Latest / Elon Musk Podcast / Tim Cook steps down, John Ternus and the new iPhone Air

Transcript

Tim Cook steps down, John Ternus and the new iPhone Air · Elon Musk Podcast

• 0:00
  Tim Cook is officially stepping down as CEO of Apple to become
• 0:03
  executive chairman, handing the top leadership role to their
• 0:06
  hardware chief John Ternis. Right.
• 0:09
  And that executive change happens exactly as Apple
• 0:13
  executes A fundamental internal software pivot.
• 0:16
  I mean they are freezing their stand alone browser answer
• 0:19
  engine projects completely. They want to completely rebuild
• 0:23
  Siri using a hybrid of in house models and Google Gemini.
• 0:27
  Yeah. So you have a hardware centric
• 0:28
  leadership team stepping in to reach shape the physical
• 0:30
  devices, but you know at the exact same time they're managing
• 0:33
  a complete overhaul of the artificial intelligence running
• 0:36
  inside those devices. Exactly.
• 0:38
  We are looking at how a hardware focused executive team is
• 0:41
  fundamentally altering the physical limits of personal
• 0:44
  devices and really how those extreme physical constraints are
• 0:47
  forcing A radical shift in how consumer artificial intelligence
• 0:50
  operates. Which brings us to the core
• 0:52
  question. How does a company obsessed with
• 0:55
  ultra thin minimalist hardware navigate the demanding,
• 0:59
  resources heavy reality of running autonomous AI systems?
• 1:02
  Well, to understand where the engineering is going, you have
• 1:05
  to look closely at John Turnus's background and you know his
• 1:08
  particular management style, right?
• 1:10
  He comes directly from the hardware engineering side.
• 1:13
  People describe his style as highly personable, but he
• 1:15
  demonstrates an intense level of decisiveness regarding product
• 1:19
  standards. Like during the lead up to the
• 1:21
  Vision Pro release, engineers uncovered A flaw causing
• 1:25
  wireless frequency audio latency between the headset and the
• 1:28
  Airpods. Oh wow.
• 1:30
  Yeah, and Turnus chose to completely delay a major feature
• 1:34
  just to fix that specific latency issue.
• 1:36
  I mean, think about the last time you wore any sort of
• 1:38
  virtual reality headset. If you turn your head and the
• 1:41
  audio hits your ears even a fraction of a second late, your
• 1:45
  brain immediately tells you something is wrong.
• 1:47
  Right. The illusion breaks completely.
• 1:49
  Yeah. And you often end up feeling
• 1:50
  physically nauseous because your visual inputs and auditory
• 1:54
  inputs are basically fighting each other.
• 1:56
  So he prioritized the absolute precision of that physical and
• 1:59
  sensory experience over hitting a predetermined release
• 2:02
  timeline. Protecting the physical hardware
• 2:05
  experience supersedes everything else, definitely, but the
• 2:08
  structural shifts extend well beyond Turnus taking the top
• 2:12
  job. Yeah, we are seeing a much
• 2:14
  broader executive shake up right alongside his promotion.
• 2:17
  Chief Operating Officer Jeff Williams is retiring.
• 2:20
  Eddie Q is taking over all health and fitness operations,
• 2:24
  and Craig Federighi is adding Watch OS to his software
• 2:26
  oversight responsibilities. Which really reads like a
• 2:29
  consolidation of control among a core group of veterans.
• 2:33
  I mean, Jeff Williams retiring alters the dynamic entirely.
• 2:37
  Operations and supply chain management are what historically
• 2:39
  made the company so incredibly profitable.
• 2:42
  Right. They mastered sourcing
• 2:43
  components and moving inventory at an unprecedented scale.
• 2:47
  And Eddie Q taking over health aligns perfectly with the
• 2:50
  reality that health tracking is entirely a data-driven recurring
• 2:54
  ServiceNow, Yeah. You pay a subscription and the
• 2:56
  servers analyze your biometric data.
