Key Highlights

  • Anthropic just leased SpaceX’s Colossus 1 supercluster — 300+ MW and 220,000+ Nvidia GPUs — from a company whose CEO has spent the last six months publicly calling them “misanthropic.” Per Anthropic’s announcement (and Theo’s deep-dive on the implications), the deal gets resolved by month’s end and is being paired with doubled Claude usage caps across paid tiers, removal of peak-hour throttling, and a 4–5× bump on API rate limits. Dario Amodei told the Code with Claude conference Anthropic saw an annualized 80× revenue/usage growth in Q1 and admitted they undershot compute planning; Theo’s read is sharper — Anthropic has world-class research, OpenAI has all three (research, data, compute), XAI had only compute, and Cursor had only data, which explains both the Anthropic↔SpaceX compute deal and the SpaceX↔Cursor $10B-for-data / $60B-for-the-company option signed weeks earlier. The unifying point worth filing: every recent “puzzling” Anthropic move (trying to remove Claude Code from Pro, peak-hour caps, banning Windsurf and XAI from API access) was a compute-allocation problem, not a pricing-power problem.
  • The five-architect panel at Milken put hard numbers on what “supply-constrained” actually means in 2026. ASML’s Christophe Fouquet: chip supply will limit hyperscalers for “two, three, maybe five years.” Google Cloud COO Francis deSouza: $20B/quarter revenue at 63% YoY growth, with backlog almost doubling from $250B to $460B in a single quarter. Applied Intuition’s Qasar Younis flagged the non-silicon bottleneck: real-world data collection — synthetic simulation can’t fully bridge training gaps for autonomy. The takeaway across the panel was unusually candid: this isn’t a demand-discovery story anymore, it’s an industrial throughput story, and the constraint moves down the stack from chips to data to physical-world capture as you go from LLMs to agents to embodied systems.
  • Google shipped a Prompt API in Chrome that requires accepting Google’s AI terms of service to call a standard web API, and silently downloaded a 4GB Gemini Nano model to users’ machines without consent. Mozilla, WebKit, and the W3C TAG all opposed the proposal; Google shipped it anyway. The author’s argument is narrow but load-bearing for the open web: no W3C-track API should require agreeing to a single advertiser’s prohibited-use policy as a precondition for being callable, and the user-side consent failure (auto-download + auto-reinstall after deletion) compounds the standards problem. Reads as the natural sequel to last week’s Telus accent-conversion story — AI features now ship across long-standing trust boundaries first, and the consent/disclosure debate happens in retrospect.

Analysis & Opinion

Anthropic, SpaceX(AI) become unlikely compute partners — The Rundown

Despite months of public hostility — Musk has repeatedly called Anthropic “misanthropic” on X — SpaceX is leasing its Memphis-based Colossus 1 supercomputer (220,000+ Nvidia GPUs, 300+ MW) to Anthropic to ease an acute serving-capacity crunch. As part of the deal, Claude usage caps double across paid subscription tiers, API users get further increases, and peak-hour rate limits are removed entirely. Musk publicly justified the about-face by saying he’d spent time with senior Anthropic staff and was “impressed,” and that SpaceX’s own training had already moved to Colossus 2 — making Colossus 1 surplus, given how little Grok inference traffic XAI is actually serving. The structural read is that the deal is mutually rational: Anthropic plugs a serving-capacity hole; XAI monetizes a data center it built for an inference business that hasn’t materialized; both companies route around OpenAI, which is currently the only major lab with research, data, and compute under one roof.

Five architects of the AI economy explain where the wheels are coming off — TechCrunch

A panel at the Milken Global Conference brought together ASML, Google Cloud, Perplexity, Applied Intuition, and Logical Intelligence — a full vertical slice of the AI supply chain — and the picture they painted is constraint-bound rather than demand-bound. Fouquet warned that hyperscaler chip supply will be limited for the next 2–5 years even as ASML accelerates manufacturing. DeSouza disclosed a Google Cloud backlog jump from $250B to $460B in a single quarter at $20B/quarter revenue and 63% YoY growth — a backlog roughly 5–6× current annualized revenue. Younis added the dimension the chip-supply story usually elides: for autonomy and physical AI, the binding constraint is real-world data collection, not silicon — synthetic simulation cannot close the last-mile gap. The combined reading is that talking about “AI capacity” as a single number is now misleading; the bottleneck moves down the stack as you move from chatbots to agents to embodied systems.

Google’s Prompt API — via Lobsters

The piece argues — narrowly and persuasively — that a web standard API should not require agreeing to a single company’s prohibited-use policy to be callable, and that Google shipped its Prompt API over the explicit objections of Mozilla, WebKit, and the W3C TAG on the strength of dubious “developer interest” evidence (a sparsely-engaged GitHub thread plus an unspecified survey). The technical critique is sharper than the procedural one: the API is currently nothing more than a thin façade over Google’s Gemini Nano model, and Chrome silently downloads ~4GB of Gemini Nano weights to the user’s machine without consent and reinstalls them after deletion. The author’s framing — “no web standard should require you to agree to an advertising company’s terms of use” — is the load-bearing claim, and it’s hard to refute on its merits: it would set precedent for any future browser-vendor model to ship as a standard with vendor-locked compliance terms.

