Is Apple ready to move at the speed of AI?

On July 30, 2026, Tim Cook sat down for his last earnings call as Apple’s CEO. In the same hour, Apple’s stock pushed the company’s market cap past $5 trillion, edging out Nvidia to reclaim the title of world’s most valuable company. Cook also admitted, almost in passing, that Apple still doesn’t have a clear plan for what heavy Siri AI usage will cost the company to run, and that memory chip prices are going through what he called a “100-year flood,” driven almost entirely by AI data-center demand.

That’s Apple’s AI story in one sitting: record-breaking, expensive, unresolved, and still not finished — nearly two and a half years after the company first showed the world “Apple Intelligence.”

If you’ve been asking whether Apple can actually keep pace with an industry that resets its own state of the art every few months, you’re asking the right question at the right time. The honest answer isn’t yes or no. It’s that Apple is running two different races at once, and it’s only winning one of them so far.

The short version

• Apple’s headline AI product — a rebuilt, conversational Siri — was promised in June 2024, delayed at least four separate times, and still isn’t fully shipped as of this writing. Full functionality isn’t expected until iOS 27 in September 2026, and even that rollout has slipped before.
• Apple settled a $250 million class-action lawsuit in May 2026 over marketing Siri features that didn’t exist yet at launch.
• Apple’s 2026 AI-related capital spending is a small fraction of what Microsoft, Google, Amazon, and Meta are each spending — Apple is projected to spend roughly $11 billion this year against rivals spending $120 billion to $200 billion apiece.
• Despite that, Apple hit a $5 trillion market cap in July 2026, its stock up around 25% on the year, outperforming every other AI megacap.
• Apple’s AI leadership has been almost completely rebuilt in the past 18 months: former AI chief John Giannandrea is gone, Craig Federighi now owns AI strategy, Microsoft/Google veteran Amar Subramanya was hired to lead AI research, and Tim Cook himself is stepping down as CEO on September 1, 2026, handing the company to hardware chief John Ternus.
• The new Siri runs on a hybrid model: Apple’s own on-device systems for fast, private tasks, plus a custom version of Google’s Gemini for the heavy reasoning, under a deal reportedly worth about $1 billion a year to Google.

None of these facts alone settles the argument. Together, they describe a company betting its next decade on being late but right, rather than first but wrong — and the jury genuinely hasn’t reached a verdict.

How Apple actually fell behind

It’s worth being specific about what “behind” means here, because the phrase gets thrown around loosely. Apple didn’t fail to notice generative AI. It announced Apple Intelligence at WWDC 2024 with a promotional blitz that included an iPhone 16 ad showing Siri reading a user’s messages, pulling up contacts from an email, and acting on personal context — capabilities that, as it turned out, weren’t close to ready.

Those exact features are what triggered the lawsuit. Plaintiffs argued Apple advertised a product it knew wasn’t functional, and Apple agreed to pay out roughly $95 per eligible iPhone owner rather than fight it in court.

The delay pattern that followed reads like a slow-motion pileup:

Two full years is a long time in any product category. In AI specifically, where the field moved from GPT-3.5-level chatbots to genuinely agentic systems that can browse, code, and execute multi-step tasks over that same stretch, two years is close to a full technological generation.

Why the delays kept happening

Bloomberg’s Mark Gurman, who has tracked the Siri rebuild more closely than almost anyone, has reported a fairly consistent set of internal problems: the assistant sometimes failed to process queries correctly, took too long to respond, struggled with fast talkers, and would unpredictably route requests to ChatGPT instead of Apple’s own systems. The underlying architecture — internally code-named Linwood, with a separate chatbot layer called Campo — had to reconcile Apple’s smaller on-device models with a much larger cloud model, and getting that handoff to feel instantaneous and reliable turned out to be much harder than Apple’s original 2024 timeline assumed.

There’s a structural reason for that difficulty that doesn’t get discussed enough: Apple wasn’t just building a chatbot. It was trying to build a chatbot that can act — reading your texts, calling your contacts, moving your photos, editing your calendar — while keeping that data off Apple’s own servers whenever possible and never letting Google use it to train Gemini. That’s a much harder engineering problem than a general-purpose assistant that assumes cloud access by default. Apple picked the more constrained, more defensible version of the feature and paid for that choice in time.

