Apple’s ambitious overhaul of Siri is finally taking shape, and the early verdict is in: the new AI assistant doesn’t just talk a good game—it understands you on a deeply personal level, all without sending your data to the cloud. This week, the first notable hands-on test of the rebuilt Siri AI appeared in the macOS 27 Golden Gate developer beta, with The Verge’s Antonio G. Di Benedetto spending 24 hours using it on a next-generation M5 MacBook. The experience reveals an assistant that finally feels like the futuristic helper Apple long promised, and for Windows enthusiasts, it raises a pressing question: can Microsoft’s own Copilot+ vision keep pace?
The Rebuilt Siri: What’s New in macOS 27 Golden Gate?
Apple’s macOS 27, codenamed Golden Gate, marks a pivotal moment for the Mac. While the OS brings a raft of iterative improvements, the star of the show is a completely re-architected Siri AI that now leans heavily on on-device processing. Gone is the dependence on round-trip cloud requests for basic tasks; instead, Siri leverages the Neural Engine inside the M5 chip to perform natural language understanding, personal context awareness, and even complex automations locally. This means Siri can now process requests like “Find the spreadsheet I was editing last night and attach it to an email to Jenna” entirely on the device, without an internet connection.
Di Benedetto’s 24-hour test highlights just how transformative this shift is. Siri’s responses are instantaneous, often completing actions before the user finishes speaking. More importantly, the assistant now maintains a persistent understanding of the user’s digital life—calendar events, recent documents, messages, and even app-specific states. For example, if you ask Siri “What time is my flight tomorrow?” it can pull the information from your email or calendar without breaking context. If you then say “Book a car to get there,” it seamlessly acts on that understanding. This level of contextual continuity has long been the missing link in voice assistants, and Apple’s implementation appears to nail it.
Crucially, all of this happens on-device, a design choice that Apple says preserves privacy. The M5’s secure enclave ensures that sensitive personal data never leaves the Mac. This is a stark contrast to previous Siri iterations, which often required cloud-based processing and left users uneasy about what was being recorded and stored.
Why On-Device AI Matters for Privacy and Performance
On-device AI isn’t just a buzzword—it’s a fundamental rethinking of how personal computing should work. By keeping data local, Apple eliminates the latency and security risks associated with transmitting information to remote servers. In practice, Siri’s new responsiveness turns routine tasks from multi-second waits into near-instant completions. Dictation, message composition, and even complex multi-step commands now feel snappier than ever.
Microsoft has also recognized the value of on-device AI, evidenced by its Copilot+ PC initiative. Windows 11’s Recall feature, which takes periodic snapshots of your screen and makes them searchable via AI, relies entirely on the on-device NPU. Similarly, Cocreator in Paint and real-time live captions leverage local hardware acceleration. However, the full Copilot assistant—the one that integrates across Office apps and the operating system—still leans heavily on cloud services for most of its intelligence. This hybrid approach can lead to inconsistent experiences: some queries are lightning fast, while others stall as they wait for server round-trips.
Apple’s decision to move Siri entirely on-device for core functionality sets a new standard. It demonstrates that a capable, context-aware assistant doesn’t need to ship your data off to a data center. For privacy-conscious users, especially those in enterprise environments, this could be a decisive factor when choosing between Mac and Windows hardware.
The M5 Factor: Apple’s Custom Silicon Flexes Its AI Muscles
The engine behind Siri’s dramatic upgrade is the M5 chip, Apple’s latest system-on-chip whose Neural Engine is reportedly twice as fast as the one found in the M4 generation. While exact specifications remain under wraps until the macOS 27 public release, developers testing the Golden Gate beta describe a machine learning performance profile that rivals dedicated AI accelerators found in servers. The M5’s unified memory architecture also plays a role: Siri can access the same memory space as running applications, allowing it to index and query live data without the overhead of file system bottlenecks.
This hardware advantage means that the M5 MacBook used in The Verge’s test never felt bogged down by Siri’s constant background processing. In fact, Di Benedetto noted zero impact on battery life after a full day of heavy use—a critical metric for laptop users. Apple’s silicon team has seemingly achieved a balance that Windows OEMs have struggled to match. While Qualcomm’s Snapdragon X Elite and Intel’s Lunar Lake processors deliver impressive NPU performance, they must still contend with the overhead of legacy x86 compatibility and less-efficient software stacks. Microsoft’s own Windows 11, though optimized for hybrid AI workloads, doesn’t yet offer the same deep integration between hardware and software that Apple achieves with its vertical control.
