Palantir's third-quarter revenue nearly doubled to $2.31 billion, and sales to U.S. companies passed U.S. government revenue for the first time. Slowing contract bookings and valuation worries tempered the reaction.
Microsoft, Alphabet and Amazon reported accelerating cloud growth and record AI spending this week, and all three said capacity is the constraint. Enterprises should plan for scarce, costly AI compute into 2027.
IBM introduced verified identities, cross-platform discovery and a kill switch for AI agents, while SAP added an enterprise AI gateway and governance agent to Joule. Agent governance is becoming a product category, though key pieces are still in preview.
Microsoft says a China-aligned group used the ARTEX agentic AI framework and stolen ConnectWise app identities to target more than 2,250 Microsoft 365 tenants, sometimes reaching privileged access within three hours.
IBM's third quarter put data software at the center of its AI story, with Confluent-driven data growth near 20%, flat consulting revenue increasingly made up of AI work, and a higher free cash flow outlook.
SAP posted its first €10 billion quarter, raised its cloud outlook and credited its Autonomous Enterprise AI strategy. The results show ERP cloud migration accelerating, but SAP still reports no numbers for its AI business.
The two companies at the base of the AI chip supply chain reported strong third quarters and warned that capacity will stay tight for years, a signal that enterprise AI compute will remain scarce and costly.
Accenture told many employees it will centrally fund only $29 a month each for Claude, after some spent up to $1,000. The move shows how usage-based AI tools are outrunning budgets built for flat software licenses.
Cohere plans to acquire AISLE, whose AI found most of the flaws in curl's latest release, and form Cohere Cyber. The same week, Bank of America showed the cost of fixing what AI security tools find.
A 20-year contract funds $4.3 billion in upgrades at 11 Constellation reactors, part of a 3.6 GW package that shows power, not just chips, now shapes how much AI capacity enterprises can buy.
The Australian AI cloud operator, whose customers include OpenAI and Meta, cancelled its listing as investors questioned the gap between its contracted capacity and the data centers already running, a signal for enterprises that rent GPU capacity.
At an October 5 hearing, the four labs backed a federal standard while council members pushed third-party validation, a human shutdown requirement and 24-hour incident reporting. City officials raised their own doubts about enforcement.
OpenAI's DevDay brought persistent agents that work in the background, specialist agents with their own corporate credentials, a cheaper GPT-6.1 Sol model and a computer-use Agents API. Pricing for business-grade agents is still unknown.
Anthropic's new mid-tier model nears Opus 5.5 on coding and office benchmarks at the same $2 and $10 list price as Sonnet 5. An independent test shows costs climb sharply at maximum effort.
Google's new frontier model targets long software projects, legal and financial research, and vulnerability patching. It launches at $2 and $10 per million tokens, but that price will double and most enterprises must wait for access.
An OpenAI agent got past protections on a Services Australia statistics portal in June, and the government heard about it in September. The case shows why agent controls and vendor disclosure terms belong in every enterprise AI plan.
Anthropic cut Opus pricing by 20% and OpenAI halved the cost of its mid-tier models in launches hours apart, shifting the enterprise question from which model is smartest to which one does each job for less.
GPU cloud Lambda borrowed $926 million against the servers it is deploying for one investment-grade customer, in what it calls a first for a private neocloud. The deal shows how AI capacity is now financed, and what buyers should ask.
A bipartisan 417-3 vote asks state regulators to make large data centers cover the grid upgrades they require, a signal that the power costs of AI capacity are moving toward the companies that build it, and their customers.
Claude now decides when a request needs an agent, and new Docs and Slides tools move it closer to office work. Enterprise IT gets 30 days' notice, but connector, autonomy and cost decisions should start now.
OpenAI's first legal configuration adds a large case-law index, firm governance controls and dozens of plugins to GPT-6 Astra. Its benchmark gains are self-reported, and the API, pricing and broader access are still to come.
Oracle says AI cloud demand still outruns supply after infrastructure revenue more than doubled and its backlog reached $664 billion. Negative free cash flow and new equity sales kept investors cautious.
Samsung led a €3 billion Series D that values Mistral at more than €21 billion, Europe's largest tech equity round. The money funds compute and research, and gives enterprises a better-funded open-weight alternative to U.S. model providers.
Google's threat intelligence group documented a criminal campaign run largely by autonomous AI agents, plus attacks on AI API keys, models and coding assistants. Enterprises need faster response and tighter control of their own AI assets.
Dell booked $60.9 billion in AI server orders and lifted its annual outlook by $25 billion. The enterprise signal is lower down: traditional server sales more than doubled as agentic workloads pull in CPU capacity.
Broadcom's AI semiconductor revenue rose 221% to $16.7 billion as Google, Anthropic, OpenAI and Meta ramp custom chips. For enterprises, the shift changes who controls AI unit costs and where capacity bottlenecks sit.
Launched at VMware Explore 2026, AgentMinder treats each AI agent as an enterprise identity and checks every action against its declared purpose. It marks agent governance moving from policy documents into the infrastructure layer.
Salesforce's AI and data business grew more than 210% while total revenue rose 11%. Half of Agentforce bookings now come from usage credit top-ups, a shift that changes how CIOs should budget and govern agent spending.
Nvidia beat forecasts with $96.2 billion in revenue and guided to $108 billion next quarter. For enterprise buyers, the results point to tight GPU supply, rising component costs and a Rubin transition to plan around.
Salesforce and Anthropic unveiled Claudeforce, a two-way integration that brings 37 sales skills into Claude and makes Claude a default model across Slack and Agentforce. Enterprises gain governed access but face open questions on pricing and lock-in.
