AI PULSE - The AI Chip Wars

Google One Surpasses 150 Million Subscribers with New AI Tier Expansion Mozilla Firefox to Promote Perplexity Search Engine Google to Integrate Gemini AI into Chrome for Windows by 2025 FBI Warns of Deepfake Voice Campaigns Impersonating Officials to Compromise Devices TikTok Introduces AI Alive for Animating Static Photos into Dynamic Videos Salesforce Enhances AI Agent Trust with New Security and Compliance Features The AI Chip Wars #AI, #ChipWars, #Technology, #Deepfake, #AIIntegration, #Cybersecurity, #AIInnovation

Episode Transcript

(00:00):
Welcome to Innovation Pulse, your quick, no-nonsense update on the latest in AI.

(00:09):
First, we will cover the latest news.
Google One hits 150 million subscribers with new AI plans.
Mozilla tests pop-up ads.
Gemini AI integrates into Chrome, FBI warns of AI scams.
TikTok launches AI video tools.
After this, we'll dive deep into Salesforce's new AI governance and security tools,

(00:34):
addressing crucial data trust and compliance issues.
Google One, Alphabet's cloud storage and AI subscription service,
has reached over 150 million subscribers,
marking a 50% increase since February 2024.
This growth includes a new $19.99 monthly AI plan,

(00:57):
contributing millions to the subscriber count,
as noted by Google VP, Shimrit Ben-Yair.
Google One is part of Alphabet's strategy to reduce reliance on advertising,
which dominated its $350 billion revenue in 2024.
The rise of AI, including threats from chatbots like OpenAI's ChatGPT,

(01:22):
challenges Google's traditional search engine dominance.
An Apple executive highlighted a decline in Google searches on Safari,
as Apple eyes AI search options.
With AI yet to integrate ads smoothly,
many firms, including Google, are exploring subscription models.

(01:42):
CEO Sundar Pichai emphasized focusing on subscriptions for monetizing AI advancements,
like Gemini.
Join us as we step into the evolving search landscape.
Mozilla Firefox is planning to test a pop-up advertisement
for the Perplexity search engine in its upcoming 139 release.

(02:03):
This move suggests a potential partnership or revenue deal
between Mozilla and Perplexity,
reminiscent of Google's substantial payments to be Firefox's default search engine.
Currently, Firefox supports the Perplexity AI chat,
which is the only Perplexity AI chatbot in the sidebar,
and there's interest from users to integrate Perplexity as a search engine option.

(02:28):
The call-out introduces Perplexity as a new search method,
offering complete, well-cited answers in a chat-like format.
Mozilla's experiment might target specific regions
or segments of Firefox 139 users.
With Chrome's future uncertain,
Google will be exploring alternative search engine partnerships,

(02:48):
although the financial details of this experiment remain undisclosed.
If successful, this could signify a shift in online searching methods.
Google is planning to integrate Gemini, a conversational AI,
into Chrome for Windows 11 and 10.
This was confirmed during Google's antitrust trial,

(03:11):
indicating a move to compete with Microsoft's co-pilot in Windows 11.
A slide from Google highlighted their aim to blend AI with Chrome,
promising more features by 2025.
In 2024, Chrome added AI features like Google Lens and Comparison Tools.

(03:31):
At Google I.O. 2025, Gemini Live will be a key focus,
bringing conversational AI to Chrome on desktops.
Future features are confidential,
suggesting more than just Gemini integration.
Gemini will function as a resizable widget,
accessible via a system tray or toolbar icon.

(03:53):
Even if Chrome is closed,
the Gemini button will remain in the system tray.
Taskbar integration is confirmed,
with updates indicating it's in development.
Gemini will rely on Chrome without a standalone app for Windows.
The FBI warns of a scam,
using AI-generated voice audio to impersonate United States officials,

(04:15):
tricking targets into clicking harmful links.
Since April 2025, these deepfakes have targeted current and former government officials,
urging people to verify messages claiming to be from senior officials.
Attackers use deepfake voices to mimic individuals convincingly,
often indistinguishable without trained analysis.

(04:39):
They might ask to switch conversations to another platform,
then send malicious links.
Recent incidents include a deepfake of LastPass' CEO targeting an employee
and a robocall campaign using President Biden's voice to influence voters.
To protect yourself, verify the identity of callers,

(05:00):
inspect emails and messages for small discrepancies,
and be cautious of urgent claims.
Recognize that these scams can fool anyone,
and when uncertain, seek help from security officials or the FBI.
Join us as we step into the world of AI storytelling.
TikTok has introduced AI Alive,

(05:23):
a tool that animates static photos into engaging short videos
using the TikTok Story Camera.
This feature allows creators to add movements and effects,
enhancing visual storytelling.
To use AI Alive, users access the Story Camera,
select a photo, and specify desired changes,
like making a photo dance or wink.

