Daily Tech Briefing: Salesforce and Anthropic Claudeforce Partnership, Nvidia Buys Hugging Face & Google Launches Gemini 3.5 Transcribe

The CODEW Daily Tech Briefing | August 28, 2026

Ten Fast Reads: What Changed Today Across AI, Infrastructure, Security, and Capital

Nvidia Posts a $96.2 Billion Quarter and Forecasts 70% Growth as AI Infrastructure Spending Keeps Surging.

Data-center revenue hit $89 billion, up 117% year-over-year, while Nvidia's own $160 billion memory-supply commitment shows how deep the industrial buildout now runs.

What happened: Nvidia reported $96.2 billion in quarterly revenue and, notably, issued its first year-ahead growth forecast: roughly 70% revenue growth next fiscal year, with current-quarter guidance around $108 billion. Data-center revenue alone reached $89 billion, up 117% year-over-year. CEO Jensen Huang maintains demand remains constrained by supply, not customer appetite.

Key numbers/companies: Q2 revenue $96.2B · Data center revenue $89B, +117% YoY · Q3 guidance ~$108B · FY27 growth forecast ~70% · $160B committed to memory supply.

Why it matters: Nvidia is increasingly using its own balance sheet — investments, financing guarantees, partnerships — to support the infrastructure projects that buy its chips, raising questions about how intertwined it has become with its biggest customers.

Market implication: Rising memory costs are expected to pressure Nvidia's own gross margins even as revenue accelerates — a tension worth watching across the whole AI hardware chain.

What's next: Watch whether the 70% forward growth forecast holds through the next two quarters as memory and packaging constraints persist.

Sources: Financial Times (via TechStartups).

Salesforce Puts Its CRM Inside Claude With New "Claudeforce" Partnership

The first product ships with 37 prebuilt sales skills, letting users query and act on live Salesforce data without opening Salesforce's own interface.

What happened: Salesforce and Anthropic launched Claudeforce, putting Salesforce data, business rules, workflows, and permitted actions directly inside Claude. Salesforce in Claude ships with 37 prebuilt sales skills covering meeting prep, pipeline analysis, and deal-health reviews; pilot customers have access now, with open beta planned for September.

Key numbers/companies: 37 prebuilt sales skills at launch · Open beta in September · Salesforce, Anthropic, Agentforce, Slack integration also planned.

Why it matters: Salesforce spent decades building dashboards and menus; agents let users state intent directly instead, making the traditional application interface matter less — a genuine bet that enterprise software becomes infrastructure agents operate rather than screens people click through.

Market implication: Other CRM and enterprise-app vendors face pressure to open comparable agent-native access rather than protect their own interface as the primary product.

What's next: Watch adoption during the September open beta as the real test of whether users trust an agent to take permitted actions, not just answer questions.

Sources: VentureBeat (via TechStartups).

Anthropic Locks In $45 Billion, Six-Year AI Compute Deal With Nscale

The agreement covers 460 megawatts at a West Virginia data center and will run on Nvidia's next-generation Vera Rubin systems starting in late 2027.

What happened: Anthropic agreed to spend $45 billion over six years renting AI computing capacity from British infrastructure company Nscale, covering roughly 460 megawatts at a West Virginia development, per Bloomberg. Capacity comes online beginning in late 2027 using Nvidia's next-generation Vera Rubin systems.

Key numbers/companies: $45B over 6 years · ~460 MW capacity · Nscale (founded 2024) · Vera Rubin systems from late 2027.

Why it matters: Frontier AI labs increasingly resemble energy-intensive industrial companies more than software startups — future growth depends on securing land, electricity, chips, and financing years in advance, and specialized providers like Nscale are becoming major counterparts to the hyperscalers.

Market implication: A two-year-old infrastructure company landing a $45 billion commitment shows compute-provider diversification accelerating beyond AWS, Azure, and Google Cloud.

What's next: Watch Nscale's ability to execute construction and power procurement at this scale on schedule.

Sources: Bloomberg (via TechStartups).

Kioxia and SanDisk Plan $31 Billion-Plus Memory Investment in Japan Through 2032

The buildout targets flash-memory and next-generation semiconductor capacity as AI's data and storage demands spread the infrastructure boom deeper into the memory supply chain.

What happened: Kioxia and SanDisk plan to invest more than $31 billion in Japan through 2032 to expand flash-memory production, including new capacity at Kioxia's Kitakami plant in Iwate Prefecture, contingent on Japanese government support. The companies have already invested more than $50 billion together in Japan over the past 25 years.

