How Apple Competes in AI

The market is still judging Apple’s AI strategy through the wrong lens. After WWDC, most of the conversation has focused on Siri demos, Gemini rumors, and whether Apple is spending enough on AI data centers...but Apple is not trying to win the hyperscaler AI capex race. Apple is trying to win the AI routing race inside the personal computing stack. My latest piece argues that Apple’s AI flywheel is structurally different from the hyperscaler flywheel. Microsoft, Google, Amazon, Meta and Oracle are building centralized AI factories. Apple is building something different: a distributed AI architecture across Apple Silicon, on-device inference, Private Cloud Compute, partner model supply, and Apple Intelligence as the orchestration layer. That distinction matters. If Apple can route routine, personal, privacy-sensitive tasks to the device, escalate heavier private workloads to PCC, and use frontier model supply only where needed, then its AI economics look very different from a cloud-first model where every interaction becomes a metered compute event. The model is becoming a component...and the experience is becoming the product. Orchestration may become the control plane. The question is not simply who has the best model...the question is who controls where intelligence runs. That is the layer where AI economics becomes platform power. Article below.

The "Apple is shit at AI" takes reveal a painful to read misunderstanding of hardware, distributed systems, and the sociotechnical realities of AI deployment. I'm not an Apple fanboy—I'm an engineer who recognises superior technical strategy when I see it. The criticism following WWDC 2025 shows exactly what's wrong with tech discourse. While everyone chases GPT hype, Apple quietly built the infrastructure that everyone is now copying: 🔹 7-year Neural Engine evolution: 0.6 TOPS (2017) → 38 TOPS (2024). That's 63x performance improvement in custom AI silicon. 🔹 Google I/O 2025 Edge AI announcement: Literally implements Apple's 2017 strategy—Gemma 3n on-device multimodal models, on-device RAG, function calling libraries. Validation doesn't get clearer. 🔹 Private Cloud Compute: The only cloud AI system with verifiable privacy guarantees. Custom silicon, stateless computation, cryptographic attestation. Technical achievement competitors can't match. The critics fundamentally misunderstand the problem space: ❌ "Siri isn't ChatGPT" → Missing the point. On-device inference vs cloud dependency. Privacy vs data harvesting. ❌ "Apple is conservative" → 38 TOPS Neural Engine while competitors scramble for edge deployment. ❌ "Behind in AI" → Foundation Models framework gives developers AI inference that's free, private, and offline. 3 lines of Swift code. Technical reality: Edge AI market exploding $20.78B → $269.82B by 2032. Apple didn't follow trends—they created them with 7 years of Neural Engine development. The sociotechnical implications matter more than benchmark porn: • 485% increase in corporate data sharing with AI tools • GDPR/EU AI Act compliance requirements • Real-world latency constraints • Energy efficiency at scale 🪃 Here's the real issue: Everyone got drunk on AGI fantasies and forgot to build AI that actually works in resource-constrained, security-conscious environments. I've watched companies burn billions chasing GPT-scale models while ignoring fundamental deployment realities🪃 Apple built sustainable competitive moats while others optimised for TechCrunch headlines. WWDC 2025: Foundation Models framework gives developers direct API access to on-device Apple Intelligence. 3 lines of Swift code for private, offline AI inference. Free of cost. The future of AI is (and SHOULD be) distributed, private, and efficient. Apple saw this coming in 2017. Sources: https://lnkd.in/garVXrGt https://lnkd.in/gSCFGWer https://lnkd.in/g6nCgXRG https://lnkd.in/gwWtGEXN https://lnkd.in/garVXrGt #AI #TechnicalStrategy #EdgeComputing #Privacy #Apple #MachineLearning

