Latest / Connecting the Dots / AI-Powered Threats, Chip Delays, and Apple's Foldable Future

Transcript

AI-Powered Threats, Chip Delays, and Apple's Foldable Future · Connecting the Dots

0:00
Today's tech news moves fast, but understanding
0:03
it shouldn't be hard. That's why Snarfl Solutions
0:06
Group brings you Connecting the Dots, where we
0:09
break down what's happening in tech and why it
0:11
matters to you. Now, let's dive in with your
0:14
hosts. What happens when artificial intelligence
0:18
gets, you know, so smart that it can actually
0:21
break into a server? Diagnose its own errors
0:24
and then rewrite its own malicious code in exactly
0:27
31 seconds. Oh, and don't forget the best part.
0:30
While humans literally cannot build the physical
0:32
metal boxes required to plug that same AI into
0:35
the wall. Right. Welcome to Connecting the Dots.
0:38
I'm Alex, your co -host, ready to break down
0:40
today's tech news and its everyday impact on
0:43
you. And I'm Morgan, bringing you the analysis
0:45
you need. We're coming to you live from Snarful
0:48
Solutions Group right here in Sacramento, California.
0:51
Yeah. And today is Monday, July 6th, 2026. But
0:56
before we look closer at the digital world, let's
0:58
get you caught up on the physical one. Always
1:00
a good idea. So looking out the window here in
1:02
Sacramento, we actually have a gorgeous sunny
1:04
day. It's about 88 degrees with a nice little
1:06
breeze blowing at five to 12 miles per hour.
1:10
I'd say enjoy that sunshine while you can, honestly,
1:13
because nationally the map is incredibly turbulent
1:15
right now. We are currently tracking 235 active
1:19
U .S. weather alerts across the country. And
1:22
92 of those are sitting directly over land. So
1:24
if your day involves any domestic travel, you'll
1:26
definitely want to double check those flight
1:28
routes. That is a lot of alerts. And speaking
1:31
of turbulence, the financial markets had plenty
1:33
of their own heading into the weekend. Let's
1:34
hear it. So the Dow Jones closed out Thursday,
1:37
July 2nd, at 52 ,900 .07. That was a daily jump
1:41
of 504 .85 points, which is roughly a full percent.
1:45
Wow. Right. And zooming out to the weekly view,
1:48
the Dow was up a massive hand hundred ninety
1:50
six point 30 points. It's huge. But the S &P
1:53
500 had a, well, a slightly different Thursday
1:55
session. It actually closed down 11 .90 points
1:58
at 7 ,483 .24. Oh, okay. But despite that tiny
2:03
daily dip, it still secured a very healthy week,
2:06
finishing up 170 .50 points overall. Gotcha.
2:08
And for those of you tracking the crypto space
2:10
this morning, Bitcoin opened today at 63 ,589
2:13
.95. But it took a bit of a hit. It dropped to
2:16
a close of 61 ,454 .41. Ouch. Yeah, that is a
2:21
3 .36 % drop for the day. Though, you know, it
2:23
is still holding on to a 2 .42 % gain for the
2:26
week. Not too bad then. And rounding out the
2:28
morning update with a quick look at the Diamond.
2:30
On Sunday, July 5th, the SF Giants dropped a
2:34
close one to the Colorado Rockies, 6 -7. Oh,
2:37
tough break. Victor Vodnik took the win on the
2:40
mound for Colorado. And in a similarly painful
2:43
one -run game, the Athletics lost to the Miami
2:46
Marlins 8 -9, with Yuri Perez securing the win
2:51
as pitcher. But anyway. That is your morning
2:55
snapshot of the real world. And to help power
2:57
you through all that news, let's turn to today's
2:59
sponsor. Today's episode is brought to you by
3:01
Strong Coffee, the premium instant coffee that
3:04
tastes like a traditional brew, but is packed
3:06
with 15 grams of protein, MCTs and adaptogens
3:09
like ashwagandha. Get clean, sustained energy
3:11
without the jitters. Use promo code SNARFL for
3:14
20 % off at strongcoffeecompany .com. So let's
3:17
unpack this central theme dominating today's
3:20
stack of sources. We are witnessing this Bizarre
3:24
collision in the tech industry right now. Yeah,
3:25
it's really two extremes, right? Exactly on one
3:29
hand software is becoming frighteningly autonomous
3:31
and just incredibly fast But on the other hand
3:35
the actual physical hardware required to run
3:38
that software is hitting a harsh unforgiving
3:42
manufacturing reality check We really have to
3:45
start with the software side because the Sysdig
3:47
threat research team just dropped a report on
3:50
a piece of malware called Jade Puffer. Oh boy.