Windows Copilot+ and the Battle for On-Device AI Supremacy
Microsoft’s Copilot+ branding is the company’s attempt to signal that a PC is ready for the AI age. Launched alongside Snapdragon X Elite-powered devices in 2024, these machines include a dedicated NPU capable of at least 40 trillion operations per second (TOPS). Windows 11 takes advantage of this with features like Recall, which creates a searchable timeline of everything you’ve done on your PC, and Cocreator, an AI image generation tool integrated into Paint. On paper, the Copilot+ experience mirrors Apple’s local-first philosophy, but in practice, the gap is still wide.
Recall, for instance, stores its database entirely on-device and uses on-device AI for matching, yet it has been criticized for being less intuitive than Apple’s approach. Instead of understanding your intent like Siri does, Recall largely functions as a photographic memory—impressive, but not proactive. Copilot itself, the primary interface for AI in Windows, remains heavily cloud-dependent. Ask it to summarize a document or draft an email, and it will rush off to Azure’s servers, often introducing latency and raising the same privacy concerns that Apple sidesteps.
This fragmentation is typical of Microsoft’s broader AI strategy. While the company has made massive investments in OpenAI and cloud AI, its on-device efforts often feel like a secondary priority. For Windows users, the result is a disjointed experience: some tools work offline, most don’t, and there’s no single assistant that ties them together as elegantly as Siri now does on the Mac.
What Windows Users Can Learn from Apple’s Siri Overhaul
Despite the platform rivalry, there’s plenty for Windows enthusiasts to admire—and demand—in Apple’s latest move. The biggest takeaway is that true personal AI requires a relentless focus on integration. Siri in macOS 27 Golden Gate isn’t just a voice on top of the OS; it is woven into the fabric of every app and file. Because Apple controls the entire stack—from the M5 silicon to the macOS frameworks to the native apps—it can afford to build deep contextual understanding that third-party Windows developers can only dream of.
For Microsoft, the path forward should involve a similar commitment to a unified AI layer. Instead of scattering AI features across different apps and services, Windows needs a persistent assistant that can learn from what you do and anticipate your needs without user friction. The upcoming Windows 12 (or whatever Microsoft calls its next major update) might be the opportunity to introduce such an assistant—one that runs entirely on-device thanks to the next wave of NPU hardware from AMD, Intel, and Qualcomm.
Moreover, Apple’s privacy marketing could force Microsoft’s hand. As data breaches and surveillance concerns continue to mount, the ability to do sensitive tasks entirely offline becomes a competitive advantage. Microsoft has already shown with Recall that it understands the importance of local processing; now it needs to extend that principle to the entirety of Copilot. If it doesn’t, Apple will have a compelling story to tell every Windows user who values their privacy.
The Road Ahead: Will Microsoft Answer?
Competition breeds innovation, and Apple’s macOS 27 Golden Gate beta is a clear shot across Microsoft’s bow. With the M5 chip expected to power the next MacBook Air, MacBook Pro, and even a new Mac Studio, Apple’s on-device AI advantage could quickly scale to tens of millions of users. Microsoft, meanwhile, must navigate a more complex ecosystem where hardware capabilities vary wildly. The answer likely lies in the evolution of Windows Copilot to a fully local model for everyday tasks, reserving the cloud for only the most demanding creative or analytic queries.
Rumors already point to Microsoft working on a “local Copilot” that would use the NPU to handle email summarization, document Q&A, and even code completion offline. If true, this could level the playing field. But execution matters, and Apple has set a high bar with Siri’s seamless, always-ready intelligence.
For Windows users today, the choice isn’t black and white. Copilot+ PCs deliver genuine AI value right now, and features like live captions and studio effects are genuinely useful. Yet, they don’t offer a unified, context-aware assistant that makes the entire operating system feel smarter. That’s the dream Apple is selling with macOS 27 Golden Gate, and it’s one that Microsoft must chase aggressively.
One thing is certain: the on-device AI race is just heating up, and consumers stand to win. Whether you’re a die-hard Windows fan or an Apple devotee, the era of truly personal, private, and powerful AI assistants is finally dawning—and it’s happening first on the Mac.