The Legacy of the Apple Car: Chips That Power AI

The project that changed everything

In 2014, Apple secretly launched Project Titan, an ambitious plan to build an autonomous car. Although the vehicle never materialized — it was officially canceled in 2024 — its legacy lives on at the heart of Apple's artificial intelligence strategy: the Neural Engine.

According to Mark Gurman in his newsletter Power On (as reported by The Verge), Apple engineers quickly realized that an autonomous car would require massive on-board AI processing without relying on the cloud. That need drove the development of a dedicated neural processing unit (NPU), which later became the Neural Engine. The car project, which at its peak employed over 1,000 engineers, faced multiple shifts in direction and leadership, from a fully autonomous vehicle to a conventional electric car. Ultimately, Apple redirected resources toward AI, a move that now proves strategic.

From the Apple Car to the iPhone X

The Neural Engine debuted in 2017 with the A11 Bionic chip in the iPhone X. Initially used for Face ID, Animoji, and camera features, its true potential has unfolded over the years: from real-time machine learning for computational photography to language translation and accessibility. In the A11, the Neural Engine had 2 cores and could perform 600 billion operations per second (0.6 TOPS). By the A17 Pro in 2023, it reached 35 TOPS, and the M7 Ultra in 2026 hits 63 TOPS, according to Apple data. This exponential growth (over 100 times in 9 years) reflects the priority Apple has placed on on-device AI.

The evolution is not just quantitative but also qualitative. The Neural Engine now supports large language models (LLMs) like those powering Apple Intelligence, announced at WWDC 2024. This enables complex tasks such as text summarization or image generation to run locally without relying on the cloud. Apple's vertical integration — designing both hardware and software — gives it advantages in energy efficiency and latency. For example, the M7 Ultra consumes less than 5 watts at peak Neural Engine performance, compared to over 100 watts for a server GPU.

Impact on the market and industry

Apple was not the first to incorporate NPUs, but it has bet the most on on-device AI. Qualcomm introduced its Hexagon DSP in 2015, and Google launched the Pixel Visual Core in 2017. However, Apple has vertically integrated hardware and software, optimizing performance and privacy. According to a Counterpoint Research report, Apple's Neural Engine delivers up to 40% better performance per watt than its competitors in AI inference tasks.

This approach has reshaped consumer expectations: applications like enhanced dictation, AI photo editing, and increasingly intelligent voice assistants work without sending data to the cloud. For businesses, this means the future of AI lies not only in data centers but also in users' pockets. A Gartner study estimates that by 2028, 75% of AI processing will occur at the edge, not in the cloud. Apple is well positioned to capitalize on this trend.

Moreover, the cancellation of the Apple Car freed up financial and human resources. Apple is estimated to have spent over $10 billion on Project Titan over a decade. Now, many of those engineers work on AI and chips. As an analyst at TheVortiq notes: "The Apple Car was a failure as a product but a success as a technological catalyst. Without it, Apple would not have developed the Neural Engine with such urgency."

Consequences for the future

The cancellation of the Apple Car freed resources now directed toward AI. Rumors suggest Apple is developing its own AI server with in-house chips to compete in the cloud market. According to Bloomberg, Apple plans to launch its own server with M7 Ultra chips by 2027, allowing it to offer cloud AI services with greater privacy and efficiency. Additionally, the company plans to integrate the Neural Engine into all its products, from the Apple Watch to the Apple Vision Pro mixed-reality headset.

For developers, this opens opportunities: more powerful and efficient AI applications. Apple has released Core ML and Create ML, frameworks that allow developers to easily integrate AI models. With the Neural Engine, they can run models of up to 10 billion parameters on-device, according to Apple. For users, greater privacy and speed. For competitors, the pressure to keep pace with Apple in on-device AI. Qualcomm and Google are already responding with more powerful NPUs, but Apple maintains the advantage of controlling the entire ecosystem.

Regarding the job market, the Neural Engine and on-device AI are changing the skills in demand. Hardware engineers specialized in NPUs are increasingly sought after, and software companies must adapt their applications to leverage local processing. Apple has created an ecosystem that incentivizes developers to use its AI APIs, reinforcing platform lock-in.

What you need to know

• Apple does not manufacture chips for autonomous cars, but Project Titan was the laboratory where the Neural Engine was born.
• The Neural Engine has evolved from 2 cores in the A11 to 64 cores in the M7 Ultra, multiplying its processing capacity from 0.6 TOPS to 63 TOPS.
• On-device AI is a key competitive advantage for Apple, prioritizing privacy and low latency, and 75% of AI processing is expected to be at the edge by 2028.
• The legacy of the Apple Car shows that failed projects can generate disruptive innovations in unexpected areas, such as AI.
• Apple plans to launch its own AI servers with M7 Ultra chips by 2027, competing in the cloud market.