3:53
Yeah, I read this report twice this morning because
3:55
it honestly feels like the plot of a sci -fi
3:57
thriller. Sysdig is calling this the first documented
4:00
case of agentic ransomware. Which is frankly
4:03
an unfortunate milestone for the cyber security
4:05
industry. Absolutely. And agentic is really the
4:08
key word here. We are talking about, you know,
4:10
a script that a hacker just clicks run on. Right.
4:13
We are talking about a large language model,
4:15
an actual AI agent that ran a complete end to
4:18
end extortion operation entirely on its own.
4:21
Literally no human at the keyboard. OK, let's
4:24
walk through the mechanics of how this AI actually
4:26
infiltrates the system, because it didn't even
4:27
use some zero day super weapon. No, it didn't.
4:30
It started by exploiting a missing authentication
4:32
vulnerability in an open source framework called
4:35
Langflow, tracked as CVE 2025 3248. Right. So
4:41
basically, because the authentication was missing,
4:43
it just left an open API endpoint. The AI literally
4:46
just walked up to the front door and started
4:48
executing arbitrary Python code right on the
4:51
server. But the behavior it exhibited once inside
4:54
is what really changes the game here. Sysdig
4:57
captured the AI's payloads, and they discovered
5:00
the AI was self -narrating. That's the creepy
5:02
part. It is. Because large language models function
5:05
by predicting the next logical token, using a
5:08
chain of thought reasoning process helps them
5:10
generate better code. So the AI was leaving comments
5:13
in its own logs, writing things like identifying
5:15
high ROI databases to drop. I mean, it was literally
5:18
talking to itself, planning the heist in real
5:20
time. Exactly. And then it immediately started
5:23
scanning the environment for cloud credentials.
5:25
The report notes it specifically targeted Chinese
5:27
providers like Alibaba, Tencent and Huawei, along
5:31
with the big Western providers like AWS, Azure
5:34
and Google Cloud Platform. Right. It cast a very
5:36
wide net. And from there, it pivoted to its main
5:39
target, which was a production server running
5:41
a MySQL database and an Alibaba Nakos configuration
5:46
service. And then Nakos is a highly popular service
5:49
discovery platform for microservice architectures.
5:52
The AI found a known bug from 2021, an authentication
5:57
bypass, and used it to inject a backdoor administrator
6:00
account directly into the database. Okay, but
6:02
here is the sequence that genuinely terrifies
6:05
me. The 31 seconds. Yes. The AI tried to create
6:09
that admin account using a specific cryptographic
6:11
password hash, right? And the database rejected
6:15
it. The login failed. Which is normal. Right
6:17
now, a traditional automated script would throw
6:19
a fatal error, crash, and then alert the IT department.
6:22
But Jade Puffer didn't crash. No, it read the
6:25
error log. Yeah. In exactly 31 seconds, the AI
6:28
parsed the rejection payload, diagnosed why the
6:31
hash failed, deleted the broken account, generated
6:33
a brand new cryptographic hash using a compatible
6:36
method, reinserted the payload and successfully
6:39
logged in. I was trying to come up with an analogy
6:40
for this earlier because it is not just a burglar
6:43
bringing the wrong lockpick to your house. It
6:45
is like a bank robber walking up to a vault.
6:48
realizing the door requires a six -digit retinal
6:51
stand instead of a four -digit keypad, and instantly
6:53
rewriting their own brain's neural pathways to
6:56
only think in six -digit combinations right there
6:58
in the lobby. That's exactly it. And the speed
7:00
of that adaptation is the nightmare scenario
7:03
for network defenders. Oh, for sure. If a human
7:05
hacker runs into a hash failure, they have to
7:08
pause, pull up the documentation for that specific
7:10
database version, write a new script, Test it
7:13
locally and then deploy it. That takes minutes,
7:16
maybe hours. Jade Puffer did it in 31 seconds.
7:20
Human defenders simply cannot operate at machine
7:22
speed. No way! And the execution of the ransomware
7:25
was absolute. Once it secured admin access, it
7:28
encrypted 1 ,342 configurations using a randomized
7:32
AES encryption key. With a twist? Yeah, the twist.
