Technology
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v5 ยท 4,525 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v5 ยท 4,525 words
Add missing citations, update stale details, or suggest a clearer explanation.
Technology is one of the largest application domains for artificial intelligence, covering enterprise IT operations, cloud computing, cybersecurity, software development, networking, semiconductor manufacturing, and the data center buildout that powers modern machine learning. Across these areas, AI has shifted from a research curiosity to a foundational layer of products and infrastructure. Hyperscale cloud providers offer hosted foundation models through unified developer platforms, observability vendors embed assistants in incident response workflows, security operations centers run autonomous triage agents, and chip designers compete to deliver the highest training and inference throughput per watt. As of mid 2026, technology is also where the largest dollar commitments to AI infrastructure are being made, including the Stargate joint venture between OpenAI, Oracle, SoftBank, and MGX, which targets up to $500 billion in United States data center investment by 2029.[5]
This gateway article surveys the major categories of AI for technology and IT, lists the leading vendors and products in each, and links out to dedicated wiki pages where they exist. It complements other gateways such as Healthcare, Finance, and Education.
See also: Technology ChatGPT Plugins
AI has become embedded in nearly every layer of the modern IT stack. Software developers use code assistants such as GitHub Copilot to write, refactor, and review code. Site reliability engineers use observability platforms with built-in large language model assistants to summarize alerts and propose remediations. Security teams run autonomous agents that triage suspected intrusions across millions of endpoints. Network operators use AI to predict capacity, detect anomalies, and automate configuration changes across thousands of switches.
Three forces shape the current landscape. First, the rise of generative AI has put natural-language interfaces on top of nearly every operational tool, lowering the skill barrier for routine tasks. Second, the move from predictive analytics to agentic AI has shifted the boundary between human and machine work, with agents now triggering remediations, opening change tickets, and taking containment actions on their own. Third, the underlying compute, memory, and power infrastructure has become a strategic constraint, leading to multi-year contracts for GPU capacity, dedicated power purchase agreements, and large public investments such as the Stargate Project.
As agents proliferated across the IT stack, the lack of a common way to connect them to tools and data became a bottleneck. The Model Context Protocol (MCP), an open standard introduced by Anthropic in November 2024, addressed this by defining a uniform interface between AI applications and external systems such as file stores, databases, and SaaS APIs. Adoption was unusually fast for an infrastructure standard. OpenAI adopted MCP across its products in March 2025, Google DeepMind confirmed support in Gemini the following month, and Microsoft wired it into Copilot Studio and Visual Studio Code.[19] By late 2025 there were more than 10,000 public MCP servers and the protocol's software development kits were being downloaded tens of millions of times per month.[20] On December 9, 2025, Anthropic donated MCP to the newly formed Agentic AI Foundation, a directed fund under the Linux Foundation co-founded with Block and OpenAI and backed by Google, Microsoft, Amazon Web Services, Cloudflare, and Bloomberg, cementing its status as vendor-neutral infrastructure.[21] A complementary effort, the Agent2Agent (A2A) protocol introduced by Google in April 2025 and donated to the Linux Foundation in June 2025, standardizes communication between independent agents rather than between an agent and its tools.[25]
The three United States hyperscalers, along with IBM and Oracle, dominate the enterprise AI platform layer. Each offers a managed service that combines hosted models, data integration, agent frameworks, vector search, and governance tooling.
| Platform | Vendor | Notable models and features | Notes |
|---|---|---|---|
| Amazon Bedrock | Amazon Web Services | Hosts Claude, Llama, Cohere, AI21, Stability AI, Mistral, and Amazon Titan and Nova families. Bedrock AgentCore for agent orchestration. | Multi-vendor model marketplace; serverless invocation with provisioned throughput option. |
| Azure AI Foundry | Microsoft | Direct access to OpenAI GPT models, Llama 3, and a model catalog spanning open-weight and proprietary options. Includes Foundry Agent Service. | Tight integration with Microsoft 365, Entra ID, and Azure Synapse data plane. |
| Vertex AI | Google Cloud | Native Gemini family, Model Garden of third-party and open models, Vertex AI Agent Builder, and a managed vector search service. | Strong fit for organizations standardized on BigQuery and Google data services. |
| Watsonx | IBM | IBM Granite models, watsonx.ai foundation model studio, watsonx.data lakehouse, and watsonx.governance. | Watsonx Orchestrate is also offered on Oracle Cloud Infrastructure as part of a 2025 partnership.[10] |
| OCI Generative AI | Oracle | Hosts Cohere, Llama, and IBM Granite models, plus the OCI Generative AI Agents service inside Oracle Fusion Applications. | Bare metal GPU clusters on RDMA fabric for training and inference. |
All five platforms now support similar primitives: managed model endpoints, retrieval pipelines built on top of vector databases, agent runtimes with tool calling, content moderation, and audit logging tied to identity systems. The differentiator for buyers is typically the surrounding data platform, the contractual relationship, and which proprietary models are available natively.
