2025 saw AI move from experimental models to a strategic force, transforming industries, energy grids, and global competition amid surging investments and next-generation innovations.
2025 will be remembered as the year AI left the laboratory and reshaped industry, infrastructure and geopolitics. What began as a surge of generative models accelerated into a broader wave of multimodal, agentic and physical AI, and the ripple effects , from chips and data centres to energy grids and venture capital , defined a new competitive landscape.
According to a CGTN America interview with Ray Wang, Principal Analyst at Constellation Research, the year saw generative AI mature into systems that handle text, images, audio and video with greater fluency, while a shift to “agentic” AI enabled firms to automate entire workflows and move from proof-of-concept to measurable decision velocity. The same trend fed advances in physical robotics: humanoid and upper‑body control systems made notable technical gains even as their limitations remained real-world constraints on rapid deployment.
Model releases and robotics research underpinned those advances. OpenAI’s GPT-5, released on 7 August 2025, broadened multimodal capabilities and was integrated into ChatGPT and Microsoft Copilot, raising the bar for conversational and multimodal assistants. Industry research into embodied control also advanced: Figure AI’s Helix model, unveiled in February 2025, demonstrated a vision‑language‑action architecture capable of high‑frequency control of a humanoid robot’s upper body, signalling how VLAs might scale to complex physical tasks.
The demand for compute and the raw resources that support it exploded in parallel. Data centres emerged as the largest category of investment in 2025, driving more than $61 billion of deals according to industry tallies cited by Constellation Research. Major cloud and platform providers pushed billions into capacity, chips and power: Microsoft alone committed large capital to neocloud partners and its own facilities, with Tom’s Hardware reporting $33 billion of deals in October 2025 that secured hundreds of thousands of Nvidia GB300 chips and underwrote new hyperscale sites. AMD, meanwhile, pushed an open‑rack, rack‑scale approach with its Helios platform and Instinct MI450 GPUs to improve power efficiency and density, and Oracle committed to deploy large volumes of the new hardware, TechRadar reported.
Energy and real‑world infrastructure became strategic assets. Alphabet’s December 2025 acquisition of Intersect for $4.75 billion signalled a deeper integration of energy and data‑centre strategy; Intersect is developing assets and aims to reach 10.8 gigawatts of capacity by 2028, IT Pro reported, underscoring projections that AI‑driven compute demand could exceed 100 gigawatts within the next decade. Sundar Pichai framed the deal as modernising energy systems and bolstering US competitiveness in AI, while Intersect’s chief executive warned that the ageing electricity grid constrains rapid build‑out and called for co‑located, integrated power and data solutions.
Those infrastructure bets reshaped the market for providers and investors. “NeoClouds” , specialist data‑centre operators such as CoreWeave, Nebius and others , attracted strategic capital and supply commitments as hyperscalers sought flexible capacity and specialised hardware access. Tom’s Hardware detailed Microsoft’s multi‑billion‑dollar Nebius agreement that secured roughly 100,000 GB300 chips for internal use, while CoreWeave’s expansion and other neocloud projects illustrated the market’s fragmentation between hyperscale, specialist and open‑rack strategies.
China’s state and corporate investments accelerated the global race. Constellation Research estimated roughly $125 billion of state funding flowed into Chinese AI, semiconductors and robotics firms in 2025, and mainland players pledged further massive data‑centre investments heading into 2026. Domestic models and open approaches gained traction: Alibaba’s Tongyi and Qianwen, DeepSeek’s R1 and Wudao 3.0, and other Chinese models advanced alongside deployments in smart cities, logistics, healthcare, agriculture and robotics. Constellation’s assessment framed the contest as close: the US leads on certain model and chip advantages, China on power efficiency, open‑source momentum and rapid sectoral integration.
Investor sentiment became more nuanced as the year progressed. Early enthusiasm in 2025 produced sharp capital flows, but by autumn some investors questioned short‑term returns, according to Ray Wang’s analysis. That did not undercut the winners: companies with clear, outcome‑driven AI deployments reported outsized gains, while those “dabbling” trailed. The narrative hardened around a long‑term, winner‑takes‑most logic: firms that commit to sustained investment in data, compute, and integration are positioned to extract exponential efficiency gains.
Profitability and sustainability remained open questions. AP News reported Microsoft’s November 2025 infrastructure partnership with Anthropic and Nvidia , including Anthropic’s promised $30 billion investment in Azure and large chip capacity commitments , as an example of deepening strategic alliances that reshape platform economics. Yet many leading model developers have yet to show profitability, leaving industry watchers to weigh whether continued capital intensity is sustainable or whether business models will adapt to produce returns.