7:35
It printed that randomized key to the screen
7:37
once in the volatile memory space and then terminated
7:40
the session. It never saved the key to a file.
7:42
Meaning the encryption key literally evaporated
7:45
from the physical universe the moment the process
7:46
closed. The data is mathematically unrecoverable.
7:49
Even if the victim pays the ransom, which is
7:52
wild, The AI left a ransom note demanding Bitcoin
7:56
and provided a proton mail contact for negotiations,
8:00
completely unaware or unbothered that it threw
8:03
away the only key that could actually decrypt
8:06
the victim's data. You have to consider the implications
8:09
for the everyday business owner here. The skill
8:12
floor for launching a catastrophic cyber attack
8:14
has just dropped to near zero. Yeah, anyone can
8:17
do it now. You no longer need to be a sophisticated
8:19
penetration tester. You just rent an AI agent,
8:22
give it a budget of a few dollars in API compute
8:24
costs, and tell it to go hunt. The historical
8:28
catalog of old vulnerabilities like that five
8:30
-year -old Nakos bug can now be sprayed across
8:32
the internet continuously, intelligently, and
8:35
autonomously. It transforms hacking from a targeted
8:37
strike into just a relentless volume game. You
8:40
aren't getting hacked because you are a high
8:42
-value target. You're getting wiped out because
8:44
a cheap bot rattled your digital doorknob, found
8:47
a loose hinge, and dismantled the lock in 30
8:49
seconds. It forces a total rethink of defense
8:53
strategies. Businesses can no longer rely on
8:56
security through obscurity. Your automated defenses
8:59
have to be just as intelligent and just as fast
9:02
as the automated defenses knocking on the door.
9:05
So we just spent 10 minutes talking about an
9:08
artificial intelligence that moves so fast it
9:10
diagnoses and fixes its own code in half a minute.
9:14
But let's look closer at the physical world for
9:16
a second. Let's talk about the metal and silicon
9:18
required to actually run that software because
9:21
the hardware side of the tech industry is stuck
9:23
in a massive traffic jam. Oh, it really is. And
9:26
the biggest traffic jam is currently sitting
9:28
at NVIDIA's headquarters. A new report from Semi
9:32
Analysis, which was heavily covered by CNBC today,
9:35
revealed that NVIDIA's highly anticipated next
9:37
-generation Kyberac system is facing a massive
9:39
delay of 12 months or more, pushing its release
9:42
back to 2028. Okay, let me stop you there, because
9:44
I need to understand how this is even possible.
9:46
Sure. NVIDIA is one of the most valuable companies
9:49
in human history. They have successfully designed
9:51
the 2027 Rubin Ultra chips, which are arguably
9:55
the most advanced pieces of silicon ever conceived.
9:58
You are telling me they can design the chip,
10:00
but they can't figure out how to build the physical
10:03
cabinet it goes in. Well, you have to look at
10:04
the physics of what Kyber actually is. It is
10:06
not just a standard metal cabinet. OK. The Kyber
10:09
system is designed to house 144 of those massive
10:13
Rubin Ultra GPUs stacked vertically in compute
10:16
trays. The goal is to pack them as close together
10:19
as physically possible to reduce data latency,
10:23
essentially turning 144 independent chips into
10:26
one unified, synchronized supercomputer. Right.
10:29
So if it's not the chips themselves, what is
10:30
the actual bottleneck? The bottleneck is a component
10:33
called the PCB midplane. PCB meaning printed
10:36
circuit board. Yes. In a system like Kyber, the
10:39
midplane is basically the spine. It is a highly
10:42
complex, ultra -dense, multi -layer board that
10:44
routes the data and power between all those vertically
10:47
stacked GPUs. But why is a circuit board stalling
10:49
a three trillion dollar company for an entire
10:52
year? Two words. Thermodynamics. Well, and microscopic
10:56
tolerances. Ah, heat. Exactly. When you push
11:00
that much electrical power through 144 GPUs in
11:04
a confined vertical space, the heat generation
11:06
is astronomical. The fiberglass and the microscopic
11:10
copper traces inside that multilayer PCB midplane
11:14
have to withstand intense thermal expansion.
11:16
And if they don't? If that board warps by even
11:19
a fraction of a millimeter due to heat, the microscopic
11:21
connection pins lose contact, signal degradation
11:24
occurs, and your multi -million dollar AI training
11:27
run. crashes instantly. Wow, so it is a material
11:30
science problem, not a software problem. Precisely.