Observability and incident management vendors have spent two years rebuilding their products around AI assistants and autonomous agents. The umbrella term AIOps, originally coined by Gartner, covers any use of AI to help operate IT systems, from log anomaly detection to automated runbook execution.
| Tool | Vendor | Description |
|---|---|---|
| PagerDuty AIOps | PagerDuty | Alert correlation, noise reduction, and automated incident triage. Integrates with PagerDuty Operations Cloud for runbook automation. |
| Bits AI | Datadog | Suite of agents for SRE, development, and security workflows. Announced at DASH 2025. Provides root cause analysis, code suggestions, and incident summaries. |
| Splunk AI Assistant | Splunk (Cisco) | Natural-language query authoring for SPL, dashboard generation, and detection content for Splunk ITSI and Splunk Enterprise Security. |
| Davis AI | Dynatrace | Causal AI engine for root cause analysis, problem detection, and predictive alerting across the Dynatrace observability platform. |
| DevOps Guru | Amazon Web Services | Operational anomaly detection across AWS services, with prescriptive recommendations and integration into Systems Manager. |
| GitHub Copilot | Microsoft and GitHub | In-IDE code completion, chat, and Copilot Workspace for issue-to-pull-request workflows. |
These tools share a common pattern: they ingest telemetry that already exists in the customer environment, apply LLM-based reasoning to summarize incidents and surface root causes, and increasingly take action through integrations with ticketing, runbook, and configuration management systems. The 2024 to 2026 trend has been a steady shift from suggestion to action, with vendors describing their assistants as agents that can isolate hosts, roll back deployments, and open change tickets without waiting for explicit approval on lower-risk operations.
Code assistance is the most widely adopted of these capabilities. GitHub reported that Copilot had reached roughly 20 million all-time users by mid 2025 and was in use at around 90 percent of the Fortune 100, and the product has since expanded from in-IDE completion to an autonomous coding agent that can take a GitHub issue and open a pull request with the implementation.[22]
Security operations have become one of the largest beachheads for agentic AI. Modern security operations centers face alert volumes far beyond what human analysts can review, and AI-native platforms have begun to autonomously triage, investigate, and respond to threats.
| Tool | Vendor | Description |
|---|---|---|
| Charlotte AI | CrowdStrike | Generative AI assistant on the Falcon platform. Performs investigation, query authoring, and detection tuning. CrowdStrike has expanded Charlotte into autonomous detection and triage. |
| Purple AI | SentinelOne | AI analyst on the Singularity Data Lake and AI SIEM. Converts natural-language prompts into investigations with timelines and suggested actions. |
| Microsoft Security Copilot | Microsoft | Reasoning over Microsoft Defender, Sentinel, Entra, Intune, and Purview signals. Designed for native fit with Microsoft security stack customers. |
| Cortex XSIAM | Palo Alto Networks | AI-driven security operations platform. Cortex XSIAM 3.0, released in April 2025, added Cortex Exposure Management and advanced email security. Surpassed $1 billion in cumulative bookings in 2025, the fastest-growing product in the company's history.[12] |
| Darktrace | Darktrace | Self-learning unsupervised AI for network and email anomaly detection. Autonomous Response acts to contain threats based on learned baselines. |
| Cisco AI Defense | Cisco | Announced January 2025 to secure enterprise AI applications. Covers model and prompt validation, runtime monitoring, and threat protection for generative AI deployments.[11] |
The practical effect is that triage and investigation work that once required senior analysts is increasingly handled by AI agents, with humans focused on policy, oversight, and high-impact incidents. Autonomous response, where the system isolates endpoints or blocks accounts without waiting for approval, is now standard at the high end of the market.