Looking toward 2026, industry observers expect the tension among chips, power and data to intensify. Constellation Research and other industry reporting forecast rapid growth in agentic AI and autonomous workflows, continued investment in TPU and GPU performance and efficiency, and a sharper contest between Western proprietary model ecosystems and Chinese open‑source alternatives for developer mindshare. The emergence of “Data Inc” business models , firms monetising proprietary, continually refreshed datasets at scale , is predicted to accelerate, with data access and sovereignty increasingly central to competitive advantage.
The broader political economy of AI is now clear: the race is as much about real assets and energy as it is about algorithms. Firms and states that secure land, power, cooling and specialised silicon stand to dominate the next phase. 2025 proved that the costs of failing to invest can be existential; 2026 will test which investments yield durable returns and which strategies can turn compute and data dominance into sustainable commercial ecosystems.
- https://raywang.org/blog/2025-12/news-analysis-why-2025-became-age-ai – Please view link – unable to able to access data
- https://www.itpro.com/infrastructure/data-centres/google-drops-usd4-75bn-on-data-centre-and-energy-firm-intersect – In December 2025, Alphabet, Google’s parent company, acquired data centre and energy firm Intersect for $4.75 billion. This strategic move aims to bolster infrastructure to meet the growing demands of artificial intelligence (AI). Intersect, which currently manages or is developing $15 billion in assets, plans to achieve 10.8 gigawatts of energy capacity by 2028. The acquisition underscores the escalating energy requirements to support the rapid expansion of AI and data centre operations, with power demand projected to surpass 106 gigawatts within the next decade. Despite Alphabet’s ownership, Intersect will operate as an independent entity, maintaining its brand and leadership. The two companies intend to collaborate on projects, including a joint data centre and energy site in Texas. Intersect will focus on energy innovation, integrating renewable and backup power sources to enhance electricity access for data infrastructure. CEO Sundar Pichai highlighted Intersect’s role in modernising energy systems and strengthening American AI competitiveness. Concurrently, Intersect CEO Sheldon Kimber criticised the outdated electric grid, advocating for integrated power solutions co-located with data centres. While Google has made strides in clean energy, including a nearly carbon-free facility in Norway, its emissions have grown by 48% since 2019 due to increasing data centre demands. This acquisition reflects the broader trend of significant investments in AI infrastructure by major tech companies to support the burgeoning AI industry.
- https://apnews.com/article/a3e4d6ba75f475eb130d91c81e522f93 – In November 2025, Microsoft announced a significant artificial intelligence (AI) infrastructure partnership with Anthropic and Nvidia, marking a departure from its previous exclusive alliance with OpenAI. As part of the agreement, Anthropic will invest $30 billion in Microsoft’s Azure cloud platform. Nvidia will contribute up to $10 billion in investment to Anthropic, while Microsoft will add up to $5 billion. The collaboration was unveiled just prior to Microsoft’s Ignite developer conference. Anthropic, founded by former OpenAI executives and creator of the chatbot Claude, aims to make Claude the exclusive frontier model available across the three major cloud platforms—Amazon, Google, and Microsoft. Despite rapid development, companies like Anthropic and OpenAI have yet to turn a profit, raising concerns about the sustainability of the AI boom. Nvidia’s contribution also includes providing Anthropic with up to a gigawatt of AI chip capacity to support its computational needs. This partnership reflects shifts in the competitive AI landscape as Microsoft diversifies its AI support and infrastructure, even while retaining a 27% stake in OpenAI. It also continues Microsoft’s strategy to lead in AI infrastructure, tools, and services amidst growing competition.
- https://en.wikipedia.org/wiki/Helix_(AI_model) – Helix, unveiled in February 2025 by Figure AI, is a generalist vision-language-action (VLA) model specifically tailored for humanoid robots. It is the first VLA capable of controlling the entire upper body of a humanoid robot, including arms, hands, torso, head, and fingers, at high frequency. Helix employs a dual-system architecture with two complementary systems trained to communicate in an end-to-end manner. System 2 (S2) is an internet-scale vision-language model specialised in scene understanding and language comprehension, while System 1 (S1) is a visuomotor policy that translates the latent representations produced by S2 into continuous robot actions. This decoupled architecture allows for both broad generalisation and fast low-level control. Helix is trained on approximately 500 hours of robot teleoperation paired with automatically generated text descriptions. The model underscores the ability of VLAs to scale to complex embodiments such as humanoid robots, highlighting advancements in AI’s integration with physical robotics.
- https://en.wikipedia.org/wiki/GPT-5 – GPT-5 is a multimodal large language model developed by OpenAI, released on August 7, 2025. It is the fifth iteration in OpenAI’s series of generative pre-trained transformer (GPT) models. GPT-5 combines reasoning capabilities and non-reasoning functionalities under a common interface, demonstrating state-of-the-art performance on various benchmarks at the time of its release. The model is publicly accessible to users of the chatbot products ChatGPT and Microsoft Copilot, as well as to developers through the OpenAI API. GPT-5’s release signifies a significant advancement in AI’s ability to process and generate human-like text, images, and other modalities, reflecting the rapid progress in AI research and application.
