Google's AI Gamble: Billions Poured into Infrastructure as Gemini 3.5 Pro Faces Delay

Alphabet's massive AI CapEx surge and the delayed Gemini 3.5 Pro launch highlight the high-stakes race for AI dominance.

/ Article
Google's AI Gamble: Billions Poured into Infrastructure as Gemini 3.5 Pro Faces Delay
Gregory Rocher (IFREMER Centre Bretagne, IRSI-RIC - ZI de la Pointe du Diable - CS 10070 - 29280 Plouzané, France), CC BY 4.0 via Wikimedia Commons

Alphabet is under intense investor scrutiny today as it reports its second-quarter 2026 results. The market is laser-focused on whether Google Cloud’s growth can justify the company’s colossal and accelerating AI infrastructure spending. This comes amidst news of a delayed launch for Google’s flagship Gemini 3.5 Pro AI model.

Google parent Alphabet previously raised its 2026 capital expenditure guidance to an eye-watering $180 billion to $190 billion. This figure more than doubles the $91.4 billion spent in 2025. The company even announced plans to raise approximately $80 billion through equity offerings to help fund this AI compute investment. This aggressive spending is driven by a simple fact: demand for AI capacity is outstripping supply. Google Cloud’s backlog nearly doubled in a single quarter to $460 billion in Q1 2026, with management indicating that revenue would have been higher if sufficient capacity existed.

The CapEx Arms Race Intensifies

This massive CapEx surge from Alphabet is not an isolated event. It is a defining characteristic of the current tech era. The market’s center of gravity has shifted from the legacy FAANG companies, focused on capturing human attention, to the Mag-7 and now the AI-native MANGOS players. The prize is no longer just eyeballs, but artificial intelligence itself. Nvidia and Google, both members of the Mag-7 and MANGOS, sit at this critical intersection.

The backdrop is a full-blown AI infrastructure arms race. Microsoft is tracking toward roughly $190 billion in 2026 CapEx, largely for AI data centers and GPU compute tied to its OpenAI relationship. Amazon is projected to spend around $200 billion, primarily for AWS data centers and Trainium chips. Meta Platforms guided to $115 billion to $135 billion, nearly double its 2025 spend. This collective spending by the five largest US cloud and AI infrastructure providers, including Oracle, is projected to hit $660 billion to $690 billion in 2026.

AI Data Center
Wikideas1, CC0 via Wikimedia Commons

Wall Street fears short-term free cash flow compression from these investments. However, it continues to reward long-term hyperscaler growth, betting on the eventual payoff from foundational AI models and the silicon that powers them. The delay of Gemini 3.5 Pro, initially planned for June, has intensified investor concerns about whether Google’s latest AI systems are meeting internal performance expectations and if the company is keeping pace with rivals. The model is designed to strengthen Google’s position in AI coding and agentic AI, areas where competition is fierce.

Winners and Losers in the AI Buildout

The immediate winners are clear: the silicon providers. Nvidia, with its new Vera Rubin NVL72 platform and Spectrum-6 Ethernet switch system, continues to dominate the hardware layer, delivering 10 times more AI tokens per watt than its previous Grace Blackwell system. Its Vera Rubin production is ramping up with partners like Google Cloud and Microsoft Azure. This underscores the MANGOS thesis: silicon dominance is paramount.

The hyperscalers, including Google Cloud, Microsoft Azure, and Amazon Web Services, are locked in a high-stakes battle for AI infrastructure leadership. Their willingness to pour hundreds of billions into data centers and custom chips like Google’s TPU v7 Ironwood demonstrates a defensive survival strategy as much as an offensive play for market dominance.

Nvidia GPU
Wikideas1, CC0 via Wikimedia Commons

The potential losers are those who cannot keep pace with this unprecedented CapEx. While Apple reportedly surpassed Nvidia in market capitalization on July 17, reaching $4.88 trillion versus Nvidia’s $4.86 trillion, its strategy of lower AI CapEx is now facing scrutiny. Tesla, for its part, has spent only $2.5 billion of its projected $25 billion in 2026 CapEx, raising questions about its commitment to AI infrastructure compared to its peers.

For Google, the challenge is to demonstrate that its massive investments translate into tangible, revenue-generating AI services and models that can compete effectively. The market is demanding evidence that the CapEx boom will yield sustained growth and profitability, not just a spending spree. The performance of Google Cloud and the trajectory of its AI initiatives will be critical in today’s earnings report and beyond.

Works Cited