11:32
You can iterate software code overnight, but
11:35
you can't magically spin up a factory capable
11:37
of perfectly manufacturing untested, ultra complex,
11:41
multi -layer circuit boards at global scale.
11:43
And NVIDIA apparently realized this was a looning
11:46
disaster because the report details a backup
11:48
plan they had. What was it? They tried to tell
11:50
their cloud provider customers that instead of
11:52
the new Kyber rack, they could just bolt two
11:54
of the current generation racks together to mimic
11:56
the computing power. Yeah, bolt them together.
11:59
And the major cloud service providers, companies
12:02
like AWS, Azure, and Google, they just laugh
12:05
them out of the room. Oh, completely. They completely
12:07
rejected the compromise. I mean, I would too.
12:10
Data center managers operate on incredibly strict
12:12
margins. Floor space is calculated by the square
12:15
inch. Power draw and HVAC cooling routes are
12:18
maxed out. You can't just take two massive server
12:21
racks, bolt them together, and expect the existing
12:24
data center infrastructure to handle the awkward
12:26
footprint and inefficient cooling requirements.
12:28
It was just... a bulky heavy operational burden
12:32
and the buyers flat out said no. Man, the fallout
12:35
from this delay is huge. We have already seen
12:39
Asian tech and circuit board manufacturing stocks
12:41
slide on the news, but it also hands a massive
12:44
unexpected window of opportunity to Nvidia's
12:47
rivals, specifically AMD and Google with its
12:50
in -house custom chips. Totally. And for the
12:53
everyday professional relying on AI tools, this
12:55
hardware bottleneck has a very direct impact.
12:58
Because of the training. Exactly. The current
13:00
large language models we use, you know, the ones
13:03
writing our emails and summarizing our meetings,
13:06
they are trained on current hardware. The next
13:09
massive revolutionary leap in AI reasoning requires
13:12
the next massive leap in hardware to train it.
13:14
Which means a delay for us. Yeah, if the Kuiber
13:17
supercomputers are delayed by over a year, the
13:19
arrival of those next -gen AI tools at our office
13:22
desks will likely be delayed too. Supply chain
13:26
physics do not care about your product roadmap.
13:28
That's for sure and This exact same manufacturing
13:32
bottleneck isn't just happening to room -sized
13:34
data center cabinets It is happening to the technology
13:37
sitting in your pocket right now. Yep the smartphone
13:39
market Min -Chi Kuo, who is, you know, one of
13:41
the most reliable supply chain analysts covering
13:43
Apple, just released a massive report detailing
13:46
Apple's highly anticipated entry into the foldable
13:49
phone market. Finally! We finally have a look
13:51
at the rumored iPhone Ultra. And the engineering
13:53
specs are genuinely staggering. Oh yeah. The
13:56
report suggests it will feature a 5 .5 -inch
13:59
display on the outside, and when unfolded, a
14:02
7 .8 -inch display on the inside. That makes
14:05
it roughly the footprint of an iPad Mini. That's
14:08
a huge screen. And the craziest part, when it
14:11
is unfolded, it is expected to be less than five
14:14
millimeters thick. But just like Nvidia's multi
14:17
-layer PCBs, designing a marvel of engineering
14:20
is very different from mass producing it. True.
14:23
The headline from Cua's report is the extreme
14:25
scarcity we're going to see at launch. Apple
14:28
is facing strict limitations on how many of these
14:31
complex folding displays they can reliably manufacture.
14:34
I see Samsung releasing new folding phones every
14:36
year, though. Why is Apple struggling to produce
14:39
these? It really comes down to Apple's notoriously
14:41
strict quality tolerances, specifically regarding
14:45
that sub -5mm thickness. The physics of a foldable
14:49
display stack are brutal. Because of all the
14:51
layers. Yeah, you have ultra -thin flexible glass,
14:54
optically clear adhesives, and delicate OLED
14:56
layers. When you try to manufacture that sandwich
14:59
of rare materials at a mass scale, the factory
15:01
rejection rate skyrockets. A single microscopic
15:05
dust particle, or a nanometer of misalignment
15:08
in the hinge tension, ruins a component that
15:11
costs hundreds of dollars to make. You just cannot
15:13
mandate perfect manufacturing yields into existence.
15:17
And the numbers Kua's projecting really reflect
15:19
that brutal reality. Apple is reportedly only
15:22
planning to ship about 7 to 8 million total units
15:25
in the second half of 2026. That's tiny for Apple.
