The Most Expensive Lesson in Building vs. Buying

What Volkswagen's pivot in China tells us about innovation in 2025

Volkswagen just spent 18 months and deployed 500 engineers to build autonomous driving technology for the Chinese market through their CARIZON joint venture. They’re also licensing autonomous driving tech from Xpeng. They’re also working with Horizon Robotics. They’re also partnering with Alibaba for computing infrastructure.

This isn’t strategy. This is desperation covered in press releases.

But it’s instructive desperation. Because what’s happening to Volkswagen in China is a preview of what’s going to happen to a lot of Western companies trying to compete in AI-enabled product categories.

The build trap

Volkswagen’s automotive software division, Cariad, was supposed to be the company’s answer to Tesla. They poured billions into developing software in-house. The vision was clear: cars were becoming software platforms, so Volkswagen needed to build software capabilities.

The execution was less clear. Cariad became known for delays, cost overruns, and technology that lagged behind Chinese competitors. By the time they delivered features for the German market, Chinese EV makers had already moved to the next generation.

This is the trap many established companies fall into: they know they need new capabilities, so they try to build them internally. They hire engineers, set up innovation labs, and announce multi-year development roadmaps. And then they discover that building these capabilities is harder, slower, and more expensive than anticipated.

Meanwhile, the market moves on.

The China speed problem

The term “China speed” appears in Volkswagen’s press releases about CARIZON. It’s not just marketing language. Chinese tech companies genuinely move faster than their Western counterparts in certain domains.

Xpeng develops and deploys autonomous driving features in months, not years. They collect terabytes of real-world driving data daily. They update their systems over-the-air every few weeks. By the time a traditional automotive OEM finishes internal reviews and compliance processes, Chinese competitors have already tested three iterations with real customers.

This creates a compounding advantage. Faster iteration means more learning. More learning means better products. Better products mean more data. More data means faster iteration.

Volkswagen looked at this cycle and realized they couldn’t match it by building everything in-house. So they’re doing something that would have been unthinkable five years ago: licensing core technology from Chinese companies.

The uncomfortable admission

When an established automotive company with over 600,000 employees decides to license autonomous driving technology from a Chinese EV startup, it’s an admission that internal development isn’t working fast enough.

This is a significant shift. Historically, automotive OEMs controlled their core technologies. They might outsource components, but the critical systems were developed internally or with tier-one suppliers they heavily influenced.

Autonomous driving was supposed to be a core technology. Something you build internally because it’s a key differentiator. Volkswagen’s pivot says: we thought it was core, but actually speed and competitive parity matter more than internal development.

The partnership pile-up

Here’s where it gets interesting: Volkswagen isn’t just partnering with Xpeng. They have separate partnerships with Horizon Robotics, with Alibaba Cloud, with Bosch (for non-China markets), and they’re still developing some technology through CARIZON.

This isn’t a clean “build vs. buy” decision. It’s a messy “build some things, partner on others, license still others, and hope they all integrate” strategy.

On one hand, this makes sense. Use partnerships to accelerate time-to-market while building internal capabilities for the long term. On the other hand, it creates integration complexity, strategic ambiguity, and risks of duplicated effort.

When you have three different autonomous driving initiatives running in parallel, you’re either hedging your bets or you don’t know which path will work.

What this reveals about AI development

Volkswagen’s situation is a microcosm of what’s happening across industries with AI. Companies know they need AI capabilities. Many try to build them internally. Most discover it’s harder than expected.

The resources required to compete in AI are immense. Computing infrastructure, talent, data, and rapid iteration capabilities. Large companies have some of these resources, but rarely all of them. And having 80% of what you need still means you lose to competitors who have 100%.

The temptation is to partner your way to competitiveness. Use OpenAI’s models. Partner with Anthropic. License technology from specialists. This gets you to market faster, but it also means you don’t build differentiated capabilities.

The data moat question

Volkswagen’s bet is that partnerships can give them competitive technology while they build something proprietary for the long term. But there’s a problem: the partners are also selling to everyone else.

If Volkswagen licenses Xpeng’s autonomous driving stack, and that same stack is available to other OEMs, what’s Volkswagen’s differentiation? The answer is supposed to be data and brand - Volkswagen’s driving data from their installed fleet, tuned to German driver preferences, integrated with Volkswagen’s UI/UX.

But if the underlying technology is the same, how much does that tuning really matter?

This is the broader question for companies using third-party AI: if everyone has access to the same foundation models, where does competitive advantage come from?