• 2:58
  Exactly, and giving Federi Watch OS aligns all the personal
• 3:02
  computing operating systems under 1 unified architectural
• 3:04
  roof instead of you know, having the watch software developed in
• 3:08
  a separate silo. I do question the underlying
• 3:10
  logic of the core strategy here though.
• 3:12
  Placing a hardware purist at the very top of the hierarchy seems
• 3:16
  counterintuitive when the entire tech industry insists that
• 3:20
  software and artificial intelligence are the only
• 3:23
  components driving value right now.
• 3:25
  You think it feels out of step? Yeah, it feels remarkably
• 3:28
  similar to an automotive company promoting their lead chassis
• 3:31
  designer to CEO, just as the entire auto industry shifts its
• 3:35
  focus toward autonomous driving software.
• 3:37
  The mechanical shell seems far less critical than the digital
• 3:40
  brain running the vehicle. Well, Apple is placing A
• 3:43
  distinct bet that physical form factor still dictates consumer
• 3:47
  desire. They believe that if you hold an
• 3:50
  object and it feels incredible, you will buy it regardless of
• 3:53
  the prevailing software trends. Right.
• 3:55
  And that physical desire directly drives the engineering
• 3:57
  behind their newest flagship device, the iPhone Air.
• 4:01
  The strategy assumes that a compelling physical object buys
• 4:04
  you the high profit margin, and it buys you the time required to
• 4:07
  figure out the complexities of the software ecosystem.
• 4:10
  And the physical reality of the iPhone Air pushes engineering to
• 4:15
  extreme limits. We are looking at a device with
• 4:18
  a 6.5 inch display that measures just 5.6mm thick.
• 4:23
  Wow, 5.6mm. Yeah, think about a stack of
• 4:27
  maybe 5 standard credit cards. That is the entire physical
• 4:30
  depth of this phone. That actually makes it thinner
• 4:33
  than Samsung's 5.8mm Galaxy S25 Edge.
• 4:37
  OK, And to achieve this structural integrity without the
• 4:41
  phone simply snapping in half when you sit down with it in
• 4:43
  your pocket, they built it using a polished grade 5 titanium
• 4:46
  frame and they covered both sides with ceramic shield.
• 4:49
  Too, the camera plateau design is a direct result of that
• 4:52
  severe thinness constraint too. Physics dictates that optical
• 4:55
  lenses need a certain amount of physical depth to focus light
• 4:58
  onto a sensor. Right.
• 4:59
  You cannot simply compress A curved glass lens and expect it
• 5:02
  to bend light correctly. Exactly.
• 5:04
  So to save space, Apple designed a distinct raised plateau on the
• 5:07
  back of the device. They grouped the 48 megapixel
• 5:09
  rear lens, the 18 megapixel front lens, the face I true
• 5:13
  depth system, and a speaker into a single raised section on the
• 5:16
  chassis. Wait.
• 5:17
  Bit back U let me make sure I'm wrapping my head around this.
• 5:19
  Are you saying they put the front facing camera and the rear
• 5:22
  facing camera in the exact same physical module?
• 5:25
  Yeah, they essentially share the same physical Z axis space,
• 5:29
  pointing in opposite directions. By grouping the camera sensors
• 5:32
  back-to-back, the rest of the phone stays at that ultra thin
• 5:36
  5.6mm. Oh wow, so this single plateau
• 5:39
  houses all the deep components? Exactly, and they isolate the
• 5:43
  sensors internally so light entering the front lens does not
• 5:46
  bleed into the rear sensor. They found other extreme ways to
• 5:49
  shave off millimeters too, like they utilized a 3D printed
• 5:53
  titanium USBC port. OK, instead of standard
• 5:56
  machining. Right.
• 5:58
  Traditional forging or CNC machining requires taking a
• 6:01
  solid block of metal and carving it away with drill bits.
• 6:04
  When you do that, you have to leave relatively thick metal
• 6:07
  walls so the component maintains its structural integrity.
• 6:10
  So it doesn't just tear during the milling process.
• 6:12
  Yeah, but 3D printing titanium powder layer by layer, using
• 6:16
  high-powered lasers to fuse the particles allows for complex
• 6:20
  microscopic honeycomb structures.
