AI Infrastructure Shake-Up Hits Chips and Cloud

December 19, 2025 at 07:15 UTC

5 min read

AI infrastructure transformation with chips and cloud servers, highlighting investment trends and tech shifts

Key Points

Google and Meta deepen collaboration to reduce reliance on Nvidia AI chips
Alphabet’s Gemini 3 and Google Cloud post rapid AI-driven user and revenue growth
Nebius lands multibillion-dollar Microsoft and Meta AI data center deals amid valuation debate
Investors reassess AMD, Nvidia and broader chip exposure as AI capex surges into 2026

Google, Meta and Custom Chips Challenge Nvidia’s Grip

Alphabet’s Google is intensifying efforts to reshape the AI chip landscape as demand for large-scale AI computing accelerates. The company is working more closely with Meta Platforms to reduce dependence on Nvidia hardware, developing software changes that make its in-house Tensor Processing Units (TPUs) easier to use with PyTorch, the dominant tool for AI developers. Meta, a major backer of AI tools, is collaborating with Google and gaining greater access to Google’s chip infrastructure, which Google plans to deploy for its own AI services and for cloud customers. This push comes as Nvidia faces mounting competition in AI hardware, even as it remains the leading supplier of data center GPUs and recently became the first company to surpass a $4.5 trillion market capitalization. Nvidia has said a widely discussed potential $100 billion deal with OpenAI is nonbinding and that its sales outlook does not depend on that agreement, emphasizing the strength of its full AI stack. Commentators such as Jim Cramer have highlighted Broadcom as another likely beneficiary of any Alphabet–Meta alignment, citing the chipmaker’s growing role in custom AI silicon and high-end networking under CEO Hock Tan.

Alphabet Builds Out AI Stack With Gemini, Cloud and TPUs

Alphabet’s broader AI strategy extends beyond chip collaborations. The company reported third-quarter revenue growth of 16% year over year, with accelerating paid clicks and strong contributions across businesses including Google Search, YouTube, Android and Waymo. Google Cloud has become a focal point, with cloud services revenue rising 34% year over year as customers use the platform to build generative AI applications and agents. Alphabet’s latest large language model, Gemini 3, has been rated a leading model on several leaderboards and has 650 million active users, which the company is working to monetize. Alphabet is also investing heavily in infrastructure, designing TPUs to train and run its models alongside Nvidia GPUs. Management expects capital expenditures of $91 billion to $93 billion in 2025, up from a prior $85 billion forecast, and plans a significant capex increase in 2026. Separate analysis of Alphabet’s investment narrative links this spending to Gemini-driven AI momentum and long-term renewable power agreements, including a 21-year deal with TotalEnergies in Malaysia and a 30 megawatt solar contract in Kedah, which are intended to support data center expansion in Asia.

Nebius Emerges as High-Growth AI Data Center Partner

AI infrastructure provider Nebius Group has quickly become a notable player in outsourced data centers. The company recently rolled out its AI Cloud 3.1 upgrade, incorporating Nvidia Blackwell Ultra hardware and expanded capacity controls, aiming to address the cost-versus-scalability trade-off for AI workloads. Nebius reported third-quarter revenue of $146.1 million, up 355% from $32.1 million a year earlier, and has secured major contracts. In September it announced a multibillion-dollar deal with Microsoft to provide AI infrastructure services from its Vineland, New Jersey, facility. Last month, it disclosed a three-year, $5 billion agreement with Meta to supply AI infrastructure, reinforcing its position as a key supplier to large technology customers. The company projects electricity needs rising from 220 megawatts this year to between 800 megawatts and 1 gigawatt next year, with contracted demand expected to reach 2.5 gigawatts by the end of 2026. Nebius targets annualized revenue of $7 billion to $9 billion by the end of next year, compared with an expected $550 million in 2025. Despite this growth, Nebius remains unprofitable, posting a third-quarter net loss of $119.6 million and an operating margin of negative 89.1%. It trades at 52 times trailing sales, and social media sentiment has been bearish even as analysts maintain unanimous buy ratings and a consensus price target around $151.50, roughly double recent trading levels.

Investors Reposition Across AI Chips and Infrastructure

The shifting AI landscape is prompting portfolio adjustments among institutional investors and Wall Street analysts. Coatue Management, led by Philippe Laffont, reduced its stakes in Nvidia and Advanced Micro Devices by 14% and 19%, respectively, in the third quarter, while more than tripling positions in Alphabet and Marvell Technology. Nvidia and AMD remain dominant in GPUs used to train large language models, but Coatue’s moves highlight growing interest in other parts of the semiconductor stack, including Alphabet’s TPUs and Marvell’s high-bandwidth memory and networking designs. Separately, Cantor Fitzgerald cut its price target on AMD to $300 from $350 but kept an Overweight rating, citing strong AI tailwinds and what it views as exponential growth in AI infrastructure spending. Cantor noted AMD’s software progress following the Nod.ai acquisition and a partnership with OpenAI to supply six gigawatts of computing capacity, while also flagging intense competition from Nvidia and ambitious data center growth targets. At the same time, questions have surfaced around Nvidia’s reported Blackwell GPU shipments, with social media analysis comparing shipment claims to data center revenue and U.S. power capacity, and investor Michael Burry publicly asking for evidence of warehoused GPUs. These debates underscore how central AI chips and infrastructure have become to both corporate strategies and market expectations heading into 2026.

Key Takeaways

Major cloud platforms are investing in custom chips and software to reduce dependence on Nvidia while still relying on its GPUs for key workloads.
Alphabet is pairing rapid AI model adoption with heavy capex and long-term clean energy deals, tying its AI growth to large, power-intensive data centers.
Nebius illustrates both the opportunity and risk in AI infrastructure, combining hypergrowth and marquee contracts with steep losses and valuation concerns.