Apple’s own operating system release notes reflect this staged approach. macOS Tahoe 26 was the first release to bake Apple Intelligence system-wide, and the follow-up macOS Golden Gate 27 is where the fully rebuilt Siri AI app finally lands on the Mac, alongside the end of Intel support entirely.

The money question: is Apple even trying to keep up?

Here’s where the picture gets genuinely interesting, because the numbers don’t support a simple “Apple is cheap” narrative — they support something closer to “Apple made a different bet entirely.”

What the hyperscalers are spending in 2026

• Amazon: roughly $200 billion in capital expenditure, mostly AWS data centers and custom Trainium chips
• Alphabet (Google): $175–190 billion
• Microsoft: roughly $120–190 billion depending on how leases are counted
• Meta: $115–145 billion, revised upward mid-year on rising memory costs
• Combined big-four total: around $725 billion for the year, up 77% from 2025

What Apple is spending

Analysts at FactSet estimate Apple’s total 2026 AI-related capex at just over $11 billion — roughly what Microsoft alone spends in a single month. Apple’s research and development budget, by contrast, hit $34 billion through the first nine months of fiscal 2026, about five times its capital spending. That ratio is unusual in Big Tech’s AI buildout: everyone else is racing to own compute; Apple is racing to own the software layer that sits on top of chips it’s already shipping in a billion-plus devices.

There’s a real strategic logic here, and it’s not just spin. Apple doesn’t need to train a frontier model from scratch if it can license the reasoning layer from a partner and focus its own spending on the neural engines, the Private Cloud Compute infrastructure, and the interface work that makes AI feel native to an iPhone instead of bolted onto one. Baird analysts put it plainly in a note to investors this year: markets that initially punished Apple for skipping the LLM investment cycle have started rewarding the company’s industry-leading free cash flow instead. Apple’s $130-plus billion in cash and marketable securities gives it room to acquire, rather than build, if a specific capability gap becomes urgent.

The tradeoff: Apple is now dependent on Google for the actual intelligence behind its flagship AI feature. If Gemini’s quality plateaus, or if the Apple-Google agreement becomes a regulatory target — and given the ongoing antitrust scrutiny of Google’s default-search deal with Apple, that’s not hypothetical — Apple’s AI roadmap inherits someone else’s risk.

The leadership reset nobody’s talking about enough

While the Siri saga dominated headlines, Apple quietly rebuilt its entire AI chain of command.

John Giannandrea, the former Google AI lead Apple hired in 2018 specifically to fix this kind of problem, lost oversight of Siri, robotics, and core AI teams back in March 2025 after Apple Intelligence’s rocky launch. Apple made his exit official in December 2025 and he formally left the company in April 2026, after staying on payroll long enough for his final stock grants to vest — what insiders call “resting and vesting.” His responsibilities were split three ways, between software chief Craig Federighi, services head Eddy Cue, and operations chief Sabih Khan, with Federighi taking the largest share of AI strategy.

Apple then hired Amar Subramanya — previously a corporate VP of AI at Microsoft and, before that, head of engineering for Google’s Gemini Assistant — as VP of AI reporting directly to Federighi. That’s a notable hire: Apple pulled someone out of the team that built the exact product it’s now licensing.

And then, in April 2026, Apple announced the biggest change of all: Tim Cook is stepping down. John Ternus, Apple’s hardware engineering chief since 2013, becomes CEO on September 1, 2026, with Cook moving to executive chairman. Ternus inherits a company that just posted record revenue and a record stock price, but also a Siri overhaul that’s still not finished and a memory-cost crunch squeezing margins on every product line. Whether Apple can “move at the speed of AI” over the next few years may have less to do with any feature ship date and more to do with how a hardware-engineering CEO — as opposed to Cook’s operations-and-supply-chain background — reorganizes a software-and-AI-driven roadmap.

The bull case: slow and expensive isn’t the same as behind and irrelevant

It’s tempting to read all of this as a company caught flat-footed. The counterargument, made increasingly by Wall Street analysts through 2026, deserves real weight.

Distribution is Apple’s actual moat. OpenAI, Google, and Anthropic all have to convince people to download an app or change a habit. Apple ships AI directly into the operating system running on well over a billion active devices. When Siri AI works, it doesn’t need marketing — it needs an iOS update.

