three tiers. The top tier deploys a triple-lidar configuration with NVIDIA Orin X processors delivering over 500 TOPS of compute, enabling autonomous navigation without pre-mapped routes. The mid-tier handles urban driving with a single lidar. Even the entry-level system, installed on cars costing less than 13,000 euros, provides lane-keeping, adaptive cruise, emergency braking, and autonomous parking.

What makes this strategically significant is scale. By late 2025, BYD had over 2.5 million God's Eye-equipped vehicles on the road, generating 150 to 160 million kilometres of driving data every single day. That volume of real-world training data, flowing back into a single company's AI systems, represents a vertical integration advantage that no European automaker currently matches. BMW's and Mercedes-Benz's driver-assistance features remain largely optional, premium-tier add-ons rather than a data-generating standard.

BYD has committed 13 billion euros to AI agent and world-model development, supported by a team of over 5,000 engineers dedicated exclusively to autonomous driving. The company has integrated DeepSeek's AI reasoning capabilities into its architecture for real-time decision-making, achieving 98.7 percent efficiency through knowledge distillation on edge devices. Beyond driving, BYD's manufacturing AI has reduced battery defects by 40 percent and improved average battery lifespan by 20 percent. The company's 90,000 engineers are not split between "AI research" and "car engineering." They are one integrated operation where AI is the connective tissue across design, manufacturing, quality control, and the driving experience itself.

This is what vertical AI looks like. Not a chatbot. A car.

The Wider Ecosystem: Batteries, Drones, and Edge Compute

BYD is the most visible example, but the vertical AI strategy runs across an entire industrial base. Consider what is happening simultaneously in adjacent sectors.

CATL, the world's largest battery manufacturer with 38 percent global market share, has turned battery design from manual trial-and-error into an AI-driven process. Its intelligent design platform, trained on more than 100,000 battery design cases and 600 terabytes of test data, generates optimised cell designs in minutes with 95 percent prediction accuracy. The company's manufacturing lines monitor over 6,800 quality-control points using real-time AI image recognition, producing one battery cell per second. CATL won the World Economic Forum's MINDS Award for AI-driven innovation in January 2026, and it is now deploying humanoid robots for end-of-line inspection that autonomously detect wiring anomalies in real time. From 2026, CATL plans to operate large sections of its factories with AI systems and humanoid robots, removing human workers from high-voltage inspection environments entirely.

For European battery manufacturers such as Northvolt, which has already faced quality and financial difficulties, this trajectory is not an abstract competitive threat. It is an existential one. When a competitor can train AI on 600 terabytes of proprietary battery data accumulated over years of production, replicating that advantage through procurement of a general-purpose model is not realistic.

DJI, which dominates the global commercial drone market with over 70 percent share, has shifted its enterprise strategy from "flying cameras" to "aerial AI platforms." Its Manifold 3 edge computing module processes sensor data, flight control, and route planning onboard, with no cloud dependency required. DJI is not competing with OpenAI. It is building the world's best AI for flying machines, with applications in infrastructure inspection, precision agriculture, and logistics. European drone manufacturers, including Parrot and Quantum Systems, are capable companies but operate at a fraction of DJI's data scale.

On the hardware side, Huawei's Ascend 910C chip delivers 800 teraflops of FP16 performance with 128 GB of memory, and the Atlas 900 A3 SuperPod clusters up to 384 of these chips for 300 petaflops of AI compute. Huawei shipped 700,000 Ascend AI chips in 2025, with 600,000 units of the 910C planned for 2026. Critically, its Pangu large models are not designed for consumer chat. They are optimised for industrial applications: predictive maintenance, network optimisation, and manufacturing quality control.

Foundation Models as Infrastructure, Not Product

A common misconception frames the picture as a binary: the West builds foundation models, and vertically integrated manufacturers deploy applied AI. The reality is more nuanced. DeepSeek, a relatively small Chinese lab, reportedly built its R1 reasoning model for a fraction of the cost of comparable Western systems. It sent shockwaves through global equity markets when it demonstrated performance rivalling OpenAI's best on key benchmarks. The model is open-weight, free to use, and already integrated into industrial applications including BYD's autonomous driving architecture.

The critical difference is purpose. Western labs, including Mistral AI in Paris and Google DeepMind in London, treat foundation models as the primary product, monetised through API access and enterprise licensing. Vertically integrated manufacturers treat foundation models as infrastructure: the plumbing that enables AI applications across manufacturing, logistics, consumer devices, and autonomous systems. The open-weight strategy is not charity. It is an accelerant for ecosystem-wide adoption across an enormous industrial base.

