Top Stories
Anthropic Launches Claude Opus 4.6 with 1M Token Context Window
Anthropic released Claude Opus 4.6 on Wednesday, its most powerful model yet. The headline feature is a 1 million token context window in beta - enough to process around 1,500 pages of text, 30,000 lines of code, or over an hour of video in a single prompt.
Beyond raw context length, Opus 4.6 improves on its predecessor across the board. On GDPval-AA, an evaluation of economically valuable knowledge work in finance, legal, and other domains, it outperforms OpenAI’s GPT-5.2 by roughly 144 Elo points and its own predecessor Opus 4.5 by 190 points. It also tops Terminal-Bench 2.0 for agentic coding and Humanity’s Last Exam for reasoning.
New capabilities include Agent Teams in Claude Code (multiple agents working in parallel on different parts of a task), Adaptive Thinking (the model decides when deeper reasoning is needed), and a Claude in PowerPoint research preview. Pricing stays at $5/$25 per million tokens, with a premium tier at $10/$37.50 for prompts exceeding 200K tokens.
The timing matters. Coming on the heels of Claude Cowork’s market-shaking plugins, this release deepens the concern that Anthropic is building the toolkit to replace entire software categories, not just augment them.
Alphabet Doubles Down on AI with Up to $185 Billion in 2026 Capex
Alphabet told investors its 2026 capital expenditures will land between $175 billion and $185 billion - roughly double the $91.4 billion it spent in 2025, and more than triple the $52.5 billion spent in 2024. The spending will go toward AI compute capacity for Google DeepMind and meeting what the company describes as “significant cloud customer demand.”
The numbers make Alphabet the biggest spender in the AI infrastructure arms race, surpassing Meta’s projected $115-$135 billion for 2026. Google Cloud’s backlog surged 55% sequentially and more than doubled year over year, reaching $240 billion. CEO Sundar Pichai said the company remains “supply constrained” and needs even more capacity.
Investors weren’t entirely convinced the bet would pay off. Alphabet stock fell as much as 5% on Thursday as the scale of spending raised questions about returns. But the earnings themselves were strong: annual revenue crossed $400 billion for the first time, a 15% year-over-year increase.
Sources: Yahoo Finance, Fortune, CNBC
The SaaSpocalypse: Nearly $1 Trillion Wiped from Software Stocks
The fallout from Anthropic’s Claude Cowork plugins continued to cascade through markets this week. What started with a $285 billion single-day wipeout when the legal and financial plugins launched has now grown to nearly $1 trillion in lost market value across software and services stocks.
Thomson Reuters fell more than 15%. RELX, parent of LexisNexis, dropped about 14%. LegalZoom cratered nearly 20%. DocuSign, Salesforce, Adobe, and ServiceNow all took 7-11% hits. Traders are calling it the “SaaSpocalypse” - the dawning realization that AI agents aren’t just supplementing enterprise software, they might be replacing it.
The fear isn’t irrational. Anthropic released its Cowork plugins under an open-source license, meaning anyone can customize and extend them. When a $5/month AI subscription can do contract review, NDA triage, and compliance checks, the value proposition of specialized SaaS tools that charge hundreds per seat gets harder to defend.
Sources: Yahoo Finance, Bloomberg
Hardware & Infrastructure
Nvidia Vera Rubin Platform Enters Full Production
Nvidia announced that its next-generation Vera Rubin platform has entered full production, with partner systems expected in the second half of 2026. The Rubin platform - successor to Blackwell - promises a 10x reduction in inference token costs and a 4x reduction in GPUs needed to train mixture-of-experts models.
The platform combines one Vera CPU with two Rubin GPUs in a single superchip, part of a six-chip architecture. AWS, Google Cloud, Microsoft Azure, and CoreWeave will be among the first to deploy Vera Rubin, with server makers Cisco, Dell, HPE, Lenovo, and Supermicro building systems around it.
This marks Nvidia’s shift from a two-year release cadence to annual updates - a pace driven by the relentless demand for AI compute that Alphabet’s spending numbers underscore.
Source: Nvidia Newsroom
Defense & Policy
Pentagon Integrates Grok into Military AI Platform, Hits 1.1 Million Users
The Department of War confirmed that xAI’s Grok models will be added to GenAI.mil, the Pentagon’s enterprise AI platform already used by 1.1 million military and civilian personnel. Five of six military branches have formally adopted GenAI.mil as their go-to AI tool.
The integration will give 3 million Department of War employees access to Grok at Impact Level 5 (sensitive but unclassified). This comes alongside the Pentagon’s broader “AI-first” strategy, which mandates that AI vendors deploy their latest models within 30 days of public release and outlines seven “Pace-Setting Projects” for accelerated AI adoption.
The speed of military AI adoption is notable - and the lack of public debate about it even more so. Giving Elon Musk’s AI models access to military systems while Musk simultaneously advises on government technology policy raises conflict-of-interest questions that haven’t been adequately addressed.
Sources: Department of War, DefenseScoop
Security
Okta Flags Authorization Gap as AI Agents Enter Shared Workspaces
Okta published research warning about a fundamental security flaw in how AI agents handle permissions in collaborative tools. The problem: an agent retrieves data using the permissions of whoever authenticated it (say, a CFO), but then outputs that data into shared spaces like Slack channels where recipients have mixed clearance levels.
Four critical vulnerabilities rated CVSS 9.3-9.4 were found in agent integrations from Anthropic, Microsoft, ServiceNow, and Salesforce in 2025. Okta argues that traditional OAuth-based access control wasn’t built for this scenario and that a new layer of “fine-grained authorization” is needed - one that computes the intersection of all recipients’ permissions before data leaves the retrieval layer.
As AI agents proliferate in enterprise environments, this class of vulnerability is likely to become a major attack surface.
Source: Okta
Quick Hits
-
Super Bowl preview: Anthropic’s pre-game and in-game ads mock AI assistants that pitch products mid-conversation - a pointed jab at OpenAI’s decision to introduce ads in ChatGPT’s free tier. Sam Altman called the campaign “funny” but “clearly dishonest.” The Super Bowl itself airs Sunday. CNBC
-
Nvidia Nemotron 3: Nvidia’s open-weight model family includes Nano (30B params, available now), Super (100B, H1 2026), and Ultra (500B, H1 2026). All use a hybrid latent mixture-of-experts architecture optimized for agentic AI. Nano delivers 4x the throughput of its predecessor. Nvidia
-
AI2 Theorizer: The Allen Institute for AI released Theorizer, an open-source tool that synthesizes scientific theories by reading up to 100 papers per query - flipping the typical researcher workflow of reading first, theorizing later.
-
Samsung + Gemini: Samsung aims to double its Gemini AI-equipped mobile devices to 800 million units by end of 2026, bringing generative AI features to mid-tier and budget phones.
Worth Watching
The SaaSpocalypse may be the most significant market event in AI since DeepSeek’s emergence. The question isn’t whether AI will disrupt enterprise software - it’s how fast. Anthropic open-sourcing its Cowork plugins was the accelerant. Now every company with an LLM can build competing agents for legal, financial, and operational workflows. The SaaS incumbents that survive will be those that integrate AI deeply enough to stay ahead of commoditization.
Alphabet’s $185 billion spending commitment deserves scrutiny too. When the CEO of a $2 trillion company says they’re “supply constrained” despite planning to spend more on infrastructure than most countries’ GDP, it raises real questions about whether we’re in a rational investment cycle or something closer to a bubble.