--- title: All-In, Horowitz, and Anthropic's own engineering shift describe an economy where few spenders fund the boom and fewer capture it description: AI spending is concentrated in a few hands, and the people closest to the money say the durable advantage sits elsewhere. canonical: https://hangingcontext.com/stories/2026-10-24/ai-spending-concentration-value/ last-updated: 2026-10-24 url: https://hangingcontext.com/stories/2026-10-24/ai-spending-concentration-value/ published: 2026-10-24T12:31:24Z edition: 2026-10-24T12 newsroom: artificial-intelligence beat: funding-deals trust: confirmed source_count: 4 cited_manifest_ids: ['1792799983072806532', '1792842071176472210', '1792815463347360757', '1792776659944244018'] --- # All-In, Horowitz, and Anthropic's own engineering shift describe an economy where few spenders fund the boom and fewer capture it *AI spending is concentrated in a few hands, and the people closest to the money say the durable advantage sits elsewhere.* Confirmed · 4 sources · Artificial Intelligence · Published 2026-10-24 12:31 UTC ## Key facts - 1% of companies represent 90% of all AI spending, according to the All-In podcast discussion. [cites: 1792799983072806532] - 80% of the top 90% of AI spending comes from unprofitable VC-backed companies, according to the same discussion. [cites: 1792799983072806532] - Durable advantage in AI applications comes from proprietary data and customer possession rather than just model architecture, according to Ben Horowitz. [cites: 1792842071176472210] - As of May 2026, Claude-written code accounted for over 80% of Anthropic's production codebase. (self-reported) [cites: 1792815463347360757] ## The concentration numbers The All-In podcast discussion, featuring Nikesh Arora and the All-In hosts, puts the concentration starkly: 1% of companies represent 90% of all AI spending, and 80% of the top 90% of AI spending comes from unprofitable VC-backed companies. The hosts also discuss how the US economy is currently highly correlated with the AI trade. Those are podcast claims, not audited figures, and they should be read as the hosts' framing of the market. But the framing itself is the news: even the people inside the boom are describing it as narrow, funded by companies that do not yet make money, and correlated with the broader economy in a way that makes the concentration everyone's problem. [cites: 1792799983072806532] The same discussion points to where enterprises are looking for an exit from that concentration. Enterprises are gravitating toward open-source and open-weight AI models to ensure explainability, auditability, and verifiability, the hosts say. That is a procurement motive, not a philosophical one. If the spending is concentrated and the vendors are unprofitable, buyers want models they can inspect, audit, and keep running if the vendor's economics change. Open weights are a hedge against concentration, and the hedge is getting more attractive as the concentration gets more visible. [cites: 1792799983072806532] ## Horowitz's answer: own the customer, not the model Ben Horowitz, speaking on the a16z show, offers a theory of where durable value sits in this market. He argues that AI is a probabilistic computer, whereas previous computing eras were based on deterministic computers, and that AI applications can reach massive user numbers instantly because they do not rely on network-effect adoption curves. That speed cuts both ways: if anyone can reach users instantly, distribution alone is not a moat. Durable advantage in AI applications, he says, comes from proprietary data and customer possession rather than just model architecture. [cites: 1792842071176472210] > **From Synorb** — Ask what Synorb covers. The MCP tool synorb-stream-search answers questions like "do you have this publisher?" without using any Manifest allowance. [Connect and ask](https://synorb.com/connect/start) Horowitz extends the argument to the craft of building companies. He says entrepreneurship requires a no choice mindset where the founder feels compelled to build the company, that the primary role of a venture capitalist is to provide momentum to founders in hiring, fundraising, and customer acquisition, and that CEO decision-making is a situational skill that requires confidence to act without knowing if the decision is correct. Successful market entry, he says, requires a product that is 10 times better than the incumbent, not just marginally better. The VC industry, he adds, has evolved toward hiring former operators to provide more relevant advice to founders. Read against the concentration numbers, the message is consistent: in a market where models are widely available and spending is concentrated, the scarce assets are customers, proprietary data, and the operators who can hold a team together while building them. [cites: 1792842071176472210] ## The labor shift inside the labs Open Source China reports a striking internal shift at Anthropic that fits the same story. The company's 200-person platform engineering team now uses Claude for all code generation, and engineers have shifted their focus to system design and validation. As of May 2026, Claude-written code accounted for over 80% of Anthropic's production codebase. The team used Claude to build components and monitor results during a Python to Rust migration, and Anthropic allows AI agents to autonomously fix tests without human intervention, though engineering leadership remains cautious about allowing agents to modify the codebase directly. [cites: 1792815463347360757] The hiring numbers in the same report are the part founders should watch. The proportion of junior staff in