As the year draws to a close, the market is no longer asking if AI will work - but who possesses the dry powder to build the "Stargate" required to power it? With global AI infrastructure boom spending already surpassing the US$1 trillion milestone this year alone, Silicon Valley, governments and other enterprises across the globe are throwing the proverbial bus at winning, or at least being a part of, this decade's enormous technological change.
From US$300 billion data centre projects to the "de-Nvidification" of custom silicon, here are the top 10 most significant artificial intelligence deals from the last 90 days that matter.
1. The $300 billion “Stargate”
September's quantification of the landmark $300 billion partnership between Oracle and OpenAI stands as the most ambitious infrastructure commitment in the history of the digital age.
The massive Oracle and OpenAI partnership effectively turns Oracle into a primary cloud heavyweight while providing OpenAI with the unprecedented scale required for its next-generation models.
Spanning five years, the agreement aims to supply vast amounts of computing power through 30 gigawatts of capacity, representing a shift toward utility-scale tech operations as Oracle follows its instinct with record-breaking hardware orders.
Industry analysts suggest the alliance is a strategic move to diversify cloud platform partnerships and reduce OpenAI's total reliance on Microsoft’s Azure ecosystem.
"We are building the largest AI training clusters in the world," Oracle Chairman Larry Ellison said.
"Our partnership with OpenAI ensures the next generation of models will have the power they need to change the world."
2. OpenAI’s PBC pivot and Microsoft’s $135 billion equity stake
OpenAI finalised its transition from a non-profit-controlled entity into a Public Benefit Corporation (PBC) on 28 October, a move designed to attract more traditional capital and simplify its corporate governance.
Microsoft’s cumulative investments were converted into a 27% equity stake in the for-profit entity, following the OpenAI corporate restructuring which valued the holding at approximately $135 billion.
The new structure allows the company to pursue massive funding rounds without the previous constraints of its non-profit roots, a shift recently detailed when Azzet spent time reverse engineering the week in business.
By removing the "profit cap" for early backers, the deal makes OpenAI significantly more attractive for private equity and sovereign wealth funds that require standard equity models for long-term holds.
"Our partnership with OpenAI has never been stronger," Microsoft CEO Satya Nadella said.
"This new structure aligns our long-term interests as we lead the era of platform-shift toward AI."
3. Nvidia’s $20 billion "talent extraction" of Groq
A definitive $20 billion strategic transaction with Groq on 25 December signalled Nvidia’s intent to neutralise its most potent rival in the low-latency inference market.
Rather than a traditional merger, Nvidia paid for a perpetual licence to Groq’s LPU architecture and the transfer of its core engineering team to bolster its own hardware division.
Such a massive "acqui-hire" brings Groq founder Jonathan Ross and other key leadership into Nvidia to scale high-speed inference technologies across the global data centre fleet.
The deal allows Nvidia to eliminate a potential competitor while obtaining high-speed SRAM-based technology that bypasses traditional memory bottlenecks, securing its dominance as it chases a $5 trillion market cap.
"Groq has built a remarkable architecture for the inference era," Nvidia CEO Jensen Huang said.
"Bringing their expertise into our ecosystem allows us to push the boundaries of real-time reasoning."
4. Amazon’s $38 billion OpenAI investment & Trainium3 breakthrough
Amazon has challenged the status quo with a gargantuan $38 billion commitment to OpenAI, though the arrangement is heavily predicated on the use of its own custom silicon.
OpenAI will adopt AWS Trainium3, a 3nm chip offering significantly better price-performance than current industry standards, as detailed in the Azzet report on how AI spending splits the week in business.
The Amazon and OpenAI investment involves providing hardware in exchange for equity, diversifying OpenAI's hardware reliance away from Nvidia’s expensive Blackwell systems.
AWS recently announced that the new Trainium3 silicon delivers up to 4.4 times more compute performance than previous generations, potentially resetting the unit economics of AI training for the entire industry.
"Custom silicon is the bedrock of our AI strategy," AWS CEO Matt Garman said.
"OpenAI choosing Trainium3 is a testament to the performance and cost advantages we are delivering for the most demanding workloads."
5. Anthropic’s cloud trifecta and $30 billion Microsoft deal
Anthropic, the "safety-first" rival to OpenAI, saw its fortunes soar on 19 December after securing a massive strategic partnership with Microsoft and Nvidia to scale its Claude models.
Backed by Iconiq Capital and Fidelity, the strategic move that saw Anthropic hit a cloud trifecta cements its status as the primary alternative for enterprise clients wary of OpenAI’s governance.
The company is currently approaching a $7 billion run-rate as demand for its Claude 4 models scales globally across corporate and government sectors.