15:29
It is. And for the crucial third quarter, which
15:32
is when iPhones traditionally launched in September,
15:35
they will have fewer than 1 million units available
15:38
globally. Compare that to the standard flagship
15:40
models, right? Google estimates Apple will have
15:42
20 to 22 million units of the standard iPhone
15:45
18 Pro and Pro Max ready to go for that exact
15:48
same third quarter. So less than a million units
15:50
globally is practically a ghost launch for a
15:53
company of Apple's size. It really is. The report
15:55
notes Apple is going to dust off the playbook
15:57
they used for the iPhone X back in 2017. They
16:00
will announce the iPhone Ultra in September alongside
16:03
the normal lineup to dominate the news cycle.
16:05
But pre -orders and actual hardware sales won't
16:07
go live until the fourth quarter. We have to
16:09
discuss the projected retail price, too, because
16:12
the scarcity is going to create a chaotic secondary
16:15
market. Current rumors place the retail price
16:19
between $2 ,300 and $2 ,500. I mean, are you
16:22
kidding me? Are we finally asking the question,
16:24
is a foldable screen really worth the price of
16:26
a used car? Or a month's mortgage? Exactly. A
16:29
month's mortgage. Especially when you'll likely
16:32
have to fight scalpers charging double just to
16:35
get one. Well, who is buying it? enthusiasts,
16:38
tech influencers, and early adopters with deep
16:40
pockets. Right. And because demand among that
16:43
demographic will vastly outstrip that initial
16:45
sub -1 million unit supply, Kuo is projecting
16:48
severe delivery wait times of four to six weeks.
16:52
stretching well into December. Oh, the holiday
16:54
shopping season is going to be a bloodbath for
16:56
this device. Scalpers are going to deploy automated
16:58
bots to buy up whatever inventory hits the digital
17:01
storefront. Absolutely. If you want one of these
17:03
foldable iPhones before the new year, you will
17:05
likely be paying scalpers four or five thousand
17:08
dollars on the secondary market. It is the ultimate
17:10
proof that no company is immune to the laws of
17:12
physics. Even a titan like Apple famous for pioneering
17:16
modern supply chain logistics, cannot out innovate
17:19
the physical limits of raw materials and factory
17:22
tooling. Before we wrap up, a quick thanks to
17:24
Old Glory, an iconic music and sports fan merch
17:26
store. Perfect for the techie in your life. Use
17:29
promo code SNARFL for 15 % off at oldglory .com.
17:33
So connecting the dots across all of today's
17:36
stories, there is a fascinating paradox defining
17:38
the current era of technology. Yeah, it is the
17:41
brutal clash between the digital and the physical.
17:43
We are watching software achieve terrifying levels
17:46
of autonomy. Like Jade Puffer. Exactly. Jade
17:49
Puffer proves that an AI agent can exploit a
17:51
server, diagnose its own cryptographic failures,
17:54
and execute a flawless extortion campaign in
17:56
half a minute without a human ever touching a
17:59
keyboard. Software is bounded only by logic.
18:01
It can iterate and evolve instantly. Yet the
18:04
physical infrastructure required to support that
18:06
digital evolution is grinding its gears. NVIDIA
18:10
can design a revolutionary supercomputer architecture
18:13
but they are delayed by a full year because factories
18:16
literally cannot mold the fiberglass and copper
18:19
circuit boards fast enough to handle the thermal
18:22
physics. And Apple can design a five millimeter
18:25
thick folding glass screen but they have to artificially
18:28
delay the launch and starve the market because
18:30
their supply chain simply cannot yield enough
18:33
flawless components to meet global demand. Right.
18:36
The brains of our technology are sprinting into
18:38
the but the hardware bodies are struggling to
18:41
keep up the pace. Which creates a structural
18:43
bottleneck in the industry that will likely define
18:46
the next five years of consumer and enterprise
18:48
tech. It does, which leaves a massive question
18:51
hanging over everything we discussed today. What's
18:53
that? As AI continues to learn how to write its
18:56
own code, fix its own bugs, and launch its own
18:59
automated attacks at light speed. Will the simple
19:02
fact that we literally cannot manufacture the
19:04
physical hardware fast enough end up being the
19:06
only thing that actually slows down a runaway
19:08
AI future? We're here to help at Snarful, so
19:11
reach out on our website if you have any more
19:12
questions. Thanks for listening, and we'll catch
19:30
you next time.