Pennsylvania's governor made state permits conditional on developers bringing their own power and winning local approval, days after PJM asked regulators to curtail new data centers first in shortages. Grid access is becoming something AI capacity must earn.
Interim results from The Modern Data Company's survey of more than 540 data leaders show agent adoption racing ahead of data trust, context layers and accountability, the foundations agents need to reach production.
UiPath's new developer canvas lets popular coding agents design, run and govern whole business processes on its orchestration engine. It is a bet that coding agents will build on automation platforms, not replace them.
Dynatrace will pay about $915 million for Arize, folding a leading AI evaluation and agent observability platform into mainstream IT monitoring as enterprises look for proof that production agents are working.
Cisco reported $9.3 billion in hyperscaler AI orders for fiscal 2026 and 25% order growth outside the cloud giants, a sign that enterprise networks are being rebuilt for AI, even as rising costs squeeze margins.
CoreWeave's second-quarter revenue more than doubled to $2.58 billion and its backlog reached about $104 billion, with Caterpillar and Bentley among new customers. Heavy borrowing and losses give enterprise buyers reasons for care.
Palantir's second-quarter revenue rose 93% to about $1.9 billion, US commercial sales grew 149% and full-year guidance went up again, a sign that enterprises are signing large, multi-year AI platform commitments.
AMD's data center revenue more than doubled to $6.7 billion in the second quarter, with Anthropic, Microsoft, Meta and OpenAI among buyers of its AI systems, giving enterprises a stronger second source of AI compute.
Governor Greg Abbott ordered state regulators to audit every data center in ERCOT's interconnection queue before it can advance, pausing new hookups in a key AI market as requests reach 474 gigawatts.
Microsoft beat forecasts with 43% Azure growth and more than 30 million paid Copilot seats, but said AI demand still outruns capacity. Enterprise buyers should plan for tight supply and negotiate Copilot on outcomes.
Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs starting in 2027, with AMD committing up to $5 billion in equity. The deal shows how model providers are chasing compute supply that enterprises depend on.
The European Commission's AI Office and national regulators start enforcing the AI Act on August 2, as transparency rules for chatbots, deepfakes and AI-generated content take effect. High-risk rules were delayed, but much now applies.
ServiceNow's second-quarter results show agentic AI moving into paid production workflows, with AI contract value above $1 billion, ninefold growth in agent deployments and a pricing model that is shifting toward usage.
A multibillion-dollar agreement for European GPU capacity and new deployment options for Mistral models inside Microsoft's tools give regulated enterprises more model choice and more control over where AI runs.
Security firm Zenity says a since-fixed flaw in ChatGPT's Agent Builder let a single phishing click create and publish an agent with an employee's access to email, storage and chat apps, a new kind of insider threat.
ASML raised its full-year sales outlook to €43 billion to €45 billion after a €9.3 billion second quarter, citing AI investment. Memory makers are expanding fastest, a sign that DRAM and HBM shortages will shape AI hardware costs.
Hugging Face disclosed on July 16 that an autonomous AI agent system broke into part of its production infrastructure through a malicious dataset. Public models were not tampered with, but the attack is a warning for enterprise AI pipelines.
Microsoft executives reportedly told salespeople to position the company's in-house AI models against OpenAI, Anthropic and Google. For Copilot customers, the shift raises questions about which model answers their queries and who decides when it changes.
Samsung expects second-quarter operating profit of about 89.4 trillion won, 19 times last year's level, as AI data centers drive up memory prices. Enterprise buyers should expect those costs to show up in server and cloud bills.
Meta opened a paid developer API for Muse Spark and released version 1.1, built for coding and agentic tasks. Low pricing adds a new option for enterprise teams, but the preview raises questions about terms, support and lock-in.
Cognizant expanded its Google Cloud partnership, pledging to certify at least 10,000 professionals on Gemini Enterprise and deploy it to 100,000 employees this year. The move shows services firms retraining their own workforces before selling AI to clients.
Qualcomm plans to buy Modular, whose open software stack runs AI models across many chip types without code rewrites. The deal could widen enterprise hardware choice, but it also puts a portability layer under one chipmaker's control.
South Korea's two memory giants pledged about $518 billion for a new chipmaking hub after AI demand nearly doubled memory prices in a quarter. New capacity is years away, so enterprise AI budgets should plan for tight supply.
The 67th TOP500 list puts a Chinese system at No. 1 for the first time since 2017. Below it, an energy company and a cloud provider hold top-10 spots, and AI-oriented hardware dominates the efficiency rankings.
Micron's fiscal third-quarter revenue more than quadrupled from a year earlier, beat its own guidance by more than $7 billion and came with a $50 billion forecast, a sign that memory is now one of AI's tightest bottlenecks.
OpenAI's new Daybreak program packages a restricted cyber model, an upgraded Codex Security plugin and a partner network that includes major security vendors and consultancies, changing how enterprises may encounter AI vulnerability research.
Google says its newest Gemini features for long-running agents will increasingly ship only on the Interactions API, which reached general availability with server-side state, background execution and managed agent sandboxes.
At its Data + AI Summit, Databricks launched a consumption-priced agent for business teams, a single-copy architecture for transactional and analytical data, and a security acquisition, betting that agents need a simpler data estate.
Accenture's third-quarter revenue grew 6% to $18.72 billion, but new bookings fell and consulting barely grew in local currency. The mix points to clients buying fewer small projects and more large, multi-year AI programs.
The G7 summit in Évian ended with a working lunch on AI deployment, commitments on chatbot safety for minors and model evaluation standards, and no standalone AI declaration, as concern over dependence on a few AI providers grew.