(05:46):
The AI processes the request, resulting in a realistic, animated version.
TikTok ensures content safety through moderation
and labels AI-generated content for transparency.
This is one of the first AI-driven image-to-video tools within an app.
Similar AI features exist on Instagram and Snapchat,

(06:07):
with the latter developing its own video generation tool.
These innovations aim to boost user engagement
by offering new interactive content options.
TikTok is also reportedly exploring features for sending photos
and voice messages in direct messages.
Salesforce has introduced initiatives to ensure trust and governance in AI agents,

(06:29):
addressing the risks of unregulated AI interactions.
Recognizing the importance of secure AI engagement with customers and employees,
Salesforce unveiled governance, security, and compliance tools.
A survey revealed that 48% of IT leaders doubt their data's ability to support AI

(06:50):
and 55% lack confidence in their guardrails for AI deployment.
Salesforce's new capabilities address these issues,
providing a unified platform for secure and governed AI.
Tools like Agent Force and Data Cloud offer real-time monitoring and data governance.
Features include Salesforce Backup and Recover for Data Protection,

(07:15):
Archive for Compliance, and Shield 2.0 for Threat Detection.
The Agent Force Testing Center allows developers to test AI agents in isolated environments,
ensuring reliable performance.
These tools help create powerful and responsible AI,
enhancing enterprise trust and security.

(07:36):
And now, pivot our discussion towards the main AI topic.
Hey everyone, welcome to another episode of Innovation Pulse.
I'm Alexander Vogue, and with me as always is my tech-savvy co-host.
That's right, Yakov Lasker here.
Great to be back with all of you innovation junkies.
We've got a spicy topic today that's been heating up the tech world.

(07:58):
It's all about semiconductors.
We're going to be talking about innovation today.
We're going to be talking about innovation today.
We're going to be talking about innovation today that's been heating up the tech world.
It's all about semiconductors, geopolitics,
and the high stakes chess game between the US and China over AI chips.
Absolutely, Alex.

(08:19):
The US has been taking some pretty aggressive steps to limit China's access to cutting-edge semiconductors,
especially those high-end AI chips like NVIDIA's A100s and H100s,
AMD's MI250s, and similar hardware.
Yeah, those are basically the Ferraris of the chip world, right?
The kind of processing power that drives everything from autonomous systems to next-gen AI models.

(08:41):
Exactly, and these restrictions represent a major shift in US policy.
This isn't just about tariffs anymore.
We're talking about actively preventing an entire category of technology
from being sold to the world's second largest economy.
So break it down for us.
What's actually driving these restrictions?
What's the US worried about?
It boils down to two main concerns.
First, there's the national security angle.

(09:03):
The US government is worried that these powerful AI chips could be repurposed for military applications,
improving weapons systems like hypersonic missiles, enhancing surveillance capabilities, and boosting cyber warfare.
I remember Jake Sullivan, the national security advisor, talked about this.
Didn't he say something about maintaining as large of a lead as possible in these foundational technologies?

(09:26):
That's right. It marks a real doctrinal shift.
Instead of just staying a generation ahead, the goal now is to prevent China from catching up at all in these critical chip capabilities.
Sullivan called leadership in AI hardware a national security imperative.
And there's the economic side too, right?
These chips are basically the oil of the 21st century.

(09:47):
Whoever controls this technology controls a lot of future innovation.
Absolutely. The Commerce Department has called semiconductors ground zero of our tech competition with China.
And we've seen this translate into action with policies like the CHIPS Act,
investing 52 billion oz in domestic semiconductor R&D in manufacturing.
So what does this actually look like in practice?

(10:08):
How is the US implementing these restrictions?
It's what policy wonks call a small yard high fence approach.
The US is tightly controlling a narrow set of critical technologies,
like those 5, 10 NMA AI chips and the equipment to make them,
while keeping trade in less sensitive tech relatively open.
I remember those October 2022 controls created a huge stir in the industry.

(10:31):
What was actually in those rules?
The October 2022 rules basically did three things.
First, they cut off China's access to top end AI chips like NVIDIA's H100s.
Second, they tried to prevent Chinese companies from developing their own advanced chips
by restricting design software and manufacturing tools.
And third, they aimed to hamper China's chip equipment industry

(10:54):
by blocking components used to build chip making machines.
Has it worked? Are these restrictions actually keeping these chips out of China?
It's a mixed picture. In the immediate aftermath,
China's access to top tier AI chips was definitely curtailed.
Their semiconductor imports dropped 15.4% in 2023, the steepest decline on record.

(11:17):
Direct purchases of cutting edge NVIDIA GPUs like the H100
basically went to zero through official channels.
But I'm guessing there's a but coming.
You know it. Despite the bans, Chinese entities, including military research institutes
and universities, have managed to acquire small batches of banned chips
through gray market suppliers.
Reuters found dozens of procurement tenders where Chinese labs bought banned GPUs

(11:42):
through underground market intermediaries.
So they're creating an entire shadow market for these chips?
Exactly. Vendors are reportedly buying up excess stock and re-rooting them through third countries,
incorporating trading firms in places like India, Taiwan, Singapore
to move them to China without detection.
There's even been documented cases of smuggling, chips mislabeled as other goods or hidden in shipments.