Apple was never out of the AI race. It was just playing a different game. . . While everyone else was racing to build the smartest AI models, Apple took a step back and focused on rebuilding its foundation. Now, things are starting to make sense. Here’s what Apple has been doing quietly: • Rebuilding Siri from scratch Not upgrading. Not fixing. Completely redesigning it to handle conversations, context, and real tasks. • Developing its own AI system On-device intelligence + private cloud. Focused on privacy and seamless usage rather than hype. • Working on deep integration AI is not a separate app. It’s being built into photos, camera, messages, and the entire ecosystem. And now comes the comeback move: • Partnering for power Instead of spending years catching up, Apple is using external AI (like Google’s models) to accelerate. • Focusing on experience over models They are not trying to win “best AI model.” They are trying to win “best AI experience.” What this means in simple terms: • Siri becomes an actual assistant, not just a voice tool • Camera becomes a search engine • Your phone starts understanding context, not just commands This is a very different strategy. Others are building AI you open. Apple is building AI you don’t even notice. That’s the real shift. Apple’s comeback is not about being first. It’s about being deeply integrated. And if this works, the competition is not just about intelligence anymore. It’s about who controls the experience.

Apple may be making one of its smartest AI moves by refusing to define success as winning the model race. That may sound counterintuitive, but founders should pay close attention. Apple already owns three powerful things, namely 1.5B devices, trust and privacy, and the default interface. If Siri becomes the gateway to multiple AI models, Apple does not need to own the deepest intelligence. It can own the point of interaction. That is where distribution, trust, and habit compound. Groq offers a useful parallel. The technology breakthrough in inference was real, yet the strategic unlock came when the company shifted from selling chips to selling tokens. That shift changed the position in the value chain: how value was captured, and how adoption could scale. Apple is shaping where AI will be used. Groq shaped how AI could be consumed. Different companies. Different layers. Same discipline. The largest outcomes often come from choosing (or finding) the right layer to own. For founders building with atoms, photons, and electrons, this is worth reflecting on. Breakthrough technology is the beginning. Strategy is deciding where you become essential. 1.Are you building a component, or are you becoming part of the system others rely on? 2. Are you creating a one-time sale, or are you growing with every use? 3. Are you impressive, or are you becoming difficult to route around? Enduring advantage comes when you sit in the flow of usage, become part of the workflow, and matter at a layer that others cannot easily replace. Apple shows the power of distribution. Groq showed the power of business model innovation. The lesson is the same: Do not only build what matters. Choose where you matter most.

If you’ve been following the Big Tech companies’ earnings reports, you know that they’re pouring more than ever into capital expenditure to pursue their AI futures. Amazon, Alphabet, Meta, and Microsoft all spent record sums last quarter on purchases of property and equipment — largely tied AI chips and data centers. And for the companies that offered forward-looking guidance, their capex plans for the year blew analysts’ already generous estimates out of the water. Amazon expects its 2026 capex to surge to $200 billion. Google is aiming for $175 billion to $185 billion. Meta estimates it will spend between $115 billion and $135 billion. All of those figures came in well above expectations and, for the most part, have weighed on their stocks. Microsoft didn’t give a formal 2026 capex outlook, but if its peers are any indication, spending will likely exceed the roughly $114 billion Wall Street expects for the calendar year. Of the Big Tech companies, just one stands apart this earnings season. Apple’s capital expenditure, already just a fraction of its peers, actually declined in the December quarter from a year earlier. For better or worse, Apple has struck its own path with AI. As we’ve argued before, it’s embracing AI but is not an AI company. Instead, it’s chosen a hybrid model, relying on both first- and third-party data centers — a move that keeps a significant amount of infrastructure spending off its balance sheet. And while Apple has said it expects capex to increase as it invests more heavily in AI, particularly to support its Private Cloud Compute, those outlays remain minimal compared with its peers. You can see that approach reflected in Apple’s decision to use Google’s Gemini, rather than an in-house model, to power the next generation of Siri and Apple Intelligence. The Google deal, reportedly worth about $1 billion a year, gives Apple access to a top-tier AI model for pennies on the dollar compared to what other Big Tech companies are spending to build their own. Of course, it also means Apple won’t fully own a technology that some see as powering the next industrial revolution. But if that revolution fails to materialize — or takes longer than expected — Apple won’t be left holding the most expensive bag in Silicon Valley history. https://lnkd.in/eDTFzE46