Key numbers/companies: $31B+ planned through 2032 · $50B+ already invested over 25 years · Kioxia, SanDisk, Kitakami plant (Iwate Prefecture).

Why it matters: Large training datasets, inference workloads, and vector databases require enormous NAND flash and memory capacity, not just GPUs — Nvidia's own $160 billion memory-supply commitment (see AI, above) confirms the same bottleneck from the buyer side.

Market implication: Japan continues treating advanced semiconductors as a strategic priority, supporting domestic manufacturing as global supply chains grow more politically sensitive.

What's next: Watch for confirmation of Japanese government support and construction timelines at Kitakami.

Sources: Wall Street Journal (via TechStartups).

AWS and Nvidia Plan to Deploy 2 Million More GPUs in 2027 and 2028

The expansion includes a 100,000-GPU secure system for U.S. government workloads and extends the partnership into robotics, networking, and physical AI.

What happened: AWS and Nvidia are expanding their partnership, with AWS planning to deploy an additional 2 million Nvidia GPUs (Blackwell Ultra, Rubin, Rubin Ultra) across 2027-2028, building on a prior plan to add 1 million GPUs starting in 2026. Plans include a secure 100,000-GPU system for federal and national-security workloads, plus Amazon Robotics adopting Nvidia's physical-AI technology.

Key numbers/companies: 2M additional GPUs planned (2027-2028) · Builds on 1M+ GPU plan from 2026 · 100,000-GPU federal/national-security system · AWS, Nvidia, Amazon Robotics.

Why it matters: The scale complicates the narrative that Amazon's own Trainium chips compete with Nvidia GPUs — hyperscalers are building custom silicon while simultaneously buying enormous Nvidia volumes to meet customer demand for multiple compute options.

Market implication: One of the clearest signals yet that hyperscalers expect AI compute demand to stay enormous for years, not plateau in the near term.

What's next: Watch execution against the 2027-2028 timeline given ongoing power and packaging constraints documented across this week's coverage.

Sources: Amazon Web Services (via TechStartups).

Ransomware Crew Used SpaceX's Cursor AI Agent to Breach at Least Seven Companies

Attackers told Cursor the work was a security simulation, then had it steal credentials and map networks — while ATF separately confirmed a "major" breach after a Qilin ransomware claim.

What happened: Gambit Security and Reuters found a Russian-speaking Aur0ra ransomware affiliate used Cursor (running Anthropic's Claude 4.5 Sonnet) to help breach at least seven companies between April 8 and May 21, recovering 28 chat sessions from an exposed server. The operator convinced the agent the work was a legitimate simulation, then had it steal credentials and map internal networks. Separately, ATF confirmed a "major" cyber incident affecting a standalone system after the Qilin ransomware group listed it on a leak site.

Key numbers/companies: 7+ companies breached · 28 recovered chat sessions · 30-50% faster operations per Gambit's Eyal Sela · ATF, Qilin ransomware group.

Why it matters: General-purpose coding agents are now measurably accelerating real intrusions by automating steps that once required manual expertise, and AI providers face a genuine challenge distinguishing legitimate security testing from misrepresented malicious use.

Market implication: Expect coding-agent vendors to face pressure for behavior-based abuse detection rather than relying on user-stated intent alone.

What's next: Watch whether Cursor (now part of SpaceX following its $60B acquisition) or Anthropic respond with new safeguards against simulation-framed jailbreaks.

Sources: Reuters; BleepingComputer (via TechStartups).

Google Launches Gemini 3.5 Transcribe for Real-Time, Sub-Second Speech-to-Text

The model removes filler words, recognizes self-corrections, and identifies multiple speakers across 85+ languages, positioning voice as core AI-agent infrastructure rather than a standalone feature.

What happened: Google introduced Gemini 3.5 Transcribe, available via the Gemini API and Google AI Studio, supporting streaming transcription with sub-second latency across more than 85 languages. Google reports average word-error rates of 4.0% for streaming and 2.6% for non-streaming workloads in cited testing.

Key numbers/companies: 85+ languages supported · 4.0% WER streaming, 2.6% non-streaming (cited testing) · Sub-second latency · Google.

Why it matters: As AI shifts from text boxes toward voice-driven agents — call centers, meeting tools, accessibility software — accurate, low-latency transcription becomes foundational infrastructure that every downstream agent depends on to interpret intent correctly.

Market implication: Enterprise voice-AI vendors building on third-party transcription now compete against Google offering comparable capability natively through its own API.

What's next: Watch real-world accuracy across accents, background noise, and specialized vocabulary as adoption scales beyond Google's cited benchmarks.