Enterprise IT support has been transformed by AI assistants that combine knowledge bases, ticket data, and access to operational systems. The pattern is similar across vendors: a generative AI layer sits on top of an existing service desk, answers user questions, and increasingly executes routine workflows.
| Product | Vendor | Description |
|---|---|---|
| Microsoft Copilot | Microsoft | Copilot in Microsoft 365 acts as an enterprise assistant across Word, Excel, Outlook, Teams, and SharePoint. Microsoft Copilot Studio lets organizations build custom copilots and agents. |
| Now Assist | ServiceNow | AI assistant embedded in the Now Platform. Summarizes incidents, drafts resolution notes, generates code for Flow Designer, and integrates with Microsoft Copilot inside Microsoft Teams.[17] |
| Zendesk AI and Copilot | Zendesk | Automation of front-line support through AI agents and an agent-side Copilot. In 2025, Zendesk became a launch partner for Microsoft Agent 365 to integrate with Microsoft 365 environments. |
The practical benchmark for these tools is deflection rate, the share of tickets resolved without involving a human agent, alongside resolution time and customer satisfaction. ServiceNow, Zendesk, and Microsoft each publish customer case studies showing deflection rates of forty to seventy percent on routine categories such as password resets, access requests, and procurement questions.
AI accelerators are now one of the largest segments of the semiconductor industry, and the design and supply of these chips is itself a major story in the technology sector.
Nvidia remains the dominant designer of AI training GPUs. The current flagship is the Blackwell generation, announced at GTC in March 2024 and entering volume production through 2025 and 2026.[1][2]
| Product | Vendor | Description |
|---|---|---|
| B200 | Nvidia | Dual-die Blackwell GPU on TSMC 4NP with 208 billion transistors and up to 192 GB of HBM3e memory. Up to 20 petaflops of sparse FP4 compute. NVLink 5.0 at 1.8 TB/s. |
| GB200 NVL72 | Nvidia | Liquid-cooled rack system with 36 Grace Blackwell Superchips, totaling 72 Blackwell GPUs and 36 Grace CPUs over fifth-generation NVLink. Designed for trillion-parameter training and inference. |
| B300 (Blackwell Ultra) | Nvidia | Refreshed Blackwell tier with higher memory and compute targeted at inference-heavy deployments. Announced at GTC in March 2025 and shipping in the second half of 2025. |
| GB300 NVL72 (Blackwell Ultra) | Nvidia | Liquid-cooled rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs, delivering about 1.5x the AI performance of the GB200 NVL72. Microsoft Azure brought the first supercomputing-scale production cluster online in October 2025, linking more than 4,600 GB300 GPUs over Quantum-X800 InfiniBand for OpenAI.[23] |
| Instinct MI300X | AMD | 2023 generation accelerator with up to 192 GB of HBM3, widely deployed for training and serving large models. |
| Instinct MI325X | AMD | 2024 refresh with HBM3e and higher memory bandwidth. |
| Instinct MI350 series (MI350X, MI355X) | AMD | Released in June 2025. AMD reports up to 4x faster AI compute and up to 35x higher inference throughput versus the MI300 series, with up to 288 GB of HBM3e.[3] |
| Instinct MI400 series | AMD | Full lineup (MI455X, MI450, MI430X) unveiled at CES in January 2026 and the first GPUs built on TSMC's 2 nm class N2 node, with up to 432 GB of HBM4 and 19.6 TB/s of bandwidth. Anchors the Helios rack-scale platform of 72 MI455X accelerators, on track for the second half of 2026 to compete with Nvidia's GB300 and Rubin systems.[24] |
| Apple Silicon M5 | Apple | Announced October 2025 on TSMC third-generation 3 nm. 16-core Neural Engine, GPU with neural accelerator in each core, up to 153 GB/s unified memory bandwidth. Up to 4x peak GPU compute for AI versus M4.[4] |
A growing set of specialized vendors target specific workloads, especially inference and large-scale training.
