Skip to main content
a16z Podcast

Steven Sinofsky on Apple at 50, Microsoft, and the Future of Computing

29 min episode · 2 min read
·

Episode

29 min

Read time

2 min

Topics

Artificial Intelligence, Software Development, Crypto & Web3

AI-Generated Summary

Key Takeaways

  • Token cost drives hardware shift: AI compute currently billed per token creates a cost ceiling that historically forces resources onto local devices. Every prior computing constraint — DRAM, processing power, storage — followed this same pattern: pay-per-use on remote infrastructure eventually migrates to free on-device. Expect AI inference to follow within 6–9 months as models shrink and local chips improve.
  • NVIDIA RTX Spark architecture: The RTX Spark chip combines an ARM CPU with NVIDIA parallel GPU processing into a unified system-on-chip with a new memory architecture. This targets PC manufacturers directly and enables local AI model inference without cloud token costs. The key unknown is whether CUDA APIs will be preinstalled, OS-integrated, or downloadable — Microsoft has not specified publicly.
  • 16GB RAM minimum for Windows AI devices: Current Windows machines require deliberate optimization — uninstalling software, registry edits — to run adequately on 8GB RAM. Sinofsky recommends 16GB as the baseline for any new PC purchase today. The Dell XPS 13 starting at 8GB is flagged as insufficient, while the MacBook Neo at $499–$599 offers a more capable baseline configuration.
  • Backward compatibility as strategic trap: Microsoft's decision to support all legacy Win32 applications on ARM-based NVIDIA Spark devices repeats a pattern Sinofsky argues undermines platform advancement. Consumers do not actually want registry access, legacy app compatibility, or fan-cooled hardware — they want sealed, stable systems like phones and Macs. Enterprise legacy app needs can be addressed via VMs or remote servers instead.
  • Apple's WWDC API decision is the pivotal moment: The critical near-term question is whether Apple will natively support CUDA APIs in its upcoming WWDC announcements. Options range from native OS integration to App Store distribution to a translation layer. Apple's choice determines whether its hardware — particularly iPhones — can run optimized open-source AI models locally, a capability currently limited to Mac mini stacks running headless agents.

What It Covers

Steven Sinofsky, former Windows division president at Microsoft, analyzes NVIDIA's RTX Spark chip announcement at Computex 2025, the shift toward on-device AI compute, Apple versus Microsoft platform strategy, and why backward compatibility decisions made today will define the next era of personal computing hardware.

Key Questions Answered

  • Token cost drives hardware shift: AI compute currently billed per token creates a cost ceiling that historically forces resources onto local devices. Every prior computing constraint — DRAM, processing power, storage — followed this same pattern: pay-per-use on remote infrastructure eventually migrates to free on-device. Expect AI inference to follow within 6–9 months as models shrink and local chips improve.
  • NVIDIA RTX Spark architecture: The RTX Spark chip combines an ARM CPU with NVIDIA parallel GPU processing into a unified system-on-chip with a new memory architecture. This targets PC manufacturers directly and enables local AI model inference without cloud token costs. The key unknown is whether CUDA APIs will be preinstalled, OS-integrated, or downloadable — Microsoft has not specified publicly.
  • 16GB RAM minimum for Windows AI devices: Current Windows machines require deliberate optimization — uninstalling software, registry edits — to run adequately on 8GB RAM. Sinofsky recommends 16GB as the baseline for any new PC purchase today. The Dell XPS 13 starting at 8GB is flagged as insufficient, while the MacBook Neo at $499–$599 offers a more capable baseline configuration.
  • Backward compatibility as strategic trap: Microsoft's decision to support all legacy Win32 applications on ARM-based NVIDIA Spark devices repeats a pattern Sinofsky argues undermines platform advancement. Consumers do not actually want registry access, legacy app compatibility, or fan-cooled hardware — they want sealed, stable systems like phones and Macs. Enterprise legacy app needs can be addressed via VMs or remote servers instead.
  • Apple's WWDC API decision is the pivotal moment: The critical near-term question is whether Apple will natively support CUDA APIs in its upcoming WWDC announcements. Options range from native OS integration to App Store distribution to a translation layer. Apple's choice determines whether its hardware — particularly iPhones — can run optimized open-source AI models locally, a capability currently limited to Mac mini stacks running headless agents.