Anthropic's technical department fell from 21% in early 2025 to 14% by October 2026, and the share of junior technical roles in Anthropic's hiring fell from 27% to 9% by October 2026. If the frontier labs themselves are hiring fewer juniors while AI writes most of their code, the entry ladder into the industry is narrowing at the same time the spending is concentrating. That has consequences for who gets trained, who gets hired, and which companies can even staff an AI team. [cites: 1792815463347360757] Albert Invent's funding rounds out the picture from the applied side. The company announced a growth equity investment from Applied Ventures, and Applied Ventures sees opportunity for Albert's technology in the semiconductor ecosystem. The investment will be used to expand Albert's platform and global footprint, and the company's AI capabilities are aimed at shortening development cycles and reducing time to market. Capital is still flowing, but the flow described here goes to applied platforms with a named industry and a named customer problem, not to general capability. That is consistent with Horowitz's thesis: the money follows possession of a customer and a dataset, not possession of a model. [cites: 1792776659944244018] ## What is still unresolved - Whether the concentration figures hold up against audited spending data. - Whether open-weight adoption by enterprises accelerates if vendor economics worsen. - Whether the junior hiring decline spreads beyond frontier labs to the wider industry. ## Decision card - **Useful for:** VCs evaluating where durable AI value sits; Founders positioning against concentrated spenders - **Not for:** Anyone seeking a market timing call on AI spending; Teams without proprietary data or customer access - **Compare against:** Model-architecture-based moats; General-purpose AI capability bets ## Provenance ledger | Manifest id | Title | Source | Claims used | Published | |---|---|---|---|---| | `1792799983072806532` | AI Cyber Panic, Ken Griffin's Bubble Warning, The Fall of France? | [all-in-podcast](https://allinchamathjason.libsyn.com/ai-cyber-panic-ken-griffins-bubble-warning-the-fall-of-france) | 1792800416815888703, 1792800416819701976, 1792800416831389275, 1792800416821310463 | 10-23 00:00 | | `1792842071176472210` | What It Takes to Win as a Founder \| Ben Horowitz | [andreessen-horowitz-a16z-show-podcast](https://a16z.simplecast.com/episodes/what-it-takes-to-win-as-a-founder-ben-horowitz-AjG_7MSY) | 1792842673625272634, 1792842673620768650, 1792842673621182596, 1792842673616029763, 1792842673628821055, 1792842673630264055, 1792842673639053884, 1792842673635303310 | 10-24 00:00 | | `1792815463347360757` | Anthropic 平台工程团队 200 人的代码，全是 Claude 写的 | [open-source-china-news](https://oschina.net/news/503177) | 1792815758439060955, 1792815758430823539, 1792815758432980228, 1792815758434445680, 1792815758436800943, 1792815758437766019, 1792815758441409042, 1792815758442544737 | 10-23 00:00 | | `1792776659944244018` | Albert Invent welcomes Applied Ventures as investor | [albert-invent-newsroom](https://albertinvent.com/press-release-post/albert-invent-welcomes-applied-ventures-as-investor) | 1792777007808325562, 1792777007819241278, 1792777007814137426, 1792777007823710296 | 10-23 00:00 | ## Run these Manifests yourself Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them. 1. **Get a Synorb Key.** Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{"client":"your-agent-name"}`. The key is the `api_key` field of the answer. Save the key. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; 3 of them are starter Beacons. 2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer `. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief. 3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them. Cited Manifest ids: - `1792799983072806532` - `1792842071176472210` - `1792815463347360757` - `1792776659944244018` ```sh for id in 1792799983072806532 1792842071176472210 1792815463347360757 1792776659944244018; do curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id" done ``` > **From Synorb** — Run the cited Manifests yourself. Every claim in this paper points to a Manifest id. Get a Synorb Key and read the Manifests behind it. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; 3 of them are starter Beacons. [Get a key](https://synorb.com/connect/start) ## Related stories - [The AI trade's doubters are naming names: short CoreWeave, peak earnings in 2027](https://hangingcontext.com/stories/2026-10-23/ai-trade-doubters-name-positions/) - [Physical AI is standardizing: Digit 5 goes all-in on NVIDIA Thor as robot safety gets certified](https://hangingcontext.com/stories/2026-10-23/physical-ai-standardizes/) - [a16z renames its crypto fund Frontier while Lightspeed leads $215M into robotic chemistry](https://hangingcontext.com/stories/2026-10-23/ai-capital-goes-frontier-physical/) - [Open weights are a few points from closed — the race now is who serves them](https://hangingcontext.com/stories/2026-10-23/open-weights-deployment-race/) ## Links - HTML: https://hangingcontext.com/stories/2026-10-24/ai-spending-concentration-value/ - JSON: https://hangingcontext.com/stories/2026-10-24/ai-spending-concentration-value/index.json - Markdown: https://hangingcontext.com/stories/2026-10-24/ai-spending-concentration-value/index.md Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify. --- Reuse (terms v1): Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product. Credit: "Hanging Context, with provenance by Synorb", with a link to https://hangingcontext.com/stories/2026-10-24/ai-spending-concentration-value/. Terms: https://hangingcontext.com/terms/reuse/