Groq does not serve Claude, so this 12-tool "AI app in 2026" card is a shopping list, not a stack

A 10-second clip at 720x1280, posted 2026-10-03 by @brochbuilds ("Ben Broch | AI Saas builder"), sitting at 497 views, 9 likes, 1 comment, 0 reposts and 5 saves at capture, over an audio track credited only as "original sound" by the same account. I analyzed all 5 extracted frames and a 27-word transcript. The transcript carries nothing: Whisper picked up sung lyrics from the background music ("No, she doesn't mind / I, she doesn't mind"), not narration. All five frames are visually identical. Each shows the same static white-card overlay burned over b-roll of a man in a palm-print shirt DJing on a Pioneer CDJ setup at a waterfront party at sunset, crowd with hands up on the right. The overlay is a headline, "How to build an AI app in 2026," underlined in yellow, above a two-column, six-row grid of twelve labeled tool cards, each with the vendor's logo: Coding: Claude Code. Model: Claude. Any model: OpenRouter. Fast inference: Groq. AI in the UI: Vercel AI SDK. Database: Supabase. Vectors: Pinecone. GPUs: Modal. Voice: ElevenLabs. Auth: Clerk. Payments: Stripe. Analytics: PostHog. No prices, no code, no URLs, no repo, no on-screen caption beyond those twelve labels.

The card is the entire video

There is no progression across the ten seconds. Frames 1 through 5 carry the same twelve cards in the same positions. The only thing that changes is the b-roll behind them, which is unrelated to the topic. The spoken track is music. So the full informational payload of this post is twelve product names and twelve category labels, which a viewer absorbs in about two seconds. The remaining eight seconds are the DJ footage.

That is worth saying plainly: this is a hook with a thin payload. It tells you what to search for. It does not tell you how any of these pieces connect, what they cost, or why one was picked over an alternative. Every name on the card is a real, currently shipping product, so the list is accurate as far as it goes. It just does not go far.

Groq does not serve Claude, so two of the boxes cannot describe the same call

The card lists "Model: Claude" and "Fast inference: Groq" as adjacent slots in one stack. Those cannot both apply to the same request. Groq's published model catalog is open-weight only. Groq's own model docs list Llama 3.1 8B at 560 tokens/second, Llama 3.3 70B at 280 tokens/second, GPT-OSS 120B at roughly 500 tokens/second and GPT-OSS 20B at roughly 1,000 tokens/second, plus Whisper variants for speech-to-text. No Anthropic model appears. Groq describes itself as "the premier neocloud for fast inference" running proprietary LPU hardware, and it does not host Claude.

The route that does cover both is the next box over. OpenRouter states that "Every Anthropic model below is served through the same OpenRouter API and API key," and its Anthropic page lists Claude Opus, Sonnet, Haiku and Fable variants through Claude 5.5. So "Any model: OpenRouter" and "Model: Claude" compose. "Fast inference: Groq" is a separate decision about running open-weight models cheaply and quickly, not a speed knob you turn on Claude.

For the record on the model itself: Anthropic's models overview lists Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5 as the current lineup. Opus 5.5 runs $4 per million input tokens and $20 per million output, with a 1M-token context window. The card says only "Claude," which spans a 10x price range across that lineup.

Supabase already ships two of the other boxes

The grid treats Database, Auth and Vectors as three separate purchases: Supabase, Clerk and Pinecone. Supabase's own site describes six products in one platform, including Auth ("user management with 20+ social providers") and Vector ("AI toolkit powered by pgvector for storing, indexing, and querying vector embeddings alongside transactional data in Postgres"). If you take the Supabase box, you have already bought auth and vector search.

That does not make Clerk or Pinecone wrong picks. Clerk sells itself as "more than authentication" with full user and org management and a free tier covering the first 50,000 monthly retained users. Pinecone now positions itself above the vector-database category as "the trusted AI knowledge company," citing billions of vectors and thousands of queries per second, which is a different scale problem than pgvector in a single Postgres instance. The point is that the card presents twelve independent slots when at least two of them are overlapping decisions, and nothing on screen flags the overlap.