The local market advantage

One thing Volkswagen is getting right: acknowledging that China requires China-specific solutions. The “In China, for China” strategy recognizes that Chinese consumer expectations, traffic patterns, and technology ecosystems are different.

This is a lesson more Western companies need to learn. You can’t just take your global platform and localize the language. You need to rebuild core functionality for local market needs.

Chinese consumers expect their cars to integrate with WeChat, Alipay, and Alibaba’s ecosystem. They expect voice assistants that understand natural Chinese language including regional dialects. They expect real-time updates and social features that would seem invasive in Western markets.

Building all that from your German headquarters doesn’t work. You need local development teams, local partnerships, and acceptance that the China product will be fundamentally different than the global one.

The cautionary tale for other industries

Automotive is ahead of other industries in confronting this build-versus-partner dilemma because the technology requirements hit them earlier. But the same dynamics will play out in retail, healthcare, financial services, and consumer goods.

Any industry where AI becomes a core product capability will face this question: do we build it ourselves, or do we partner? And if we partner, how do we maintain differentiation?

The companies that figure this out will be the ones who are clear-eyed about their actual capabilities versus their aspirational capabilities. Volkswagen wanted to believe they could build world-class automotive software because they’re a world-class automotive company. But those are different competencies.

The speed-quality tradeoff

There’s a risk in Volkswagen’s approach that isn’t being discussed openly: rapid iteration and “China speed” can mean shipping features that aren’t fully tested. Chinese consumers are more accepting of beta features and occasional software issues. German consumers, not so much.

If Volkswagen adopts Chinese development practices for the Chinese market but German quality expectations persist, there’s a mismatch. They need to educate consumers that rapid updates and new features come with some risk of instability.

This is a broader tension in software development: move fast and break things, or move slowly and ensure quality. Chinese tech companies have largely chosen the first path. Western automotive companies have historically chosen the second.

Adopting your competitor’s speed means adopting their risk tolerance too.

What happens to Cariad

The really interesting question is what happens to Volkswagen’s internal software development. Cariad was supposed to be the future. Now it’s being downsized to an “integrator and coordinator” rather than a developer.

That’s a significant downgrade. It’s also probably the right decision. If you can’t compete on development speed, don’t pretend you can. Focus on integration, on ensuring partner technologies work together, and on maintaining overall system architecture.

This is a role that many corporate tech organizations will move toward: less building, more integrating. Less innovation, more orchestration. It’s not as glamorous, but it might be more realistic.

The precedent this sets

If Volkswagen can succeed by licensing core technology from Chinese suppliers, other Western automotive companies will follow. We’re already seeing similar moves from other OEMs exploring partnerships with Chinese tech companies.

This creates an interesting dynamic: Chinese companies become technology suppliers to their Western competitors. That’s a reversal of the historical pattern where Western companies provided technology and Chinese companies provided manufacturing.

It also raises questions about long-term strategic positioning. If your core product technology comes from suppliers, what exactly is your sustainable competitive advantage? Brand? Distribution? Trust?

Those are real advantages, but they’re different than technology leadership.

The talent implications

One underappreciated aspect of this shift: what happens to the engineers? Volkswagen hired software engineers to build autonomous driving technology. Now they’re licensing it instead. Those engineers either get reassigned to integration work (less exciting) or let go.

This has ripple effects. If talented engineers see that companies are choosing to license technology rather than build it, they’ll gravitate toward the companies that are building. That further concentrates talent in a smaller number of technology companies, making it even harder for traditional companies to build capabilities later.

It’s a vicious cycle: companies fall behind, so they partner instead of building, which makes it harder to attract top engineering talent, which means they fall further behind.

What this means for Western tech strategy

The broader lesson here is about honest assessment of capabilities. Volkswagen spent years trying to convince themselves and the market that they could build automotive software as well as tech companies. The China market brutally revealed that they couldn’t.

That clarity, while painful, is valuable. Now they can make realistic decisions about where to compete and where to partner. Companies still pretending they can build everything internally are going to have their own reckoning.

The winning strategy in 2025 isn’t necessarily “build everything” or “partner on everything.” It’s clarity about which capabilities are truly core, which can be sourced, and which require hybrid approaches.

Volkswagen’s messy reality - multiple partnerships, parallel development paths, licensed technology mixed with internal development - might actually be the most realistic approach. It’s just not a clean story to tell to investors and the press.