Notable Moment

Sinofsky revealed that when he originally designed Surface in 2011, the ARM-based tablet was intentionally meant to break backward compatibility and force a new OS API ecosystem. Microsoft overruled this, spent eight years reverting to Intel x86, and is now repeating the same backward-compatible mistake with NVIDIA Spark.

Know someone who'd find this useful?

Episode Transcript

Having lived through, like, a half dozen component shortage things, you just sort of wait them out, and you let some local max or local min determine the future. This will correct itself in short order. This world where you're all gated on dollars per token is a thing that's gonna move to your own device, which is exactly what happened with all of computing. Anytime there's a resource constraint that you have to pay for, it moves to your device and becomes free. AI introduces yet another opportunity to change that dynamic for the PC to have it be forward looking, not backward looking. And I think this is incredibly important opportunity for Microsoft and for the industry as a whole. Few people have had a front row seat to the personal computing revolution quite like Steven Sinofsky. Over nearly three decades at Microsoft, he helped shape products that defined the PC era, including Windows, Office, and Surface. Along the way, he also witnessed one of the technology industry's longest running rivalries, Microsoft and Apple. As Apple celebrates its fiftieth anniversary, questions about product design, platforms, hardware, software, and the future of computing remain as relevant as ever. Theo Jaffe speaks with Steven Sinofsky about Apple, Microsoft, and the evolution of personal computing. I'm in the situation room with Steven Sinovsky who might have been, like, the first ever guest on MTS back when we were still doing test streams. I think he was he was the first person I interviewed, on a test stream. He was the president of the Windows division at Microsoft. He created the Surface program at Microsoft, which we have some very interesting news about today. We're thrilled to have you on. Steven, welcome to MDS. Welcome back. Well, thanks so much. Good to see you. So Hi, everyone. Yeah. Hi. First question would be, NVIDIA and Microsoft and Arm and a few other companies just announced, something very interesting at Computex. What exactly did they announce, and what does it matter? Sure. Well, just so folks know because, it hasn't it doesn't go get in the news much, but, Computex is this big giant trade show in Taiwan. And it's it's the weirdest show because it's, like, this total inside baseball, you know, silicon supply chain show. And normally, you never hear about it. Like, in fact, I I never went to it even. I wanted to. Well, it actually turns out it was always right around the same time as a a big Microsoft sales meeting, so I never went. But you could think of it as the the ecosystem show for everything it takes to to build a a computing device of any kind. Totally well, but Jensen in his keynote last night did this incredible slide where he walked up and down the whole length of the stage pointing to partners that he was very excited to be there. And I would bet anyone that anyone watching would have no …

Get the full transcript (4,873 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all a16z Podcast transcripts →

You just read a 3-minute summary of a 26-minute episode.

Get a16z Podcast summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free

Keep Reading

Books, tools, and gear mentioned in this episode

SignalCast may earn commission on purchases via these links. As an Amazon Associate, SignalCast earns from qualifying purchases.

Gear

  • by Apple

    Apple's choice determines whether its hardware — particularly iPhones — can run optimized open-source AI models locally, a capability currently limited to Mac mini stacks running headless agents.
  • MacBook NeoRecommended

    by Apple

    The Dell XPS 13 starting at 8GB is flagged as insufficient, while the MacBook Neo at $499–$599 offers a more capable baseline configuration.
  • by Dell

    The Dell XPS 13 starting at 8GB is flagged as insufficient, while the MacBook Neo at $499–$599 offers a more capable baseline configuration.
  • by NVIDIA

    Steven Sinofsky, former Windows division president at Microsoft, analyzes NVIDIA's RTX Spark chip announcement at Computex 2025, the shift toward on-device AI compute... The RTX Spark chip combines an ARM CPU with NVIDIA parallel GPU processing into a unified system-on-chip with a new memory architecture.
  • by Microsoft

    Sinofsky revealed that when he originally designed Surface in 2011, the ARM-based tablet was intentionally meant to break backward compatibility and force a new OS API ecosystem.

More from a16z Podcast

We summarize every new episode. Want them in your inbox?

Similar Episodes

Related episodes from other podcasts

Explore Related Topics

This podcast is featured in Best Business Podcasts (2026) — ranked and reviewed with AI summaries.

Read this week's AI & Machine Learning Podcast Insights — cross-podcast analysis updated weekly.

You're clearly into a16z Podcast.

Every Monday, we deliver AI summaries of the latest episodes from a16z Podcast and 192+ other podcasts. Free for one show.

Start My Monday Digest

No credit card · Unsubscribe anytime