Why Google Almost Missed the AI Race (and How It Found Its Way Back)
Today not selling roti prata but I'm going to dive deep into how Google pulled off this epic comeback in just about 1,000 days. And it all starts with the brilliant mind behind Gemini—the CEO of Google DeepMind, Demis Hassabis.
Demis Hassabis was born in London to a Greek-Cypriot father and a Singaporean-Chinese mother. He was a classic child prodigy: starting chess at age 4, and by 12, he taught himself programming. On an old Commodore Amiga computer, he coded his first game—a city-building simulation—to practice Reversi against himself.
At 13, he reached an Elo rating of 2300, making him the second-highest ranked player in the world under 14, just behind the legendary Judit Polgár. Later, he pursued a PhD in neuroscience at University College London (UCL), inspired by the human brain's potential to unlock AI secrets.
In 2010, Hassabis founded DeepMind. Early investors included PayPal co-founder Peter Thiel, and Elon Musk soon joined as well. One day, Musk mentioned DeepMind to Google co-founder Larry Page on a private jet. Page was instantly intrigued, and right after landing, he had his team reach out. In January 2014, Google acquired DeepMind for a staggering £400 million (about $500 million at the time)—what turned out to be one of Google's most pivotal investments.
Post-acquisition, Hassabis set out to prove that general reinforcement learning could tackle not just simple games, but the ultimate test of human intellect: Go. At the time, experts believed it would take at least another decade for a computer to beat a human Go champion.
Undeterred, Hassabis teamed up with his old UCL classmate David Silver and Taiwanese engineer Aja Huang. With a tiny team of under 20 people, they took on the Everest of AI: Go. In 2016, AlphaGo defeated world champion Lee Sedol 4-1, shocking the world. It even played a never-before-seen move on the 37th turn—a stroke of creativity that made everyone realize AI had surpassed humans in certain domains. This was Hassabis's global coming-out moment.
Then, in 2020, AlphaFold predicted the structures of 200 million proteins and made them freely available to the world. For this breakthrough, Hassabis shared the 2024 Nobel Prize in Chemistry. Yes—the driving force behind Gemini is a Nobel laureate.
But here's where the problems started. For years, Hassabis ran DeepMind like an academic institution rather than a product-focused division. This meant incredible tech, but it wasn't turning into profitable products fast enough.
Everything changed with the 2022 crisis. On November 30, ChatGPT burst onto the scene. Google was stunned—they had something similar internally, codenamed Sparrow, developed earlier that year. But Hassabis, chasing perfection, held off release. OpenAI beat them to market and defined the category, leaving Google playing catch-up.
Two months later, Google rushed out Bard to counter ChatGPT. Disaster struck: in the official demo video, Bard gave wrong info about the James Webb Space Telescope. The world thought Google had fallen from its pedestal. Panic selling wiped $100 billion off Alphabet's market cap overnight.
This finally woke up Google's leadership to the real issue: organizational chaos. Google had two major AI labs—Google Brain, founded in 2011 and led by legend Jeff Dean (Google's 30th employee), and DeepMind, acquired in 2014 under genius scientist Hassabis.
Absurdly, Bard wasn't even built by either—it came from a separate social products team. Over the years, CEO Sundar Pichai hadn't resolved the siloed structure. One product spanned three departments, with the two elite AI teams duplicating efforts—a massive waste of talent.
With ChatGPT at the gates, only the founders' authority could force change. Larry Page and Sergey Brin endorsed Pichai, giving him the political capital for a massive reorganization.
Before the merger, Pichai met privately with Hassabis. He promised: "If you can beat OpenAI, I'll back you fully—manage the team however you want, cancel projects, unify codebases, anything." In exchange, Hassabis had to step out of his ivory tower and align unconditionally with Google's commercial goals.
Deal struck. Pichai then promoted Jeff Dean to Chief Scientist—effectively sidelining his direct management power—and handed full command of the AI empire to Hassabis.
In April 2023, Pichai ordered the merger of Google Brain and DeepMind into one entity: Google DeepMind. That's why it's called Gemini—the twin stars united.
With control, Hassabis took the helm of Google AI. But he inherited a mess: half the team didn't respect him, split between London and California offices, constant cultural clashes—even arguments over frameworks. Google Brain used TensorFlow; DeepMind stuck with JAX. Some even wanted to sneak in PyTorch.
This fragmentation killed efficiency and iteration speed. Hassabis made a risky but crucial call: ditch TensorFlow entirely and standardize on JAX.
The switch meant Google spent most of 2023 getting crushed by OpenAI with no real counterpunch. But by December, Gemini 1.0 finally launched.
GPT-4 was already dominating reasoning and writing. Head-to-head, Google might tie at best. But Hassabis spotted OpenAI's Achilles' heel: limited context window. GPT-4 topped out at around 128K tokens—enough for maybe 300 pages, far too little for enterprise needs.
Hassabis bet everything: redirect Google's precious compute resources to ultra-long context. This wasn't just adding servers—it required gutting core algorithms, risking model collapse.
He won the bet. Gemini 1.5 delivered a dimensional leap over GPT-4. Looking back, the timing was frantic: just two months between Gemini 1.0 and 1.5—like Apple dropping iPhone 16 right after 15, with 10x performance.
In 2024, Google abandoned normal product cadence. Whatever cards they had, they played all out.
Then, on November 18, 2025, without warning, Google released Gemini 3. The tech world exploded. It earned near-universal praise from users. Forums filled with cancellation screenshots: "Goodbye ChatGPT—Gemini 3 is the GPT we always wanted." Google was truly back.
To the outside world, Hassabis is Google's AI savior. But there's another unsung hero in this war: compute power, controlled by the man Pichai demoted—Jeff Dean.
Google's TPUs are now threatening Nvidia's dominance. But Dean foresaw this back in 2013: if every user searched with voice for just 3 minutes daily, data centers would need to double—impossible cost-wise with CPUs or GPUs.
To squeeze maximum efficiency, Google built custom TPUs. AlphaGo beat Lee Sedol in 2016 on Google's homegrown TPUs.
Fast-forward to late 2025: amid the worst global AI chip shortage, while OpenAI waits endlessly for Nvidia deliveries (delaying GPT-5), Google's data centers quietly run hundreds of thousands of self-developed Trillium and next-gen Ironwood TPUs—full throttle, no queues, no relying on Jensen Huang.
This is Gemini's secret weapon against ChatGPT. Training is a one-time burn, but every second of chatting burns power and compute. Analysts estimate Google's TPUs have the lowest cost for massive real-time interactions—a moat rivals dependent on Nvidia can't easily cross.
Through this comeback from the brink, we see the real battle isn't just who releases the best model first. It's end-to-end strength: talent, custom chips, models, data.
The Google that lost $100 billion overnight now boasts a record market cap of around $3.8 trillion in December 2025, protected by moats rivals struggle to match.
Will OpenAI disrupt everything again? What's their terrifying counter brewing? How will chip giants reshape the war? We'll keep tracking this century-defining battle.
One more thing: the real fracture between Musk and Larry Page came over OpenAI—specifically the brutal recruiting war for Ilya Sutskever. It was mostly Demis on one side, me on the other, both pulling hard. Ilya flipped back and forth—stay at Google, leave, stay, leave—until he finally joined OpenAI. That was one of the toughest recruiting fights ever.
@TigerObserver @Daily_Discussion @Tiger_comments @TigerPM @TigerStars
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