• 6:22
  So you maintain all the strength of solid titanium, but you use a
• 6:26
  fraction of the space and a fraction of the material.
• 6:28
  Exactly. They also eliminated the SIM
• 6:31
  tray entirely, making the phone esim only worldwide.
• 6:35
  Removing the mechanical reader, the springs and the tray itself
• 6:38
  frees up vital cubic millimeters inside the casing.
• 6:41
  Stripping the internal volume down South aggressively creates
• 6:44
  a fascinating set of physical limits, though the phone only
• 6:47
  has room for a 3149 milliamp hour battery.
• 6:52
  Which is pretty tiny for a modern flagship.
• 6:54
  Yeah, driving a bright 6.5 inch OLED screen with that capacity
• 6:58
  provides roughly 27 hours of continuous video playback.
• 7:02
  Apple addresses this physical limitation by selling a custom
• 7:05
  $99 Magsafe battery pack to boost the total playback time up
• 7:09
  to 40 hours. The severe thinness also forces
• 7:12
  the complete removal of the dedicated telephoto and ultra
• 7:15
  wide lenses you usually see on flagship devices.
• 7:17
  Users have to rely solely on the single 48 megapixel fusion lens.
• 7:21
  Right. And that Fusion lens works by
• 7:23
  taking a massive, highly detailed 48 megapixel image and
• 7:26
  digitally cropping directly into the center 12 megapixels.
• 7:30
  So you simulate an optical zoom with an actual actually needing
• 7:32
  a long physical telephoto barrel protruding from the back of the
• 7:35
  phone. When you strip the hardware down
• 7:37
  to its absolute physical limits, you force extreme efficiencies
• 7:41
  internally. You cannot rely on brute force
• 7:44
  processing power because you lack the thermal mass to cool a
• 7:48
  hot computer chip. And you lack the battery
• 7:50
  capacity to run it at maximum power for very long, which leads
• 7:53
  directly into their custom silicon and the entire AI
• 7:57
  overhaul. Right.
• 7:58
  Apple initiated an abrupt software pivot.
• 8:01
  They completely froze their separate Safari AI answer engine
• 8:04
  projects to concentrate entirely on overhauling Siri.
• 8:08
  Yeah, they are building a hybrid system.
• 8:10
  They are running an in house 3 billion parameter model for
• 8:13
  sensitive local tasks directly on the device itself.
• 8:16
  While you know, partnering with Google Gemini to handle complex
• 8:19
  world knowledge in the cloud. So what exactly is a parameter
• 8:22
  in this context? A parameter is essentially A
• 8:24
  variable the AI learned during its training phase.
• 8:27
  It represents a connection between concepts.
• 8:29
  A 3 billion parameter model is remarkably small in the AI
• 8:32
  space. Meaning it fits on the phone
• 8:34
  easily. Exactly.
• 8:36
  The entire model can load directly into the iPhones 12
• 8:39
  gigabytes of RAM instantly. Because it lives in the RAM,
• 8:43
  there is 0 cloud latency. So it is perfect for reading
• 8:47
  your private text messages or summarizing long emails or
• 8:50
  executing local commands without constantly engaging the radio
• 8:54
  antennas and draining the battery.
• 8:55
  And Gemini handles the heavy lifting, like explaining quantum
• 8:58
  mechanics or generating blocks of programming code.
• 9:01
  The hardware routing those specific cloud requests to
• 9:04
  Google is highly specialized too.
• 9:06
  The new Apple Design C1X modem handles the data transfer.
• 9:10
  Right. It operates faster than on their
• 9:12
  previous versions, but it is explicitly limited to sub six
• 9:15
  gigahertz 5G without any millim wave capabilities.
• 9:19
  And that particular limitation ties right back to the physical
• 9:21
  hardware design. Millim Wave antennas provide
• 9:24
  incredible Gigabit level download speeds in dense
• 9:26
  environments like stadiums or downtown urban centers.
