Google Spends $180 Billion and Its Stock Falls: The AI Capex Paradox

👁 119 reads

Record investment, falling share price

When Alphabet’s February earnings report revealed $180 billion in capital expenditure for 2025, the market’s reaction was anything but enthusiastic. Google’s stock dropped 7.2% on the day the results were published, even as the company posted impressive numbers: Gemini 3 had passed 750 million monthly active users, and AI services revenue was up 340% year on year. That contradiction neatly captures the growing scepticism on Wall Street toward astronomical AI spending.

$180 billion in capex exceeds the GDP of most European countries. For context, Finland’s entire GDP in 2025 came to roughly $170 billion. Google spent more on AI infrastructure than the combined market value of all companies on Poland’s benchmark WIG20 index. It is a figure that stops even seasoned tech analysts in their tracks.

Never before in the history of technology have we seen capital investment of this magnitude directed at a single technology. This is a moonshot on steroids, said Mark Patterson, senior analyst at Wedbush Securities, at a press conference on the day Alphabet’s results were published.

Gemini 3: technical triumph or Pyrrhic victory?

Gemini 3, the latest iteration of Google’s language model, has genuinely delivered impressive results. Released publicly in October 2025, it reached 750 million monthly active users in just four months, a pace that even outstrips ChatGPT, which needed six months to hit comparable numbers.

The technical capabilities are not in question. The model supports context windows of up to 10 million tokens, meaning it can process entire books or lengthy technical documents in a single call. Its multimodal architecture handles text, images, audio and code simultaneously, often matching or exceeding human specialists in precision.

The operational cost of that success, however, is staggering. Each Gemini 3 call costs Google an average of $0.23 in compute infrastructure, while average revenue per user per month is just $1.40. That means every active user must make fewer than six queries a month for Google to break even. The actual average is 47 queries per month per user.

The unit economics of AI are brutal. The better the model, the more expensive it is to run. Google is building the most expensive money-burning machine in the history of technology, said Dr Sarah Chen, professor of technology economics at MIT, on the “AI Economics” podcast.

Infrastructure on a nation-state scale

The infrastructure Google is building for AI bears comparison with government-scale projects from the space-race era. The company announced it brought 47 new data centres online across six continents in 2025, each costing between $3 billion and $5 billion. Together they draw 2.3 gigawatts of electricity, roughly the consumption of a mid-sized European city.

The most expensive facility, opened in Oklahoma in December 2025, cost $12 billion and covers an area equivalent to 340 football pitches. It houses 180,000 of the latest TPU v6 chips, each priced at around $40,000. The total compute power of this single centre surpasses the combined capacity of every supercomputer in the world as of 2020.

Data centres are only part of the story. Google also invested $15 billion in its own subsea cables, building a global fibre network with a throughput of 400 terabits per second. A further $8 billion went toward acquiring startups focused on data-centre cooling and energy optimisation.

Market sentiment: from euphoria to scepticism

Markets were initially enthusiastic about Google’s AI spending. In the first half of 2025, Alphabet shares rose 34% and analysts lifted their price targets. Sentiment shifted sharply in the third quarter once it became clear that AI investment was not translating into proportional gains in profitability.

Google’s operating margin in its AI segment now stands at minus 23%, meaning every dollar of AI services revenue generates a loss of 23 cents. That is a stark contrast with the traditional advertising business, where operating margins exceed 60%. Investors have started asking whether the company has fallen into a trap of its own technical success.

Google reminds me of Tesla in 2018: a great product, but the unit economics make no sense. The difference is that Tesla was selling cars for $100,000, and Google is giving AI away for free, wrote Jim Covello, chief investment strategist at Goldman Sachs, in a quarterly report.

Google’s pricing strategy has drawn particular criticism. While competitors such as Anthropic and OpenAI use subscription models charging $20 to $200 per month, Google offers Gemini 3 free of charge, betting on monetisation through advertising. That strategy worked for search; at AI’s cost structure, it generates losses.

The competition isn’t standing still

Competitive pressure complicates Google’s position further. OpenAI, working with considerably smaller resources, recorded $27 billion in revenue in 2025, driven largely by its subscription model. Sam Altman’s company invested “only” $45 billion in infrastructure and achieved a gross margin of 67%.

The bigger surprise was Anthropic, whose Claude 4 attracted 200 million users paying an average of $89 per month. The company was valued at $140 billion despite total infrastructure investment of around $18 billion, ten times less than Google spent in a single year.

Microsoft, Google’s traditional rival, took a hybrid approach. It integrated AI into existing products such as Office 365 and raised subscription prices by an average of 30%, generating an additional $34 billion in annual revenue at relatively modest additional infrastructure cost.

Google is fighting on two fronts: on one side against pure-play AI companies like OpenAI, on the other trying not to cannibalize its lucrative advertising business. It is a near-impossible equation, said Benedict Evans, independent technology analyst, at CES 2026.

Long-term vision versus short-term returns

Alphabet CEO Sundar Pichai continues to defend the strategy, arguing that AI investment represents “the foundations of the next decade of technology.” In his February 2026 letter to shareholders he wrote that Google is building infrastructure that will serve the company for the next 20 to 30 years, much as its search investments of the 1990s did.

The company has also laid out ambitious monetisation plans. By 2028, Google intends to launch Gemini Enterprise, a suite of AI business tools priced at $500 to $2,000 per user per month. The next step would be Gemini Cloud, a platform allowing developers to build their own AI applications on Google’s infrastructure.

Analysts are divided on how realistic those plans are. A McKinsey survey from January 2026 found that 73% of companies planning enterprise AI deployments prefer Microsoft solutions because of integration with existing systems. Google would need not only to match Microsoft on features, but also to persuade those customers to switch providers.

Implications for the wider industry

Google’s paradox has broader consequences for the technology sector. The astronomical capex of the big tech companies is starting to resemble an arms race, in which every player must spend ever more just to stay competitive. Amazon committed $156 billion to AWS and AI in 2025, Microsoft $142 billion, Meta $89 billion.

Those combined investments, totalling more than $560 billion, exceed the annual defence budgets of most NATO members. The question is whether the industry has trapped itself in a cycle of over-investment, where every participant must keep spending more to avoid being left behind.

Signals from the venture capital market are particularly concerning. According to a PitchBook report, investment in AI startups fell 34% in the second half of 2025, largely because of fears about the dominance of the major tech platforms. Small companies simply cannot compete with the infrastructure spending of Google, Microsoft or Amazon.

We are watching the oligopolisation of AI in real time. In five years there will be two, maybe three players capable of training frontier models. That is not healthy for innovation, warned Prof. Erik Brynjolfsson at the Stanford HAI conference.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Where this leaves us

Google’s $180 billion bet illustrates the central challenge of technology today: how to reconcile the revolutionary potential of artificial intelligence with its brutal economics. Gemini 3 has been a spectacular technical achievement, reaching 750 million users, yet the unit economics remain catastrophically negative.

Investor scepticism is not without foundation. In the short term, Google has built the most expensive capital-destruction machine in history, wagering that future AI revenues will justify today’s spending. It is a very high-stakes bet: if it pays off, the company cements its dominance for decades. If it doesn’t, Google risks repeating the story of the dot-com era.

The next 18 months will be decisive. Google must demonstrate it can actually monetise its AI investment. The launch of Gemini Enterprise and other paid services will be the test of whether Sundar Pichai’s vision can survive contact with reality, or whether the company becomes a cautionary tale about uncritical investment in future technologies. The capex paradox will turn out to be either the greatest transformation in Google’s history, or the most expensive lesson Silicon Valley has ever had to learn.

Kinetyra