Late Wednesday night, Google unveiled Gemini 4 Argon, the most advanced artificial intelligence model it has built to date. A model that beats its major competitors in 13 out of 19 tests. And as if that weren't enough, it's preparing to send its own chips... into space. Yes, you read that correctly.
And yet, the stock closed lower yesterday.
"How does that make any sense, Christo?" you may be wondering. The answer probably lies within Google itself.
GEMINI 4 ARGON
According to Google, Gemini 4 Argon set a new record in programming under real-world conditions. In cybersecurity, it shares first place with OpenAI's GPT-6 Astra and Grok 4.7. It also ranks first in professional tasks, such as finance and law. Overall, it beat Astra, Anthropic's Fable 5.1 and Opus 5.5 in 13 out of 19 tests.
"And what does it actually do?" you may be wondering. Google is already using it internally. It improved memory management in its data centers with the model. And it freed up hundreds of terabytes of memory without buying any additional equipment.
There is, however, a "but." We cannot test it yet. Google is rolling it out gradually, starting with trusted cybersecurity partners through the Fairwind program. At the same time, it is conducting security checks together with the US government.
"That gives us greater confidence," explained Google's Tulsee Doshi. And, as she said, this means a model that is strong in cyber defense can reach those protecting systems more quickly.
In fact, the announcement came one day after Google's CEO, Sundar Pichai, signed a voluntary agreement at the White House with Trump and other technology executives concerning AI safety.
WHAT WALL STREET SAYS
And this is where Wall Street comes in, having reacted positively.
Nick Jones of BNP Paribas, with an Outperform rating and a $420 price target, focused on a comment from Pichai: "you'll see us bringing out new versions at a rapid pace." For Jones, Google remains fully committed to the battle for the top spot. And it has all the tools it needs for it.
He expects more models, at different price points, for different uses. And he would not be surprised if Google eventually released a digital assistant for the general public, similar to Meta's Muse.
"And why does this matter so much?" you may be wondering. Because almost a year ago, Gemini 3 had put Google back in the lead. But the next step took a long time. A very long time.
Initially, they expected it to launch around June as Gemini 3.5 Pro. Delays, caused by stricter safety safeguards and efforts to improve its coding capabilities, led to a change in direction. And since then, Alphabet's market capitalization has fallen by almost 20%. Because investors are concerned about the competition.
GOOGLE'S CHIPS IN SPACE
And now we come to the second part.
Alphabet planned to send its own artificial intelligence chips, the TPUs, into orbit on Thursday. They will travel aboard a SpaceX Falcon 9, inside a prototype Planet Labs satellite powered by solar energy.
This is the first in-orbit test for Project Suncatcher, which Google unveiled in November 2025. The goal? Artificial intelligence infrastructure in space that draws energy from the sun and operates continuously.
"And why in space, Christo?" you may be wondering. Because in low Earth orbit, satellites have almost continuous sunlight. And they can generate up to eight times more solar energy than on Earth. How they will withstand the conditions in orbit, however, is something Google still doesn't know.
And at this point, it's worth mentioning that Alphabet has a stake in SpaceX worth more than $82 billion. They are close partners, although they compete in artificial intelligence. That's because SpaceX is also planning its own data centers in orbit. Musk says that within two, or at most three, years, this will be the cheapest way to train artificial intelligence.
Of course, it's not all rosy. Experts say this is still a very distant prospect, assuming it ever happens. Launches are expensive and infrequent. Cooling is required, along with chips capable of withstanding extreme temperatures and protection from radiation. And then there is also space debris.
THE BIG BUT
Up to this point, everything sounds perfect. And yet, the stock closed lower yesterday. Why?
The answer probably lies in a Bloomberg report. According to people with direct knowledge, there are doubts within Google itself. Gemini 4 performs very well on tests. But when employees put it to work in practice, it performs worse. And it struggles with certain coding tasks.
And this is where things get even more interesting. Some employees believe Anthropic's Fable and OpenAI's Astra are improving faster than Gemini. And that Gemini 4, even in its best version, will remain behind in certain areas. Others, however, believe it has caught up with the leaders.
Google, for its part, denies this. It says the claims that the model is lagging in coding are not true. One employee familiar with its development spoke of a "broad consensus" that Gemini 4 is at the top. And Google's DeepMind chief, Koray Kavukcuoglu, said last week: "I have absolute confidence in the team. For me, it is certain that we will always be at the top."
So you might ask... who is right? We don't know. But after so many months of waiting, perhaps that uncertainty was enough to send the stock lower.