This feedback loop compounds over time. Open models accelerate adoption. Adoption generates proprietary industrial data. Proprietary data feeds back into better models. A European carmaker licensing GPT-5 via API does not participate in that loop. It simply pays the toll.

What European Industrial Policy Is Getting Wrong

The contrast with the European AI ecosystem is stark, and it is not flattering. Europe has produced genuine AI research talent, world-class semiconductor equipment through ASML in Eindhoven, and credible frontier-model efforts through Mistral. But the regulatory and investment architecture is structured primarily around horizontal concerns: safety constraints for general-purpose models, data governance, and liability frameworks under the EU AI Act.

Those are not illegitimate concerns. But they are insufficient as an industrial strategy. The EU AI Act, which entered application in stages from 2024, focuses heavily on risk categories and prohibited uses. It says relatively little about how to accelerate the deployment of AI in manufacturing at the pace that competitors are achieving.

Jeremy Fleming, the former director of GCHQ and now a senior fellow at the Cambridge Centre for AI in Medicine, has argued publicly that Western democracies risk ceding industrial AI leadership not through inferior research but through slower deployment cycles and fragmented policy frameworks. The observation applies with particular force to European manufacturing, where AI adoption in production environments consistently trails both the United States and leading Asian manufacturers.

Philipp Gerbert, a partner at Boston Consulting Group's BCG Henderson Institute and one of Europe's most cited analysts on industrial AI, has noted that European manufacturers often treat AI as a layer applied on top of existing processes rather than as a redesign of those processes from the ground up. That distinction matters enormously for the competitive outcomes being generated in battery factories and vehicle platforms right now.

The numbers frame the challenge clearly. Combined, BYD and CATL now control the majority of a global EV battery market where China holds roughly 69 percent share. BYD is on track to produce five million vehicles in 2025. DJI holds over 70 percent of the global consumer drone market. These are not AI companies by Silicon Valley's definition. They are AI companies by the definition that matters: organisations using artificial intelligence to dominate their industries.

The Strategic Comparison

Set against the European approach, the strategic divergence is clearest when examined across several dimensions. On primary focus, the vertical approach embeds AI into specific industries at scale; the European and broader Western approach prioritises building the most capable general-purpose models. On success metrics, the vertical approach measures deployment penetration and industrial output gains, while European labs measure benchmark scores, model capability, and API revenue. On data strategy, vertical manufacturers build sector-specific data flywheels: driving data, battery data, manufacturing data. European AI draws on broad internet-scale training data. On integration speed, God's Eye is standard across every BYD product line; European automakers still treat advanced driver assistance as an optional, premium-tier feature.

The risk profiles differ too. The vertical approach faces quality and safety challenges at scale, and there is evidence that deployment-first strategies can outrun their own guardrails, as seen in some rollbacks of agentic AI in consumer devices. The horizontal approach risks model-capability plateaus and slow industrial adoption. Both risks are real. But only one of them threatens to permanently cede market share in automotive, energy storage, and industrial automation.

What European Manufacturers Should Do Now

The strategic takeaway is not that Europe should abandon frontier model research. Mistral's work is genuinely valuable, and Google DeepMind's London operation produces research that benefits the entire sector. The point is that model capability and industrial deployment are two different races, and Europe is currently entered seriously in only one of them.

European automotive and manufacturing companies need to stop treating AI as a procurement decision and start treating it as a design principle. That means building proprietary data flywheels from production lines and vehicle fleets, investing in edge AI deployment rather than cloud-dependent architectures, and measuring success by defect rates and cycle times rather than by which foundation model they have licensed.

European policymakers, for their part, need to complement the AI Act's risk framework with an affirmative industrial deployment strategy. The European Commission's AI factories initiative and the planned AI Continent Action Plan are steps in the right direction, but they remain oriented toward compute access rather than vertical integration incentives. Closing that gap is the more urgent task.

The AI race is not a single competition with a single finish line. It is multiple races across multiple industries. Europe has a credible position in the race to build capable general-purpose AI. In the race to build the most AI-integrated manufacturing economy, it is, at present, falling behind. The question is not whether European industry can build a better ChatGPT. The question is whether it can build a better battery, a better car, and a better factory floor before the window closes.