Recent reports highlight a surging valuation for Anthropic as it prepares for even larger-scale infrastructure deployments in early 2026, including potential new data centres in Europe to meet sovereign data requirements.
"Our goal is to build reliable, steerable, and interpretable AI," Anthropic CEO Dario Amodei said.
"This expanded partnership provides the compute we need to bring safety-first intelligence to everyone."
6. ServiceNow’s $7.75 billion acquisition of Armis
Software giant ServiceNow executed its largest acquisition ever on 23 December, snapping up cybersecurity leader Armis for $7.75 billion.
The deal integrates real-time threat detection into ServiceNow’s platform to create a new AI control tower that can autonomously manage security risks across complex enterprise networks.
Targeting the security of connected devices across the U.S. and global industrial sectors, ServiceNow joins the broader U.S. tech force to modernise federal infrastructure.
Analysts believe the move will triple the market opportunity for ServiceNow in the risk and compliance sectors by 2027, as companies seek to protect their AI-driven workflows from increasingly sophisticated attacks.
"Cybersecurity is the fundamental challenge of the AI era," ServiceNow CEO Bill McDermott said.
"By bringing Armis into the fold, we are creating the world's first AI-powered platform to protect every asset across the enterprise."
7. Disney and OpenAI’s $1 billion "Sora" IP licensing deal
A landmark bridge between Hollywood and Silicon Valley was formed on 11 December when Disney agreed to license its character library to OpenAI's video generation platform, Sora.
Under the terms of the agreement where Disney allows OpenAI to license its iconic roster for $1 billion, users can now generate short social videos featuring over 200 legendary characters within strict guardrails.
The Disney and Sora licensing deal establishes a blueprint for how legacy IP holders can monetise generative AI safely while maintaining control over their brand equity in the digital age.
Disney will also become a major customer of OpenAI, deploying custom APIs to streamline its internal animation pipelines and creative storyboarding processes for future Star Wars and Marvel content.
"Disney has always been at the forefront of storytelling and technology," Disney CEO Bob Iger said.
"This partnership allows us to explore new creative frontiers while ensuring our intellectual property is protected."
8. Meta’s $72 billion "all-in" capex strategy
Meta’s Q3 earnings on 29 October revealed a staggering increase in capital expenditure, with 2025 spending now forecast to hit a record-breaking $72 billion.
Such a massive reinvestment of ad profits into a gigawatt-scale data centre project is clear evidence of the Meta invests $1.5 billion strategy in new facilities to support its growing model library.
Mark Zuckerberg is betting the company on "Personal Superintelligence," creating a formidable barrier to entry for smaller rivals who cannot match the sheer scale of Meta's physical infrastructure.
The record Q3 financial results underscore the shift toward long-term infrastructure assets over short-term earnings, as Meta seeks to own the default AI interface for billions of users.
"We are playing the long game with AI infrastructure," Meta CEO Mark Zuckerberg said.
"The investments we're making today will define the next decade of social connection and personal intelligence."
9. Apple’s $1 billion Gemini-Siri licensing pact
Proving that even the world’s most valuable company can’t build everything in-house, Apple finalised a deal on 5 November to license Google’s Gemini model for its mobile ecosystem.
Apple will pay Google approximately $1 billion annually to power its overhauled Siri digital assistant, as the company addresses previously delayed Siri improvements to remain competitive.
The Apple and Google partnership ensures that "Advanced Intelligence" features remain current on the iPhone 17 without delaying the hardware launch cycle during a critical year for the product line.
By licensing the Gemini model, Apple buys critical R&D time while developing its own in-house 1 trillion parameter model for future privacy-focused on-device processing.
"We are excited to bring Google’s best-in-class models to the iPhone," Apple CEO Tim Cook said.
"This partnership enhances our intelligence features while staying true to our core values of privacy and security."
10. Cisco’s acquisition of NeuralFabric
Networking titan Cisco moved to secure the "Enterprise AI" layer by acquiring NeuralFabric on 13 November for an undisclosed multi-billion-dollar sum.
Acquiring the modular model-development platform brings a valuable asset into Cisco’s portfolio, allowing businesses to build tailored Small Language Models (SLMs) that run on their own hardware.
The move allows Cisco to offer compliant on-premise infrastructure to highly regulated sectors like banking and healthcare, following the broader trend as Alphabet buys data centre partner Intersect to boost its own capacity.
Integration will strengthen the foundation of AI Canvas, Cisco's generative collaborative workspace, positioning the company as the primary facilitator of "Private AI" for the world's most sensitive data.
"Enterprise AI must be secure and private by design," Cisco CEO Chuck Robbins said.
"NeuralFabric gives our customers the tools they need to build and deploy AI safely on their own terms."