- https://www.tomshardware.com/tech-industry/artificial-intelligence/microsoft-inks-usd33-billion-in-deals-with-neoclouds-like-nebius-coreweave-nebius-deal-alone-secures-100-000-nvidia-gb300-chips-for-internal-use – In October 2025, Microsoft invested a total of $33 billion in ‘neocloud’ data centre providers, with a major $19.4 billion deal struck with Nebius. This agreement secures approximately 100,000 of Nvidia’s latest GB300 chips—believed to be housed in GB300 NVL72 server racks—for Microsoft’s internal use. These chips will support the company’s work on large language models and AI assistants via its CoreAI team, developers of GitHub Copilot and similar AI tools. The Nebius deal is part of a broader investment strategy, where the remaining $13 billion has been allocated to other neocloud partners such as CoreWeave, Nscale, and Lambda. CoreWeave, previously focused on cryptomining, has become a key Nvidia customer and is building a $6 billion data centre in Pennsylvania. Meanwhile, Microsoft is also expanding its own infrastructure, including a massive new 315-acre AI data centre in Wisconsin. This expansion comes amid growing concerns about the environmental and energy impacts of AI centres, with local energy prices up 267% over five years and reports of pollution from AI facilities. Microsoft’s substantial investments in neoclouds and its own infrastructure underscore the escalating demand for AI computing power and the company’s commitment to maintaining a competitive edge in the AI industry.
- https://www.techradar.com/pro/amd-embraces-meta-backed-open-rack-wide-form-factor-with-new-mi450-powered-helios-racks-oracle-is-the-first-big-client-with-a-50-000-gpu-commitment-with-more-to-come – In October 2025, AMD unveiled its ‘Helios’ rack-scale platform at the Open Compute Project (OCP) Global Summit. Built on Meta’s Open Rack Wide (ORW) standard, Helios is designed to optimise power efficiency, cooling, and serviceability for AI systems. The platform utilises AMD’s Instinct MI450 GPUs, based on the CDNA architecture, alongside EPYC CPUs and Pensando networking components. Each GPU features up to 432GB of HBM4 memory and 19.6TB/s of bandwidth, with a full rack achieving 1.4 exaFLOPS in FP8 and 2.9 exaFLOPS in FP4 precision. The ORW-based design supports open standards like DC-MHS, UALink, and the Ultra Ethernet Consortium, enabling improved scalability and reliability. Meta played a pivotal role by contributing the ORW design to the OCP community to spearhead open AI infrastructure. Oracle is the first major client, committing to deploy 50,000 AMD GPUs, a strong indicator of commercial viability. However, widespread adoption will depend on broader industry support. AMD expects volume rollout by 2026 and aims to turn open collaboration into practical, deployable AI hardware solutions. This development reflects the industry’s shift towards open, standardised infrastructure to support the growing demands of AI applications.
Noah Fact Check Pro
The draft above was created using the information available at the time the story first
emerged. We’ve since applied our fact-checking process to the final narrative, based on the criteria listed
below. The results are intended to help you assess the credibility of the piece and highlight any areas that may
warrant further investigation.
Freshness check
Score:
10
Notes:
The narrative is original and published on December 29, 2025, with no prior appearances found. The content is based on a recent interview with Ray Wang, Principal Analyst at Constellation Research, conducted by CGTN’s Sally Ayhan. This indicates high freshness and originality. The article includes updated data and insights, justifying a higher freshness score. No discrepancies or recycled content were identified.
Quotes check
Score:
10
Notes:
The direct quotes from Ray Wang in the narrative are unique to this publication, with no earlier matches found. This suggests the content is potentially original or exclusive. No variations in quote wording were noted.
Source reliability
Score:
10
Notes:
The narrative originates from Ray Wang’s personal blog, ‘A Software Insider’s Point of View,’ which is a reputable platform for his analyses and insights. Ray Wang is the founder and CEO of Constellation Research, a well-established technology research and advisory firm. This association enhances the credibility of the content.
Plausability check
Score:
10
Notes:
The claims made in the narrative align with recent developments in AI, including advancements in generative and agentic AI, significant infrastructure investments, and China’s rapid AI development. These points are consistent with other reputable sources and reflect current industry trends. The language and tone are appropriate for the subject matter, and the structure is coherent without excessive or off-topic details.
Overall assessment
Verdict (FAIL, OPEN, PASS): PASS
Confidence (LOW, MEDIUM, HIGH): HIGH
Summary:
The narrative is original, timely, and authored by a credible source. The content is consistent with current AI industry developments and is presented in a clear and appropriate manner. No significant issues were identified, indicating a high level of reliability.