🤖 Apple partners with Alibaba to bring AI features to China Apple’s latest move—partnering with Alibaba to bring AI-powered experiences to Chinese iPhones—is a fascinating case study in navigating AI’s global landscape. While OpenAI’s models power Apple Intelligence in Western markets, in China, it’s Alibaba’s Qwen models stepping in to fill the gap. At first glance, this is a pragmatic response to regulatory constraints. China’s AI laws require domestic partnerships for AI deployment, and Alibaba has the data infrastructure and compliance framework to make it work. But there’s more to this than just legal maneuvering. 🔍 What does this mean technically? → Apple gets a localized AI stack that aligns with Chinese regulations while still delivering an advanced AI experience. → Alibaba’s Qwen models (which have been making strides in NLP and multimodal AI) will be optimized for Apple’s ecosystem. → This ensures on-device privacy while also leveraging cloud capabilities within China’s regulatory framework. 🧠 My take: This is a necessary move for Apple, but it’s also a reminder that AI isn’t one-size-fits-all. The future of AI isn’t just about building the best model—it’s about deploying it where and how it makes sense. While Apple’s global AI strategy prioritizes on-device intelligence, the reality is that data access, compute infrastructure, and regulatory landscapes vary dramatically across regions. This deal shows that AI’s future isn’t just about building cutting-edge models—it’s about deploying them in ways that make business and geopolitical sense. 🌍 Big picture? Expect more AI localization efforts as companies realize that competing in global markets requires regionalized AI strategies, not just universal models. I am going to seek an eye out to see how this plays out—either we will see Chinese users embrace Apple’s AI experience powered by Alibaba, or homegrown alternatives (like Huawei and Xiaomi’s AI efforts) will still to dominate! What do you think—does this move help Apple stay competitive in China? Drop your thoughts below! 👇 ------- Share this with your network ♻️ Follow me (Aishwarya Srinivasan) for more AI news, insights, and educational content to break into AI.

Apple is building the foundation for where AI will live in the real world. Apple has been playing the long game on AI interfaces through disciplined dealmaking and partnerships. Instead of chasing headline-grabbing models, it has acquired the core building blocks required to run intelligence efficiently, reliably, and privately across billions of devices. → DarwinAI strengthens model optimization and visual inspection → Datakalab deepens machine vision capabilities → WaveOne advances compression critical for spatial and video-heavy AI workloads → WhyLabs improves model monitoring at scale → Q, acquired in early 2026, centers on high-bandwidth human communication The acquisitions reinforce Apple’s push to make AI feel native and seamless rather than like a separate destination. Beyond acquisitions, Apple’s expanding partnership network serves the same strategy. Its collaboration with Google on next-generation foundation models, alongside relationships with Anthropic and NVIDIA, reflects pragmatism. Apple is partnering to accelerate capability where it makes sense, while keeping the delivery layer firmly inside its ecosystem. It does not need to win the model race outright. It needs models that power experiences across iPhone, Watch, AirPods, and the next generation of spatial devices. Similarly, supply chain and manufacturing relationships are foundational to the AI hardware roadmap. Key partners are enabling the miniaturization, advanced displays, sensor integration, and manufacturing scale that make new AI-first form factors viable. Apple is laying the groundwork for intelligence that is distributed, persistent, and embedded into everyday life. Not a chatbot you open, but an ambient layer that moves with you across contexts and devices. Apple's endgame is clear: AI that lives on your body, in your ear, in your vision, and across the Apple ecosystem. P.S. CB Insights Strategy Maps are available for any company.

Apple just admitted something the rest of us learned the hard way. Intelligence is a commodity you rent. Context is infrastructure you own. Apple and Google just signed a deal to use Gemini AI to power Siri's future capabilities. Think about that for a second. Apple - the company that builds everything in-house, controls the entire stack, and famously never relies on external partners - is renting AI intelligence from Google. Because they figured out what every team running agents in production already knows: The model isn't the differentiator. The context is. Apple already owns the richest context layer in consumer tech. They own your messages, calendar, photos, contacts, app usage, and location history (interestingly, so does Google…). That's the infrastructure that makes AI useful. That's what they're keeping. The intelligence layer? That's interchangeable. This is exactly what I've been seeing with teams trying to run AI agents at work. They obsess over which model to use, the fine-tuning, and the prompt engineering. Meanwhile, their agents fail because they can't access the context they need. The CRM data is in Salesforce, the project status is in Asana, the customer history is in Netsuite, and the budget information is in some spreadsheet nobody can find. The agent might be brilliant, but if it can't see the full picture, it's useless. Apple gets this and they're not trying to win on intelligence. They're betting on context infrastructure. Rich, real-time, connected data across every system you use. That's the foundation that makes any AI model valuable. The model is the easy part. You can rent that. The context layer? That you have to own.