• 9:29
  The problem is that the waves are so short they cannot pass
• 9:31
  through human hands or walls easily, so engineers have to put
• 9:34
  multiple bulky antennas along the outer edges of the phone.
• 9:38
  And they draw massive amounts of power.
• 9:40
  Exactly 6 gigahertz 5G travels much further, penetrates
• 9:45
  concrete walls easily, and requires fewer, significantly
• 9:48
  smaller antennas. Cutting aluminum wave entirely
• 9:52
  keeps the phone incredibly thin and protects the smaller
• 9:55
  battery. It is basically like you have a
• 9:57
  secure underground vault for your private diary that is the
• 10:01
  on device AI and right next to it there's a red telephone that
• 10:05
  dials straight to the Library of Congress for everything else.
• 10:08
  That is a great way to think about it You.
• 10:09
  Keep your most sensitive personal data.
• 10:11
  Totally be private in the local vault, but you still get
• 10:15
  infinite knowledge on demand whenever you pick up the phone
• 10:17
  to call the library. Relying heavily on that Public
• 10:20
  Library completely alters Apple's structural position in
• 10:23
  the broader tech market, though It shifts them away from
• 10:26
  operating a totally closed, tightly controlled ecosystem.
• 10:29
  Right. It forces a heavy reliance on a
• 10:31
  direct competitor's infrastructure.
• 10:32
  Yeah, and this exposes Apple to unpredictable massive usage
• 10:36
  costs. Every time a user makes a
• 10:38
  complex query, that data routes through Gemini.
• 10:41
  And running those those queries on Google servers burns
• 10:43
  electricity and utilizes incredibly expensive graphics
• 10:47
  processing units. Industry analysts estimate those
• 10:50
  specific usage costs could potentially reach a billion
• 10:53
  dollars. It also invites intense
• 10:55
  regulatory scrutiny. Global governments are already
• 10:59
  extremely sensitive to monopolistic search engine
• 11:01
  deals, and they will scrutinize these AI routing partnerships
• 11:05
  just as closely. But fetching facts from Google
• 11:08
  only solves half the problem. For the user to be genuinely
• 11:11
  valuable, the intelligence has to execute complex commands
• 11:15
  across different apps on the device itself.
• 11:18
  Right, which requires a completely different engineering
• 11:20
  approach than just answering trivia questions.
• 11:22
  That requirement brings us to Agentic AI.
• 11:25
  Raw large language models are incredibly powerful text
• 11:28
  generators, but they are essentially directionless.
• 11:31
  Yeah, they operate by simply predicting the next most likely
• 11:34
  word in a sequence based on their massive training data.
• 11:37
  Because of this predictive nature, they frequently
• 11:39
  hallucinate facts or just stop reasoning before successfully
• 11:42
  completing a multi step task. So agentic engineering
• 11:45
  completely changes the dynamic. Engineers focus on designing
• 11:49
  software systems that can reason through a problem, take an
• 11:52
  action, observe the results of that specific action, and
• 11:55
  recover gracefully if the action fails.
• 11:58
  You see this implemented through the React pattern, which stands
• 12:00
  for Reason and ACT engineers build a continuous control loop.
• 12:05
  Give me an example of how that loop works for the user.
• 12:08
  Well, imagine you get a text message saying a flight is
• 12:10
  delayed and you need to push a dinner reservation.
• 12:13
  The AI thinks the user wants to move a meeting.
• 12:16
  It acts by checking the Calendar application programming
• 12:19
  interface. The API.
• 12:21
  Yes, it observes the result, seeing a scheduling conflict at
• 12:25
  the requested time. Then it reasons about the next
• 12:27
  logical step, scanning the calendar to find the next
• 12:30
  available open slot. And to make this loop function
• 12:32
  reliably, developers utilize structured outputs.
• 12:36
  They force the AI to output its decisions in strict Jason data
• 12:39
  formats. Right, which prevents downstream
• 12:41
  system failures, ensures the AI can actually trigger app
• 12:44
  functions reliably without crashing the software.