Naming nitpick, and the rest checks out

One name on the card is out of date. The overlay reads "Vercel AI SDK." The product's own documentation now calls it the "AI SDK," branded "AI SDK by Vercel," and describes it as "the TypeScript toolkit for building AI applications and agents with React, Next.js, Vue, Svelte, Node.js, and more." The "Vercel AI SDK" form is the legacy name.

Everything else on the card is spelled correctly and does what the label implies. Claude Code is "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools," available in terminal, IDE, desktop app and browser. Modal sells serverless GPUs priced per second, from an Nvidia T4 at $0.000164/sec to a B300 at $0.001972/sec, with $30/month free credits on the Starter plan. ElevenLabs covers text-to-speech, speech-to-text and voice agents across 70+ languages. Stripe charges 2.9% + 30 cents per successful US online card transaction. PostHog is MIT-licensed with 40.1k GitHub stars and still ships product analytics alongside session replay, flags and experiments, even though its homepage now leads with "your product's context layer."

Key Takeaways

The clip is 10 seconds, 720x1280, and all 5 extracted frames are visually identical. The 27-word transcript is background song lyrics, not narration, so 100% of the information is on-screen text.
Correction: the card pairs "Model: Claude" with "Fast inference: Groq," but Groq's published catalog is open-weight only (Llama 3.1 8B at 560 tps, Llama 3.3 70B at 280 tps, GPT-OSS 120B at ~500 tps, GPT-OSS 20B at ~1,000 tps). It does not serve Anthropic models.
Correction: the overlay says "Vercel AI SDK." The current official name on ai-sdk.dev is "AI SDK," branded "AI SDK by Vercel."
Supabase already ships Auth and Vector (pgvector) per its own site, so Clerk and Pinecone are overlapping purchases, not three independent slots. The card does not say this.
"Claude" on the card spans a 10x price range: Anthropic lists Haiku 4.5 at $1/$5 per million tokens and Fable 5.1 at $10/$50, with Opus 5.5 at $4/$20 and a 1M-token context window.
Hard numbers the card omits entirely: Stripe is 2.9% + 30 cents per US online card charge, Modal runs $0.000164/sec on a T4 up to $0.001972/sec on a B300, Clerk is free to 50,000 monthly retained users.
Engagement at capture was 497 views, 9 likes, 1 comment, 0 reposts and 5 saves. The save-to-like ratio (5 saves on 9 likes) is the only signal that viewers treated it as a reference list.
Unverified: whether @brochbuilds ships anything on this stack. The clip names no project, repo, URL or price, and the account bio string "Ben Broch | AI Saas builder" is the only claim of credentials. Also unverified: why Groq and Modal both appear, since the clip gives no workload that needs both.

Resources

AI SDK documentation — confirms the official name is "AI SDK" by Vercel, not "Vercel AI SDK," and that it is a TypeScript toolkit for React, Next.js, Vue, Svelte and Node.js.
Groq models documentation — confirms Groq's catalog is open-weight (Llama, GPT-OSS, Qwen, MiniMax, Whisper) with published tokens-per-second figures, and contains no Anthropic models.
Groq homepage — confirms the "premier neocloud for fast inference" positioning and LPU hardware.
OpenRouter Anthropic models — confirms Claude models through Opus 5.5 and Sonnet 5.5 are served on one OpenRouter API key.
Anthropic models overview — confirms the current lineup (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 4.5), per-million-token pricing and 1M-token context windows.
Claude Code overview — confirms Claude Code is Anthropic's agentic coding tool across terminal, IDE, desktop and web.
Supabase homepage — confirms Supabase ships Auth and Vector (pgvector) alongside Postgres, Storage, Edge Functions and Realtime.
Clerk homepage — confirms full-stack auth and user management, and the free tier at 50,000 monthly retained users.
Pinecone homepage — confirms the repositioning to "the trusted AI knowledge company" and the billions-of-vectors scale claim.
Modal pricing — confirms per-second serverless GPU pricing from $0.000164/sec (T4) to $0.001972/sec (B300) and $30/month free credits.
ElevenLabs homepage — confirms text-to-speech, speech-to-text and voice agents across 70+ languages.
Stripe pricing — confirms 2.9% + 30 cents per successful US online card transaction.
PostHog on GitHub — confirms MIT license, 40.1k stars and the analytics, session replay, flags and experiments product set.