My hot take on the news of Apple using Google Gemini to overhaul Siri. For those who’ve been in tech long enough, this feels like a repeat of 1997. Back then, Steve Jobs took $150M from Microsoft and made Internet Explorer the default browser on Mac. To the public, it was a "surrender." In reality, it was Apple buying the time to rebuild the company. Eventually not only they survived they bounced back to become the richest company on the planet. Safari browser which was built much later became the killer app for the iPhone launch. From an architectural and business standpoint, here is why this is a massive win for Apple and a structural threat to OpenAI: 1. The "Google Maps" Strategy (Rent vs. Own) Apple is "renting" Google’s LLM infrastructure today just as they rented Google Maps in 2007. They get the best tech on day one, satisfy their 2 billion users, and avoid the "Siri is dumb" narrative—all while they quietly build their own vertically integrated, silicon-optimized models. 2. Liability Abstraction: . The training of frontier models is a legal minefield of copyright and public data misuse. By using Gemini as the foundation, Apple offloads the legal liability to Google. Google takes the heat for how the data was scraped, while Apple sits a layer above, and they can apply on-device semantic indexing, private cloud compute and app intents to differentiate their offering from vanilla Gemini. 3. The Commoditization of the LLM: This deal is a blow to OpenAI. LLMs are quickly commoditizing and the winners in this space are the ones with the best distribution (Apple has 2 billion+ devices), best ecosystem, and the best UX layer. Above brand loyalty and trust. The Bottom Line: OpenAI is in a dangerous spot. In tech, the winner isn't the one who builds the greatest engine—it’s the one who builds the car that everyone drives.

Apple is integrating Google’s Gemini AI technology into the next major Siri upgrade. Apple just made the most strategic AI decision of the decade — and it signals a paradigm shift few are talking about. Apple has selected Google’s Gemini to power the next generation of Siri — It’s a masterclass in strategic alignment. Here’s why this move matters at the system level: 1) AI is now infrastructure, not feature. The frontier models with true reasoning, planning, long context, and multi-modal capacity are beyond the scope of a single product team. They require national-scale compute, global data flows, and continuous iteration. That’s Google’s game. 2) Apple has always been a systems company. They don’t invent every component. They compose them. • They didn’t invent CPUs — they curated best-in-class silicon partners, then disrupted them. • They didn’t invent wireless modems — same playbook. • Now: intelligence is a core component in modern computing — but not the whole system. Apple builds the experience, trust, and privacy envelope around it. 3) This is pragmatic over pride. Building a trillion-parameter model — with real reasoning — isn’t a 6-month sprint. It’s a multi-year, multi-billion dollar industrial program. Apple choosing to begin with a proven partner is the same logic that made the iPhone successful: pick the best building blocks — ship — then improve. 4) Why Google’s Gemini? Because at this stage of the game, leadership in AI is defined by systems that are: • multilingual • dynamic in reasoning • scalable for automation • tested in production at scale Not just “big parameter counts.” 5) Hybrid AI is the future — and Apple just bet on it. No single company will own every layer of the AI stack. The winning architecture isn’t a monolith — it’s an ecosystem. This means: → Best-in-class models from the infrastructure tier → Differentiated experiences at the UX layer → Privacy, trust, and distribution as first-class principles → Product companies composing AI ecosystems, not owning them end-to-end That’s the playbook for the next decade. And Apple — the world’s most disciplined integrator — just signaled how it sees the future. Not by shouting about who has the biggest model, but by asking: Who can deliver useful intelligence at scale — today? Silicon Valley should pay attention. This isn’t just a deal — it’s a declaration.