Another week, another head spinning amount of AI news.
On Tuesday, Microsoft introduced two new AI agents for Microsoft 365 users called Researcher and Analyst.
As their names suggest, these smart assistants use OpenAI’s technology to handle deeper research and data analysis.
For context, Microsoft Office has more than 1.5 billion users worldwide. This kind of reach shows you how rapidly AI-powered automation can spread today.
Not to be outdone, that same day OpenAI CEO Sam Altman announced a big update to ChatGPT’s image-making abilities.
The AI can now create and edit pictures using the GPT-4o model, which was previously limited to text.
Here’s a sample of what it’s able to do:
Image: OpenAI
On Thursday, China’s Alibaba released its newest AI model, Qwen2.5-Omni-7B.
This model understands text, images, audio and video and provides real-time responses. It’s also designed to run efficiently on devices like smartphones.
According to the company: “This unique combination makes it the perfect foundation for developing agile, cost-effective AI agents that deliver tangible value, especially intelligent voice applications.”
But to me, this week’s biggest breakthrough came from Alphabet.
Google just released Gemini 2.5 Pro, a free AI model with improved reasoning skills.
It’s designed to make fewer mistakes, give more logical answers and better understand context.
The company calls it their most powerful AI yet. I’ve tried it, and it does seem to have noticeably improved reasoning skills.
But as amazing as Gemini 2.5 Pro Experimental is, what we really need to talk about is Google’s recent DeepSeek moment.
Google’s DeepSeek Breakthrough
For months, the AI industry has been obsessed with efficiency.
The cost of running large language models is staggering, and it goes up as models become more powerful.
Image: Merlin Search Technologies
That’s why companies are racing to figure out how to get the most power out of the least amount of hardware.
DeepSeek AI rocked the tech world by showing impressive efficiency in GPU usage in its R1 model.
Some people saw it as proof that China was going to win the race to artificial superintelligence (ASI).
But as I noted at the time: “China might have launched an amazing model, but America will win the race to ASI.”
And this month, Google might already be proving me right.
That’s because Google’s latest open-source model, Gemma 3, claims to rival DeepSeek AI’s R1 in performance…
But it only uses a fraction of DeepSeek’s computing power.
Google claims that Gemma 3 achieves 98% of the performance of DeepSeek’s R1 model, as measured by an Elo score (a ranking system originally used for chess and sports).
As you can see in the following chart, R1 scores 1363, while Gemma 3 hits 1338.
Source: Google
That’s a small difference, but it’s not the most important takeaway from this chart.
Instead, look at the dots underneath each column. Each one represents one of Nvidia’s powerful H100 GPUs.
DeepSeek’s model reportedly needs 32 of them to reach that level.
Gemma 3 runs on just one.
The AI Race is Evolving
If that’s accurate, it’s a game-changer. Because it shows that AI progress isn’t just about making bigger, more powerful models…
It’s also about making them more accessible.
If a model with near-top-tier performance can run on a single GPU, that opens the door for smaller companies, independent developers and even AI applications that run on personal devices instead of massive cloud servers.
It’s a huge leap toward democratizing AI.
Google is marketing Gemma 3 as the most capable model that can run on a single GPU or TPU (Google’s custom AI chip).
They’ve packed it with features that make it useful for real-world applications, from understanding multiple languages to reasoning over images and videos.
Plus, it can process a massive amount of text at once thanks to an expanded context window of 128,000 tokens. That means it can “remember” more of a conversation or document, leading to smarter responses.
If Gemma 3’s efficiency claims hold up, it signals a new phase in AI development.
The early days of large language models were all about raw power. It was about bigger models. More GPUs. More compute.
But now, we’re seeing a shift toward optimization. And companies that can get the most performance out of the least hardware could have a serious edge.
In other words, the AI arms race is no longer just about having more.
It’s also about how smart you can be with what you’ve got.
Here’s My Take
To be clear, we still need to build advanced data centers. And our energy infrastructure still needs a ton of work to adequately power our future.
That hasn’t changed.
I believe it will take a mix of more compute and better optimized AI models to win the race to ASI.
But Gemma 3 shows us that the race is speeding up.
And that means the market could be about to enter a period of massive turmoil due to AI.
For the first time since the dawn of the internet bubble, we’re running headlong into an exciting technology that can cause melt-ups and meltdowns in equal measure.
And with tariffs, global wars and political strife also agitating the market, making money today seems harder than ever.
That’s why I’m going live next Wednesday with an urgent event I’m calling Panic 2025.
Because all this volatility we’re experiencing could cause big gains to slip through your fingers. It could also result in major market players breaking down faster than you realize.
To get through this time of crisis, investors need to cut their emotions out of the picture.
And that means building a regimented process for understanding when to buy any stock, how much to buy…
And when to sell it.
Join me this Wednesday, April 9 at 1 p.m. ET and I’ll show you how to navigate this panic with the help of a genius software architect who’s changing the way tens of thousands of everyday folks invest.
We’ll show you how you could walk away from all this recent turmoil with the biggest gains of your life.
Click here now to reserve your spot.
Regards,
Ian KingChief Strategist, Banyan Hill Publishing
Editor’s Note: We’d love to hear from you!
If you want to share your thoughts or suggestions about the Daily Disruptor, or if there are any specific topics you’d like us to cover, just send an email to [email protected].
Don’t worry, we won’t reveal your full name in the event we publish a response. So feel free to comment away!