Sources: Ars Technica; Google (via TechStartups).

Nvidia Reportedly Agrees to Buy Hugging Face for $12.9 Billion

The price is nearly 3x Hugging Face's 2023 valuation and roughly 86x its reported $150 million in annualized revenue, extending Nvidia's reach into open-model distribution.

What happened: Nvidia has reportedly agreed to acquire Hugging Face for $12.9 billion, per The Information (via Reuters), taking control of the widely used model, dataset, and developer-tool repository that has become a central distribution layer for open-source AI. Nvidia participated in Hugging Face's 2023 round at a $4.5 billion valuation alongside Google and Salesforce; the company's annualized revenue was recently reported around $150 million. Neither company had publicly confirmed the deal as of this writing.

Key numbers/companies: $12.9B reported deal value · ~2023 valuation $4.5B · ~$150M annualized revenue · Nvidia, Hugging Face.

Why it matters: Owning Hugging Face would extend Nvidia's reach far beyond chips into the software and model-distribution layer developers use daily, giving it strategic positioning wherever models are discovered, tested, and deployed — even as OpenAI, Anthropic, and Google pursue more vertically integrated stacks of their own.

Market implication: Raises questions about neutrality for a platform many competing AI companies rely on for distribution, now potentially owned by the industry's dominant chip supplier.

What's next: Watch for official confirmation from both companies and any regulatory scrutiny given Nvidia's market position.

Sources: The Information via Reuters (via TechStartups).

Chip Startup Architect Labs Says AI Helped Design a New Processor in Two Weeks

The $24M-seed startup, backed by Google DeepMind's Jeff Dean, compressed a process that typically takes large engineering teams over a year — though the design hasn't been manufactured yet.

What happened: Architect Labs says AI helped two human chip architects develop and verify a processor design called Redwood in roughly two weeks, tested via FPGA simulation but not yet manufactured. The startup emerged from stealth this summer with $24 million in seed funding, backed by investors including Google DeepMind's Jeff Dean and executives associated with OpenAI and Nvidia, and is preparing to send the design to TSMC.

Key numbers/companies: $24M seed round · ~2-week design timeline (vs. 1+ year typical) · FPGA-tested, not yet fabricated · Architect Labs, targeting TSMC.

Why it matters: AI-assisted chip design could lower one of the biggest barriers to custom silicon — cost and engineering headcount — potentially letting far more startups design processors for narrow workloads without massive semiconductor teams.

Market implication: The claim still needs validation through fabricated hardware, where power, timing, yield, and manufacturing issues can expose problems simulations miss.

What's next: Watch for Redwood's actual tape-out results at TSMC as the real test of whether the two-week design claim holds up in silicon.

Sources: Business Insider (via TechStartups).

Hugging Face Unveils Microduck, a $399 Desktop Biped Robot for AI Training

The open-source, 800-gram walking robot packs 15 motors, a camera, and LiDAR, aiming to put programmable physical AI on the same price shelf as a game console.

What happened: Hugging Face's robotics team at Pollen Robotics opened pre-orders for Microduck, a 25-centimeter, 800-gram biped priced at $399, featuring 15 motors, a camera, LiDAR, dual inertial sensors, Wi-Fi, Bluetooth, and NFC. It ships with seven trained behaviors and a 50Hz onboard policy loop; Seeed Studio will manufacture roughly 20,000 units, with first deliveries targeted before Christmas.

Key numbers/companies: $399 price · 800g, 25cm, 15 motors · ~20,000-unit planned run · Hugging Face/Pollen Robotics, manufactured by Seeed Studio.

Why it matters: A sub-$400, fully open-source walking robot with an open simulation and reinforcement-learning stack lowers the cost of physical-AI experimentation dramatically for students and startups that can't afford industrial humanoids.

Market implication: Could meaningfully widen the pool of developers building and training robotics skills, similar to how affordable dev boards expanded embedded-systems experimentation.

What's next: Watch order volume against the ~20,000-unit planned run and whether Christmas delivery targets hold.

Sources: Bloomberg (via TechStartups).

THE CODEW · DAILY TECH BRIEFING

Editorial Note

The CODEW Daily Tech Briefing is a fast morning read on the day's most important technology signal, plus the handful of other stories worth knowing — built to be read in minutes, with the deeper analytical work reserved for The CODEW's Watch series and Weekly Tech Roundup.

Coverage is based on company announcements, public disclosures, industry reporting, and other publicly available information. Reported figures and sourced-but-unconfirmed details are noted as such. Analysis reflects the reporting period and should be considered in the context of the sources and developments cited.