| Vendor | Product | Notes |
|---|---|---|
| Cerebras | WSE-3 wafer-scale engine, CS-3 system | 4 trillion transistors, 900,000 AI cores, 44 GB on-chip SRAM, 21 PB/s memory bandwidth, on TSMC 5 nm. CS-3 in a 15-kilowatt envelope.[18] |
| Groq | LPU and GroqCloud | 230 MB SRAM per chip and 80 TB/s on-die bandwidth. Reported 300 tokens per second on Llama 2 70B. Raised $750 million in September 2025 at a $6.9 billion valuation. |
| Tenstorrent | Wormhole, Blackhole, Grayskull | RISC-V-based AI accelerators led by chip architect Jim Keller. Backed by Hyundai, Kia, LG, and Samsung. |
| Etched | Sohu ASIC | Transformer-specific accelerator targeting inference for transformer models. |
| SambaNova Systems | DataScale, Reconfigurable Dataflow Unit | Targeted at fine-tuning and serving open-source models in regulated and on-prem environments. |
TSMC is the dominant foundry for advanced AI silicon. Monthly 3 nm wafer output reportedly stood near 120,000 to 130,000 wafers at the end of 2025 and is projected to climb toward 180,000 by the end of 2026, an increase of more than 40 percent year on year.[16] The 2 nm node (N2), TSMC's first gate-all-around process, entered high-volume manufacturing in the fourth quarter of 2025 and is expected to ramp toward 100,000 wafers per month during 2026, with demand already exceeding the initial allocation. AI accelerators from Nvidia, AMD, and others are the primary driver of this expansion; AMD's MI400 series is among the first to adopt N2. CoWoS advanced packaging and HBM memory remain among the tightest links in the supply chain.
Edge AI runs models close to where data is collected, reducing latency, bandwidth, and dependency on cloud connectivity. The category covers everything from camera-grade vision processors to robotics modules and on-device assistants in laptops and phones.
| Platform | Vendor | Description |
|---|---|---|
| Jetson AGX Orin | Nvidia | Up to 275 TOPS of AI performance, 12-core Arm Cortex-A78AE CPU, 2048-core Ampere GPU. Used in robots, drones, and industrial vision. |
| Jetson Thor | Nvidia | Robotics-class platform built on Blackwell-class compute for humanoid and physical AI applications. |
| Apple Neural Engine | Apple | 16-core Neural Engine in the M5 generation. Drives on-device AI features in macOS and iPadOS. |
| Qualcomm AI Engine | Qualcomm | Hexagon NPU plus Adreno GPU and Kryo CPU. The Dragonwing IQ8 processor at the heart of the Arduino Ventuno Q is rated at 40 TOPS. |
| Hailo-8, Hailo-15 | Hailo | Specialized edge AI processors known for high inference throughput at very low power, used in smart cameras, NVRs, and industrial vision. |
| Coral Edge TPU | Small form-factor accelerator for vision and audio inference at the edge. |
In parallel, AI is moving onto laptops and phones through neural processing units integrated into modern CPUs from Intel, AMD, Apple, and Qualcomm. The Copilot+ PC category, defined by Microsoft in 2024, requires at least 40 TOPS of NPU performance and a minimum amount of memory and storage, and has driven a wave of new device launches.
Quantum computing is a separate technology domain, but it has growing intersections with AI. Vendors are exploring hybrid pipelines where classical AI handles preprocessing, large-scale optimization, and post-processing while quantum hardware tackles narrowly defined subproblems.
| Project | Organization | Notes |
|---|---|---|
| Heron | IBM | Family of fixed-frequency superconducting processors with 133 or 156 qubits and tunable couplers. Used in joint experiments with the Fugaku supercomputer at RIKEN and bond trading research at HSBC. |
| Willow | Google Quantum AI | 105-qubit superconducting processor announced December 2024. Demonstrated below-threshold quantum error correction with logical error rates around 0.14% per cycle on a 7x7 patch. T1 times approaching 100 microseconds. In October 2025, Google reported the Quantum Echoes benchmark on Willow.[8][9] |
| Majorana 1 | Microsoft | Topological qubit chip announced February 2025, based on a new class of materials. The claim has drawn academic skepticism, with some researchers arguing decoherence times are too short to support practical qubits. |
| Quantinuum H-series | Quantinuum | Trapped-ion processors with high gate fidelities and a separate roadmap toward error-corrected logical qubits. |
| IonQ Forte | IonQ | Trapped-ion systems offered through major cloud platforms. |
For most enterprise buyers in 2026, quantum remains exploratory. The practical implication is to monitor developments in post-quantum cryptography, since classical machine learning workloads will continue to run on conventional accelerators for the foreseeable future.
