Need to know
- Salesforce and Anthropic launched Claudeforce, embedding Claude across CRM, Slack, and 37 prebuilt sales skills via a plugin.
- Nvidia reported $96.2B in Q2 revenue, up 106% year-over-year, as AWS added 2 million more Nvidia GPUs.
- Nvidia agreed to acquire Hugging Face for $12.9 billion, per The Information.
- Okta posted record non-Q4 bookings and raised full-year guidance, with new AI-driven products now 30% of bookings.
- CrowdStrike reported Q2 revenue up 26% and record net new ARR of $333M, with Falcon Flex ARR doubling year-over-year.
New Releases
Salesforce and Anthropic announced Claudeforce, a formal expansion that puts a 37-skill Salesforce plugin inside Claude and embeds Claude as the agent layer across Salesforce products.
- Salesforce in Claude ships today as a plugin giving sellers live pipeline context, deal health review, and approved CRM write-actions without leaving Claude's interface.
- Salesforce expects to spend roughly $300M on Anthropic tokens this year and holds an equity stake now valued at approximately $5B, making this a financial relationship as much as a product one.
HubSpot launched a native Claude connector that surfaces pipeline, contact, campaign, and support data inside Claude conversations with no app switching.
- The connector supports read and write actions, letting users query churn signals and ticket urgency from within Claude's chat interface using live HubSpot data.
- Timing is notable: HubSpot is shipping its Claude integration the same day Salesforce announced Claudeforce, accelerating a pattern where CRM vendors embed AI assistants rather than waiting to be embedded.
Google DeepMind released Gemini 3.5 Transcribe, a speech-to-text model that drops filler words, self-corrects mid-sentence, and reaches final transcripts 70% faster than Chirp 3 across 85-plus languages.
- Word error rate in streaming mode is 4.0%, with the model performing comprehension-driven transcription rather than verbatim capture, meaning it writes what the speaker meant rather than every stumble.
- Two variants ship today: gemini-3.5-transcribe for batch use and gemini-3.5-transcribe-live for real-time applications, both available via the Interactions API.
IBM released Granite 4.2 in 3B, 8B, and 30B parameter sizes under Apache 2.0, each trained with a reinforcement learning pipeline designed to act inside tools rather than just answer questions.
- All three sizes are built on roughly 15 trillion tokens and are fully open-weight, making them deployable on-premises without licensing restrictions relevant for regulated enterprise environments.
- IBM's explicit bet is that agentic skill, not raw benchmark scores, is the differentiator for small models in enterprise workflows where latency and cost per call matter more than frontier-model capability.
Funding
Nvidia agreed to buy Hugging Face for $12.9 billion, per The Information, a move that would give the chip leader direct control over the dominant open-source model hub and the developer community that depends on it.
Anthropic reportedly signed a multi-year, roughly $45B compute agreement with Nscale tied to next-generation Nvidia chips, signaling that frontier-model training costs are scaling faster than any single hyperscaler can absorb.
Keenable raised a $26M seed round led by Accel to build a search index purpose-built for AI agents, founded by former Yandex search executive Andrey Styskin, with undisclosed AI lab customers already in production.
Case Studies
Okta reported Q2 FY27 revenue of $805M, up 11% year-over-year, with record non-Q4 bookings and new AI-driven products now representing 30% of total bookings.
- Large enterprise traction accelerated, with customers spending over $1M annually growing more than 20%, and the company raised full-year guidance for the second consecutive quarter.
- Agent SSO and workforce identity for AI agents were cited on the earnings call as the primary demand drivers, confirming that non-human identity is a measurable revenue line, not a roadmap item.
CrowdStrike posted Q2 FY27 revenue of $1.47B, up 26% year-over-year, with net new ARR surging 51% to a record $333M and Falcon Flex ARR doubling to $2.29B.
- Falcon Flex ARR crossed $2.29B, up 101% year-over-year, demonstrating that the flexible consumption model is pulling customers toward platform consolidation faster than per-module sales.
- Net new ARR of $333M is the highest in company history for any quarter, reversing several quarters of ARR deceleration that followed the 2024 outage and validating the recovery trajectory.
Trending on X
- Nvidia earnings vs. OpenAI Jalapeño chip Practitioners are debating whether OpenAI's Hot Chips benchmark win over Nvidia Blackwell matters at all given Nvidia just reported 106% revenue growth, with the emerging consensus being that benchmarks flatter custom silicon while Nvidia owns the production ecosystem.
- Salesforce Claudeforce and SaaSpocalypse Enterprise observers are arguing that Claudeforce disproves the SaaSpocalypse thesis, since Salesforce's stock jumping 14% on AI-driven CRM usage growth suggests vertical AI and systems of record are reinforcing each other rather than competing.
- Altman AGI claim for end of 2026 Sam Altman's statement to TIME that OpenAI will achieve AGI by year-end is drawing skepticism from practitioners who argue the term has been redefined to the point of meaninglessness as a marketing instrument.
- OpenAI vs. Anthropic enterprise seat gap A circulating data point showing Anthropic at 44% vs. OpenAI at 40% enterprise share across 70,000-plus US businesses is generating debate about whether a closing gap favors the incumbent or signals Anthropic's ceiling.
- Zhipu Ox Alpha identity revealed Confirmation that the anonymously released Ox Alpha model is Zhipu AI's GLM-5.3-Flash, deployed on 100,000 domestic Chinese chips and consuming tens of trillions of tokens in six days, is being read as evidence that Chinese open-weight AI is advancing faster than Western observers track.