How we replaced NGINX-Ingress at Stack Overflow — Stack Overflow

Stack Overflow used the November Ingress-NGINX retirement announcement as a forcing function to migrate to Kubernetes Gateway API rather than swap in another Ingress controller. Their evaluation narrowed to three Gateway API candidates (NGINX Gateway Fabric, Traefik, Istio) plus two Ingress fallbacks, eliminating cloud-specific solutions because they run on both GCP and Azure. The methodologically interesting bit: they exported every production Ingress object as YAML and used Claude to bucket them into use-case categories, which produced ~6 functional test scenarios and 2 scalability benchmarks they used to drive the bake-off. Quietly representative of how infra teams are now using LLMs not for code generation but for inventory analysis at scale — turning unstructured config across hundreds of objects into a structured evaluation framework.

New Products & Tools

Anthropic financial-services reference agents — GitHub Trending

Anthropic open-sourced a reference repo of agents, skills, and data connectors aimed at FSI workflows: investment banking pitches, equity research, private equity diligence, fund admin, GL reconciliation, valuation review. Notable for the dual deployment surface — install as Claude Cowork plugins, or deploy via the Managed Agents API (/v1/agents) behind a custom workflow engine — which is the first time Anthropic has shipped vertical-specific agent scaffolding in this format.

Spotify’s AI DJ now supports French, German, Italian, and Brazilian Portuguese — TechCrunch

Each language ships with a distinct named persona (Maia, Ben, Alex, Dani), and the rollout adds eight countries — bringing AI DJ to 75+ markets. The product is now well past commentary-only: text prompts, mood/genre adjustments via chat, and full conversational request handling.

Parloa builds service agents customers want to talk to — OpenAI

OpenAI customer story on Parloa, an enterprise voice-agent platform built on OpenAI’s models for real-time customer service. Standard product spotlight; useful mainly as another data point on the speed at which voice-AI is moving into production CX deployments.

NVIDIA Spectrum-X gets Multipath Reliable Connection (MRC) — NVIDIA

NVIDIA, Microsoft, and OpenAI co-announced MRC, an RDMA transport extension that lets a single RDMA connection spread traffic across multiple paths — essentially multi-pathing for AI training fabrics. OpenAI’s Sachin Katti is quoted saying it lets them avoid “typical network-related slowdowns and interruptions” at frontier training scale. Useful signal: the bottleneck conversation in 2026 is increasingly about fabric and interconnect, not just chip count.

Interviews & Conversations

Anthropic just…wait what — Theo - t3.gg (35:13)

Theo’s response to the Anthropic-SpaceX-Cursor triangle is the most useful single explainer of why everyone is doing what they’re doing right now. His core framework: success in AI requires research, data, and compute — and as of this week, only OpenAI has all three. Anthropic has world-class research and excellent data (from Claude Code runs) but is severely compute-constrained — Dario admitted at Code with Claude that Q1 came in at 80× annualized growth, far past their 10× plan. XAI has compute (Colossus 1, ~442 MW, 280K H100s) but lacks code-behavior data and serious researchers. Cursor has the best agentic-coding-trace dataset on the planet (every back-and-forth across every model their users invoked) but no compute. That explains the SpaceX-Cursor deal ($10B for data only, or $60B for the whole company) and now the SpaceX-Anthropic deal — Anthropic plugs compute, XAI plugs data via Cursor, and both pair up against OpenAI. Theo’s secondary read is also worth the watch: every “weird” recent Anthropic move (peak-hour limits, the attempted Claude Code Pro repricing, banning XAI and Windsurf) was a compute allocation problem misread by the market as a pricing-power problem, and the new doubled caps + removed peak throttling reflect Colossus 1 capacity finally landing. He also speculates — with charts — that Azure’s Claude hosting may currently be a passthrough to Anthropic’s own infrastructure rather than independent inference, based on near-identical latency curves between the two; AWS Bedrock and Google Vertex meaningfully outperform both. Strong cross-reference with The Rundown’s coverage of the same deal.

References

  1. The Rundown, “Anthropic, SpaceX(AI) become unlikely compute partners,” 2026-05-07 [blog]
  2. TechCrunch, “Five architects of the AI economy explain where the wheels are coming off,” 2026-05-06 [blog]
  3. wil.to, “Google’s Prompt API,” Lobsters, 2026-05-06 [blog]
  4. Stack Overflow, “How we replaced NGINX-Ingress at Stack Overflow,” 2026-05-06 [blog]
  5. Anthropic, “financial-services,” GitHub, 2026-05-07 [blog]
  6. TechCrunch, “Spotify’s AI DJ now supports French, German, Italian, and Brazilian Portuguese,” 2026-05-07 [blog]
  7. OpenAI, “Parloa builds service agents customers want to talk to,” 2026-05-07 [blog]
  8. NVIDIA, “NVIDIA Spectrum-X — the Open, AI-Native Ethernet Fabric — Sets the Standard for Gigascale AI, Now With MRC,” 2026-05-06 [blog]
  9. Theo - t3.gg, “Anthropic just…wait what,” YouTube, 2026-05-07 [video]