(12:05):
But if they're only getting small batches, that's still a win for the US policy, right?
Wasn't the goal to slow down China's AI development, not completely halt it?
That's right. Chris Miller, who wrote that book, Chip War, noted that the goal is largely to
throw sand in the gears of China's AI development to slow it down, raise costs, and limit scale,

(12:26):
rather than achieve 100% denial.
And that seems to be happening.
Training a state-of-the-art AI model like GPT-4 requires something like
30,000 plus NVIDIA, a 100 GPUs, working in parallel.
If Chinese labs are only buying 5 to 10 GPUs at a time,
that's nowhere near what they need to build cutting-edge AI from scratch.

(12:51):
Right, and it's forced NVIDIA and AMD to forego significant sales.
NVIDIA's portion of revenue from China has been shrinking, from about 26% in 2022 to projected 10% in 2025.
In fact, in April 2025, NVIDIA disclosed a $5.5 billion charge for unsold inventory
after the US government banned its last China-bound AI chip, the H20.

(13:15):
Ouch! So American companies are definitely feeling the pain here, too.
For sure. But the bigger question is, how is China responding to all this?
From what I understand, they're not taking it lying down.
Not at all. They're pursuing a multi-pronged strategy.
First, they're pouring unprecedented funding into their domestic chip sector.
In 2023, Beijing launched a new big fund with an additional $27 billion for semiconductor development.

(13:43):
And that's on top of local subsidies, tax breaks, and state-guided loans.
That's serious money. But can you really just throw cash at this problem?
These technologies took decades to develop.
While they're doing more than just throwing money at it, they also engaged in significant stockpiling before restrictions hit.
Chinese tech giants like Alibaba, Tencent, Baidu, and others rushed to import high-end GPUs while they still could.

(14:11):
In the first quarter of 2025 alone, Chinese companies collectively ordered $16 billion worth of NVIDIA's H20 chips,
right before those, too, fell under a license ban in April.
Smart move. So they've built up a war chest of sorts.
Exactly. Industry reports suggest China's top cloud and internet companies had amassed around 1 million units of advanced NVIDIA GPUs by 2023.

(14:35):
Though exact figures are hard to verify, this stockpiling gives them a buffer for the short to medium term.
But stockpiles eventually run out. What's their long-term plan?
The most interesting part is their push to develop homegrown alternatives.
Chinese companies from giants like Huawei to startups like Biren, Cambrikan, and more Threads are racing to design high-performance AI chips.

(14:59):
Huawei's Ascend series of AI processors has been making headlines.
I heard about their Ascend 910C. Apparently it's comparable to NVIDIA's A100H100 chips.
That's right. Huawei engineered it by essentially combining two of its 7Ni Ascend 910 chips on one package,
what they call advanced chiplet packaging, to double the compute and memory.

(15:20):
It's not quite the same manufacturing advancement, but a clever architectural work-round.
They reportedly began mass production in Q1 2025.
And they're making progress on the manufacturing front, too. Right?
I remember when that Huawei Mate 60 Pro phone came out with a chip that wasn't supposed to be possible under US sanctions.
That was a big moment. In 2022, China's flagship foundry smith surprised everyone by producing chips at an effective 7Ni Mate 30 process.

(15:48):
By 2023, that process was used to manufacture Huawei's Kirin 9000S smartphone chip, found in the Mate 60 Pro.
It demonstrated that China could produce a semiconductor maybe 5 to 7 years behind the leading edge, but still a breakthrough under such heavy sanctions.
So where does this all lead? Is the US strategy working or is it just accelerating China's drive for self-sufficiency?

(16:14):
That's the million dollar question. On one hand, the restrictions have definitely slowed China's AI progress in the short term.
But they've also galvanized China's determination to break free from foreign dependence.
As one report noted, the export controls prompted China to implement an all-out, government-backed effort that has already resulted in a number of startling achievements.

(16:35):
It's like squeezing a balloon. The pressure just pops up somewhere else.
Exactly. The US restrictions have kicked off this high stakes race. The next few years will reveal whether denying current tech slows China enough for the US to extend its lead, or whether China's crash program will innovate around the choke points.
And the stakes couldn't be higher. We're not just talking about who makes the fastest smartphone. AI is going to transform warfare, economic competition, everything.

(17:03):
Absolutely. This chip war will shape who leads in the AI era, and that has implications for everyone, not just tech companies.
On that sobering note, I think we should wrap up. This has been a fascinating look at the semiconductor battle between the US and China.
And it's far from over. I suspect we'll be revisiting this topic again soon as the situation evolves.
Well, that's all the time we have for today's Innovation Pulse. Thanks for joining us for this deep dive into the high stakes world of AI chips.

(17:31):
And remember, whether your Team USA or Team China in this tech race, the pace of innovation benefits us all in the long run. Keep pushing the boundaries.
We'll see you next time.
Yakov Lasker signing off. Until next time, peace.

(18:01):
Don't forget to like, subscribe and share this episode with your friends, colleagues, so they can also stay updated on the latest news and gain powerful insights.
Stay tuned for more updates.