On-device processing is a real cost and privacy advantage, not just a talking point. In his final earnings call, Cook specifically called Apple’s ability to handle a meaningful share of AI requests on-device “a competitive weapon.” Every query Apple can answer using the Neural Engine in your pocket is a query it doesn’t have to pay cloud compute costs for — costs that are currently squeezing every one of Apple’s AI-native competitors’ margins.

Apple avoided the capex trap that’s now spooking investors elsewhere. In the same week Apple hit its $5 trillion valuation, Alphabet’s stock dropped 7% after it raised its 2026 capex guidance above $200 billion and reported negative free cash flow for the first time in company history. Meta took a similar hit. Apple’s restraint, which looked like timidity in 2024 and 2025, started looking like discipline once the market began questioning whether hyperscaler AI spending can actually generate a return.

The cash position gives Apple options nobody else has. With over $130 billion in liquid reserves and comparatively low AI capex, Apple can acquire specific capabilities — a model, a research team, a chip design — rather than build everything from first principles under time pressure.

The bear case: options aren’t outcomes

The counter-counterargument is just as real, and it starts with a simple observation: talk of strategic patience is much easier to make after the fact than it is to plan in advance. Apple’s restraint through 2024 and 2025 wasn’t originally framed internally as a deliberate contrarian bet — contemporaneous reporting describes a company that was genuinely caught off guard by how fast generative AI moved, then spent nearly two years trying to catch up to a promise it had already made publicly.

A few specific risks are worth naming directly:

• Apple is now structurally dependent on a competitor. Google is simultaneously Apple’s AI supplier and one of its most direct rivals in mobile and search. That arrangement works fine until their interests diverge, or until regulators force a separation — which is a live possibility given ongoing antitrust litigation touching the Apple-Google relationship.
• “Good enough, eventually” has a shelf life. Siri’s core value proposition — that it’s baked into the device you already carry — only protects Apple for as long as users don’t develop the habit of reaching for ChatGPT or Gemini directly instead. Habit formation is exactly what competitors are optimizing for right now.
• Monetization is still an open question. Cook’s own comments in July 2026 revealed Apple hasn’t finalized how it will pay for heavy Siri AI usage, floating a possible iCloud+ tier for power users. Committing to a feature before knowing its unit economics is the same mistake that produced the original 2024 overpromise.
• The talent Apple is relying on to fix this came from the companies it’s racing against. Subramanya is excellent on paper, but hiring your rival’s former engineering lead to catch up to your rival is also an admission of where the underlying expertise currently sits.

What “speed” actually means for Apple, versus everyone else

Most coverage treats “AI speed” as a single axis — who ships the smartest model fastest. That’s the wrong frame for a company like Apple, which isn’t primarily in the model business. It’s more useful to break “speed” into the dimensions that actually matter for a hardware-and-platform company:

1. Model speed — how fast the underlying AI improves. Apple has explicitly opted out of this race by licensing Gemini rather than training a frontier model in-house. On this axis, Apple is intentionally not competing.
2. Shipping speed — how fast promised features actually reach users. This is where Apple has visibly struggled, with a two-year gap between announcement and functional delivery.
3. Distribution speed — how fast a new capability reaches scale once it does ship. This is Apple’s strongest axis: an iOS update reaches more devices in weeks than any competing AI app reaches in years.
4. Trust speed — how fast users are willing to hand over sensitive data and workflows to an AI system. Apple’s privacy architecture (Private Cloud Compute, on-device processing, contractual limits preventing Google from training on Siri data) is built to win this axis over time, even if it costs speed on the other three.

Judged only on model speed and shipping speed, the “Apple is behind” narrative is accurate and well-supported by the record. Judged on distribution and trust, Apple is arguably better positioned than any AI-native competitor, precisely because it never had to build a user base — it already has one.