Network operators have been some of the earliest adopters of machine learning for anomaly detection, capacity forecasting, and intent-based configuration. The leading vendors now combine these features with generative AI assistants for operations.
| Product | Vendor | Description |
|---|---|---|
| Cisco AI Defense | Cisco | Announced January 2025. Secures the development and deployment of AI models, applications, and agents inside enterprise environments. |
| Cisco Networking Cloud | Cisco | Unified management plane across Catalyst, Meraki, and ThousandEyes with AI-driven assurance and remediation. |
| CloudVision with AVA | Arista Networks | Network operations platform with the Autonomous Virtual Assist analytics engine for telemetry-driven monitoring and anomaly detection. |
| Mist AI | Juniper Networks | AI-driven Wi-Fi, wired, and SD-WAN operations with the Marvis virtual network assistant. The pending HPE acquisition of Juniper closed in early 2026, combining Mist with Aruba. |
| Nile Services Cloud | Nile | Network-as-a-service offering with AI-driven service assurance for campus networks. |
Specialty AI networking is also driving change in the data center. Vendors compete on Ethernet-based AI fabrics that compete with Nvidia NVLink and InfiniBand, with Arista, Broadcom, and Cisco all targeting back-end training networks for hyperscale and enterprise AI clusters.
AI workloads have driven a wave of investment into vector databases, data lakehouses, and unified analytics platforms. The basic pattern is that organizations want to apply retrieval-augmented generation and analytics over their own proprietary data, which requires storage layers that can serve embeddings, structured records, and documents to LLM applications at scale.
| Platform | Vendor | Notes |
|---|---|---|
| Snowflake Cortex | Snowflake | Fully managed service exposing LLMs (Llama, Mistral, Snowflake Arctic) through SQL functions. AI_CLASSIFY, AI_TRANSCRIBE, AI_EMBED, and AI_SIMILARITY became generally available in late 2025. Cortex Search competes with managed retrieval services.[13] |
| Mosaic AI | Databricks | Successor to MosaicML. Provides Mosaic AI Vector Search, Mosaic AI Agent Framework, and tooling for fine-tuning and serving custom models on the Databricks lakehouse. |
| Pinecone | Pinecone | Fully managed vector database. Dedicated Read Nodes for steady high-throughput query workloads were introduced in late 2025. |
| Weaviate | Weaviate | Open-source vector database with native hybrid search across vector similarity, keyword matching, and metadata filtering. |
| Milvus | Zilliz | Open-source vector database designed for billions of vectors. Supports IVF, HNSW, and GPU acceleration. |
| Chroma | Chroma | Embedding database with a Python-first developer experience. Popular for prototyping RAG applications. |
| Qdrant | Qdrant | Open-source vector search engine with rich filtering and high-throughput indexing. |
The trend in 2025 and 2026 has been convergence. Snowflake, Databricks, Microsoft, Google Cloud, and AWS all now provide native vector indexing in their primary analytics platforms, while specialty vector vendors emphasize performance, hybrid search, and lower total cost at very high scale.
The demand for AI training and inference has created a large and visible buildout of new data centers. Several public sources track the magnitude of the change.
The response has included long-term power purchase agreements with nuclear, geothermal, and solar developers, as well as on-site generation. Several hyperscalers have signed deals to restart or extend the lives of nuclear plants. Liquid cooling, high-density racks (such as the Nvidia GB200 NVL72), and ARM-based control planes have become standard in new builds focused on AI.
Stargate is the most visible single project in the current data center cycle. Announced January 21, 2025, by President Donald Trump alongside Sam Altman of OpenAI, Larry Ellison of Oracle, and Masayoshi Son of SoftBank, Stargate LLC plans to invest up to $500 billion in United States AI infrastructure by 2029, with $100 billion to be deployed initially.[5][7] Investors include OpenAI, SoftBank, Oracle, and the investment firm MGX. The flagship campus is in Abilene, Texas, with five additional United States sites announced through 2025.[6] By late 2025, Stargate had committed to nearly 7 GW of planned capacity and over $400 billion in investment, putting it on track to reach the announced 10 GW and $500 billion targets ahead of schedule. International expansion includes a UAE Stargate facility expected to open in 2026.
See also: Technology ChatGPT Plugins
| Plugin | Image | Model | Release Date | Description | Available | Working |
|---|---|---|---|---|---|---|
| Speechki (ChatGPT Plugin) |  | GPT-4 | May 20, 2023 | The easiest way to convert texts to ready-to-use audio - download link, audio player page, or embed! | Yes | Yes |