Claudeforce, AWS's 2 Million GPU Bet, and the Earnings Week That Buried the SaaSpocalypse

Salesforce and Anthropic launch Claudeforce as Marc Benioff and Dario Amodei explain the collaboration together on CNBC, AWS commits to 2 million more NVIDIA GPUs on top of its GTC pledge, plus six new earnings this week from NVIDIA, Salesforce, Synopsys, HP, Everpure, and Marvell test every bear thesis on AI infrastructure and software at once. Patrick Moorhead and Daniel Newman also cover Hot Chips 2026, the NVIDIA-Hugging Face acquisition rumor, and debate whether Anthropic's SaaS reassurances hold up on Ep. 317 of The Six Five Pod.

The handpicked topics for this week are:

Claudeforce Turns Salesforce Into Anthropic's Enterprise Front End. Salesforce and Anthropic launched Claudeforce, positioning Claude as the interface across roughly 27 Salesforce services while Anthropic supplies the underlying AI engine. Marc Benioff and Dario Amodei appeared together on CNBC to make the case for the partnership, and both stocks rallied on the news. Moorhead flags one open question: how Anthropic protects Salesforce customer data without collecting the usage traces that AI systems typically retain. (The Decode)

AWS Adds 2 Million NVIDIA GPUs on Top of Its GTC Commitment. AWS committed to another 2 million NVIDIA GPUs, layered onto the 1 million it pledged at GTC five months earlier, alongside its own Trainium and Graviton silicon build-out. Moorhead estimates the deal at 6 to 7 gigawatts and $80 billion to $120 billion in NVIDIA revenue, and reads the pairing of NVIDIA's Vera CPU with Graviton as evidence that agentic workloads need capabilities Amazon's own silicon doesn't yet cover. The commitment reinforces Moorhead's argument that wafer, packaging, and memory supply set the ceiling on AI infrastructure buildout, regardless of how many custom silicon projects come online. (The Decode)

Hot Chips 2026 Draws Mainstream Attention as Custom and Merchant Silicon Both Scale. What was once an academic gathering turned into a press and social media event, with OpenAI's Jalapeño inference chip drawing the most attention for its bandwidth-heavy first-generation performance. IBM and Arm detailed a joint development agreement enabling IBM Z mainframes to run Arm code natively at sub-nanosecond switching latency, and AMD showed a full Helios rack while Arm walked through its AGI chip architecture. Moorhead notes that Jalapeño almost certainly relied on Synopsys or Cadence EDA tools rather than in-house design tooling. (The Decode)

The NVIDIA-Hugging Face Acquisition Rumor