What actually changes for you, the user

If you’re an iPhone or Mac owner rather than an investor, the practical timeline looks like this as of mid-2026:

• If you’re on iOS 26.4 or later: you already have Phase 1 of the new Siri — on-screen awareness (Siri can reference what’s currently visible on your display), better contextual understanding, and basic actions across apps, powered by a Gemini-assisted backend that Apple says Google cannot use to train its own models.
• Expected with iOS 27 (September 2026): the fuller conversational rebuild, multi-turn dialogue closer to ChatGPT or Gemini’s own apps, more complex multi-app task execution, and a dedicated Siri app for direct chat-style interaction — the same Siri AI app that ships with macOS Golden Gate 27 on the Mac side.
• Possible but unconfirmed: a paid tier through iCloud+ for users who want heavier AI usage, hinted at but not detailed by Cook in July 2026.
• A general note on device requirements: the more advanced features increasingly favor newer hardware with more capable Neural Engines; older devices may see reduced functionality or heavier reliance on cloud processing.

If Siri AI still feels underwhelming when it lands in your hands, you’re not imagining a gap between the marketing and the product — that gap is the entire reason Apple paid out a quarter-billion-dollar settlement.

In the meantime, if you want native AI tools on your Mac that already work today rather than waiting on Apple’s timeline, it’s worth browsing AllMacWorld’s AI Tools category — apps like Topaz Photo AI are already doing on-device AI image work independent of whatever Siri eventually ships.

Frequently asked questions

Is Apple actually behind in AI, or is that a myth?

Both things are true at once, depending on which axis you measure. Apple is meaningfully behind on shipping timelines and on building its own frontier models — its 2024 promises weren’t fulfilled until 2026, and even then only partially. Apple is not behind on distribution, hardware integration, or cash reserves, all of which matter for whether AI features actually reach and stick with users.

Why did Apple partner with Google instead of building its own model?

Reporting from The Information and others indicates that some Apple leaders concluded large language models are becoming commoditized — meaning the underlying model matters less than the product experience wrapped around it — making a costly in-house frontier model harder to justify than licensing one from a specialist. Apple reportedly pays Google around $1 billion a year for the arrangement.

Does Google get access to my Siri data through this partnership?

According to reporting on the deal’s terms, the agreement specifically prevents Google from using Apple users’ Siri interactions to train future Gemini models, and queries are processed through Apple’s own Private Cloud Compute infrastructure rather than Google’s standard servers.

Will Siri eventually cost extra?

Tim Cook indicated on Apple’s July 2026 earnings call that a paid tier for heavy Siri AI usage, likely bundled with iCloud+, is under consideration, though Apple hasn’t finalized pricing or scope.

How much is Apple actually spending on AI compared to Microsoft or Google?

Estimates for 2026 put Apple’s AI-related capital expenditure at roughly $11 billion, compared with $120–200 billion each for Microsoft, Google, Amazon, and Meta. Apple’s total R&D spend, which includes AI software work not classified as capex, is considerably higher.

Does the CEO change from Tim Cook to John Ternus signal a change in AI strategy?

Nothing has been announced publicly beyond the leadership transition itself, effective September 1, 2026. Ternus’s background is in hardware engineering rather than software or AI research specifically, which has led some analysts to speculate about closer integration between Apple silicon roadmaps and AI feature development, though this remains speculative.

The verdict, such as it is

Apple isn’t racing at the speed of AI. It’s running a different race, on a track it built for itself, and betting that the finish line everyone else is sprinting toward — the most capable standalone model — matters less in the long run than the finish line Apple has always competed on: the product people actually use every day without thinking about the technology underneath it.

That bet might be right. Apple’s stock performance in 2026 suggests investors increasingly think so. But “the market rewards patience” and “the product justifies the wait” are two different claims, and only the second one is decided by whether Siri AI, fully shipped in September 2026, actually feels as good as the version Apple first promised in June 2024. Two years and a quarter-billion-dollar settlement later, that’s still an open question — and it’s the only one that will actually determine whether Apple was ready, or just lucky that the rest of the industry’s spending started looking as risky as Apple’s caution once did.

Sources referenced: Apple Newsroom, MacRumors, Bloomberg (via Mark Gurman’s “Power On” reporting as cited by multiple outlets), CNBC, Fortune, TidBITS/Six Colors earnings transcript, The Information, AppleInsider, 9to5mac, TechCrunch, Benzinga, Tom’s Hardware, and NBC News. Figures on capital expenditure and market capitalization reflect publicly reported estimates current as of early August 2026 and are subject to change with subsequent earnings reports.