• 12:47
  Hold on. So instead of the AI just
• 12:49
  talking to you conversationally, it is actually generating
• 12:52
  strings of code and talking directly to the phone's
• 12:54
  operating system. Exactly.
• 12:56
  The intelligence operates as a translation layer between human
• 13:00
  intent and machine execution. The operating system only
• 13:03
  understands strict syntax and exact formatting.
• 13:06
  So if the calendar system expects a very specific data
• 13:09
  structure to move an appointment, the AI must output
• 13:12
  perfectly formatted code, not conversational English.
• 13:16
  Right, but I see a massive vulnerability in Apple specific
• 13:20
  strategy here. They are licensing the
• 13:22
  underlying intelligence model from Gemini while building the
• 13:25
  agent harness themselves. The harness being the workflow
• 13:28
  control system that manages the Jason outputs, yeah.
• 13:31
  Competitors deeply involved in this space have found that the
• 13:34
  intelligence models and the agent harnesses need to be
• 13:36
  tightly integrated, not modular. Why does separating the
• 13:40
  intelligence from the control control harness creates such a
• 13:43
  high risk of failure? Because the integration is
• 13:45
  incredibly fragile, think about how models are updated.
• 13:49
  If Google tweaks Gemini's underlying weights to make the
• 13:52
  model slightly more helpful or chatty, the model might start
• 13:55
  outputting conversational filler alongside the required code.
• 14:00
  It might generate something like sure, I can help you move that
• 14:02
  meeting, followed by the strict Jason structure.
• 14:05
  That extra conversational text instantly breaks Apple's rigid
• 14:09
  Jason parser. Because the operating system
• 14:12
  expects an open bracket, receives the letter S, and the
• 14:15
  harness fails entirely, the user just experiences a broken
• 14:19
  feature. Exactly.
• 14:20
  When you separate the brain from the nervous system, you
• 14:23
  introduce points of failure you absolutely cannot control.
• 14:26
  This friction opens up entirely new possibilities for the future
• 14:29
  of Apple's silicon development, though.
• 14:31
  Running these continuous control loops locally, constantly
• 14:34
  monitoring the screen and evaluating data will demand
• 14:37
  massive amounts of power. Right, it sets the stage
• 14:40
  perfectly for the rumored A20 Pro chip.
• 14:43
  By moving to A2 nanometer manufacturing process, they can
• 14:46
  handle these complex agentic workflows efficiently in the
• 14:49
  next iteration of the hardware. The manufacturing scale is
• 14:52
  critical to making all of this work inside of thin chassis.
• 14:56
  Nanometers measure the gate length of a single transistor on
• 14:58
  the chip. And shorter gates mean electrons
• 15:01
  a much shorter physical distance.
• 15:03
  When electrons travel shorter distances, they move faster and
• 15:06
  generate significantly less heat.
• 15:08
  Which is crucial thermal efficiency.
• 15:10
  You have an AI agent constantly looping in the background,
• 15:13
  reading your screen, checking your calendar and organizing
• 15:16
  your data. You need that chip to process
• 15:18
  the information without melting a 5.6mm titanium phone in your
• 15:23
  hand. So Apple is betting its entire
• 15:25
  future on a high stakes combination.
• 15:27
  They are pushing physical hardware to extreme minimalist
• 15:31
  limits. It's under John Turnus,
• 15:32
  utilizing materials like 3D printed titanium to save
• 15:36
  fractions of a millimeter. Yeah, while simultaneously
• 15:38
  transforming Siri into a highly complex hybrid agentic AI system
• 15:43
  that requires extreme computing efficiency to function.
• 15:46
  It really makes you wonder, will average consumers actually care
• 15:49
  about having an autonomous agent managing their calendar
• 15:51
  workflows, or will they simply buy the device because it is the
• 15:55
  thinnest piece of titanium available on the market?
• 15:57
  If you're not subscribed yet, take a second and hit follow on
• 16:00
  whatever app you're using. It helps us keep making this.
• 16:03
  We appreciate you being here. Also, check out our YouTube
• 16:06
  channel for more business and tech updates.
• 16:08
  There's a link in the description.