06-reference

stratechery deployco 70s apple intel

2026-05-13·reference·source: Stratechery·by Ben Thompson
ai-enterprise-deploymentfde-waveopenai-deploycotop-down-implementationpalantirapple-intelsemiconductor-supply

Stratechery: The Deployment Company, Back to the 70s, Apple and Intel

Three-section Stratechery Update. Section 1 (DeployCo) is the load-bearing one for RDCO; Sections 2 and 3 are summarized briefly.

Why this is in the vault

This is the first major industry-analyst piece to frame the OpenAI DeployCo + the broader Forward-Deployed Engineer wave as a structural strategic move, not just a hiring story. Thompson treats it as the inevitable enterprise-go-to-market shift for AI, and explicitly invokes his 2024 "Enterprise Philosophy" thesis: AI's value capture happens via top-down enterprise implementation, not bottom-up SaaS adoption.

Section 1: The Deployment Company (load-bearing for RDCO)

Thompson's framing of the DeployCo move:

Does he name the productization-gap dynamic? He acknowledges the engineer-hiring and stacks the announcements, but does not coin a "productization gap" frame. He frames it as the enterprise-philosophy thesis being validated: the agent layer needs "years of work necessary to get data in a state where humans can be replaced", and the labs themselves are now doing that work. The unspoken implication is that the model-quality-alone narrative is over.

Predictions about how DeployCo will compete with traditional consulting / SI vendors:

Section 2: Back to the 70s (brief)

This is Thompson restating his "AI = mainframe wave, not internet wave" analogy: "Transformers are the transistor, and mainframes are today's models. The GUI is, arguably, still TBD." Background framing for Section 1. Not a new thesis. Relevant context: he is doubling down that the right mental model for AI rollout is 1970s top-down IT deployment, not 1990s consumer-internet land-grab.

Section 3: Apple and Intel (brief)

WSJ-confirmed preliminary deal for Intel to manufacture some Apple chips (likely the basic M-series on Intel's 18A process per Ming-Chi Kuo). Thompson's read: the deal is economic, not geopolitical-pressure. Apple has been supply-constrained at TSMC for two quarters because AI compute is eating TSMC capacity. Cook explicitly said the constraint is "advanced nodes" availability. Thompson cites his own "TSMC Risk" thesis: hyperscalers and fabless companies finally realize that TSMC's monopoly + capex restraint is a bigger risk than China-blow-up. "Today's shortages may prove to be peanuts" vs end-of-decade foregone AI revenue. He thinks Intel is finally getting the anchor customer it needs.

Not directly RDCO-relevant. Filed for completeness.

Mapping against Ray Data Co

Strong convergence. This is the most authoritative analyst voice we have on the same FDE-wave thesis we filed yesterday. Thompson independently arrives at the same conclusion: the labs are not productizing - they are deploying humans because the work that closes the gap between model capability and enterprise value is data-reworking, ontology-mapping, and process-redesign, none of which scale via API alone.

Implications for RDCO:

  1. The productization-gap concept hardens. Thompson sees the same dynamic we do, but frames it through the enterprise-philosophy lens. Our framing (productization gap = the wedge for asymmetric edge) and his framing (top-down implementation = where AI's value gets captured) describe the same phenomenon from different sides. Both should live in the concept article. See [[concepts/2026-05-13-fde-asymmetric-edge-rdco-positioning]].

  2. Top-down vs bottom-up market segmentation. Thompson is explicitly talking about Fortune-500 + PE-backed enterprises. DeployCo, Google's FDE team, Anthropic's PE JV all target the same buyer: executive-suite, brutally-rational, EV-calculating, layoff-empowered. This leaves the SMB and operator-tier market structurally unserved. RDCO sits there. The Zack-on-Meta-ads-MCP item we filed today ([[2026-05-13-zack-igclaims-meta-ads-mcp-angle-mining]]) is the SMB-side worked example of the productization gap that DeployCo will not touch.

  3. The Palantir comparison is the right one for RDCO's larger ambition. Thompson's Foundry framing - "The Ontology-Powered Operating System for the Modern Enterprise" - is structurally what RDCO becomes if the operator-tier wedge works and we expand upstream into larger orgs. The ontology + data-reworking layer is unavoidable; the question is whether RDCO learns to do it at SMB-scale-out-economics, while DeployCo/Palantir do it at F500-bespoke economics.

  4. Distribution flywheel risk. The PE-buyout pattern Thompson calls out is the most aggressive version: PE buys ops-heavy SMBs, deploys AI, fires staff, captures margin. If that's the actual go-to-market for AI value capture in the next 36 months, RDCO needs to decide whether to (a) sit beneath it as the operator-tier tool that PE-backed portfolio companies use post-acquisition, (b) compete with it head-on at pre-PE-target SMBs, or (c) ignore it as a different market. Open question to surface to founder.

  5. Apple-Intel section is not RDCO-relevant but is a useful semiconductor-supply signal for any future RDCO surface that depends on inference-cost trajectories. Filed but not load-bearing.

Sharp verdict: this is the highest-quality external validation we have for the FDE-wave thesis. Section 1 belongs in any future Sanity Check issue on the productization gap as the canonical analyst-side citation.

Related