Autoresearch: Tony Seba / RethinkX technology-disruption S-curves and the investing lens

Generated by /autoresearch on 2026-06-16. Synthesized across 3 rounds from 6 web pages (one fetch failed: ARK 403), no Grokipedia entry found for Tony Seba/RethinkX. See Provenance. Treat as raw material — review before promoting into a project or thread. Context: vault/threads/technology-adoption-s-curves Pairs with the same-day academic-research clipping (diffusion/Bass/Wright's-Law canon). Promote both together.

Summary

Tony Seba and RethinkX (founded 2016 with James Arbib) frame technology disruption as the product of converging cost curves: several independent technologies each improving exponentially in cost/capability eventually intersect, opening a "vast new possibility space," and adoption of the resulting product is non-linear, tracing an S-curve from ~1-2% penetration to ~80% with a tipping point in between (RethinkX). Mainstream analysts miss this because they "extrapolate the present into the future in a straight line" and assume cost curves improve "while all else remains equal." The framework's track record is genuinely mixed and instructive: Seba's cost-curve calls (solar becomes cheapest power; cheap lithium batteries) were right and early, dismissed at the time (energi.media); his adoption-timing calls (95% of US miles via autonomous "Transportation-as-a-Service" by 2030, used-ICE prices to zero) are badly behind. The recurring critique — "right spot, wrong timeframe" — is the single most important caveat for an investing application: cost curves are forecastable; adoption timing and the ceiling are not. The math is a logistic with three parameters (ceiling L, steepness k, inflection x₀) (Dr Li), and time-to-50%-penetration has itself been shrinking across eras (telegraph 56y → PC 16y → internet 7y → smartphone 5y → AI tools ~3y). For investing, the actionable translation (ARK-style) is that where a technology sits on its curve and how steep that curve is governs how much growth remains — and the steepest, cost-driven curves (AI inference falling ~10×/yr) can even expand the addressable market rather than merely fill it.

Findings

The Seba / RethinkX framework: converging cost curves → convergence → S-curve adoption

The framework's mechanism is explicit: "technologies within our system are improving in costs and capability at different rates," and when multiple exponentially-improving technologies intersect at the same time, they enable products that linear forecasters miss entirely. The canonical example is the smartphone — the convergence of 2.5G connectivity, touchscreens, sensors, processing power, and energy-dense lithium-ion batteries (RethinkX; WebSearch summary). Adoption then follows a sigmoid: slow at first (<1-2% penetration, "an illusion of gradual change"), an acceleration phase past a tipping/"rupture" point where "change is almost inevitable," and a saturation phase near ~80%. RethinkX stresses the qualitative nature of the shift — "A butterfly is not a faster caterpillar" — i.e. disruption is a phase change, not a faster version of the incumbent. The same dynamics, they argue, scale from products to the "five foundational sectors" (information, energy, food, transportation, materials), which they predict will see 10× cost reductions and 90% fewer resource inputs in the 2020–2033 window.

Why incumbents and analysts systematically miss it

Two named failure modes. First, linear extrapolation: forecasters "examine risks and opportunities by extrapolating the present into the future in a straight line," missing the amplifying feedback of interconnected systems (RethinkX). Second, a methodological flaw in incumbent cost accounting: RethinkX's 2021 "Great Stranding" report argues conventional Levelized Cost of Energy (LCOE) math systematically overvalues coal/nuclear/gas and undervalues solar+battery, driving capital misallocation (Wikipedia: RethinkX). RMI frames the same point from the other side: "the adoption rate of innovations is non-linear; it is slow at first, then rapidly rises," which is "why wind, solar, and battery technologies surprised analysts — prices dropping faster and further than many believed possible." RMI names four feedback mechanisms that bend the curve into an S and set its steepness: learning curves, economies of scale, technological reinforcement, and social diffusion (RMI).

The cost curve is the engine — and AI is the steepest one yet

The S-curve of adoption is downstream of a cost curve (Wright's Law: cost falls a roughly constant % per doubling of cumulative output). The clearest live example is "LLMflation": the cost of LLM inference at a fixed capability is falling ~10× per year. a16z's worked figures: an MMLU-42 (entry-level) capability cost $60/million tokens in Nov 2021 (GPT-3) and $0.06 by Nov 2024 (Llama 3.2 3B) — a ~1,000× decline in three years; high-performance MMLU-83 fell ~62× since GPT-4's launch. This is "even faster than … compute cost during the PC revolution or bandwidth during the dotcom boom," and the log-linear fit is explicitly an experience-curve dynamic (a16z LLMflation). A crucial nuance for stock-picking: while per-capability cost collapses, the cost of running frontier models has risen ~3–18×/yr (more inference for marginal gains) — so "the cost curve is falling" and "frontier spend is exploding" are simultaneously true. The energy analogues: utility solar PV under ~4¢/kWh and ~$100/kWh lithium packs, both of which Seba forecast early (energi.media track-record piece).

The math of position and steepness

The standard model is the logistic y = L / (1 + e^(−k(x−x₀))), with three parameters that map directly onto the investing questions (Dr Li):

• L = market potential / ceiling ("how big does this get?") — the parameter the academic literature flags as hardest to estimate.
• k = growth rate / steepness ("how violent is the re-rate?") — "steeper k values indicate faster transitions."
• x₀ = inflection / midpoint ("are we before or after the knee?") — the transition from early adoption to mainstream.

Position can be read against Rogers' adopter segmentation (innovators 2.5%, early adopters 13.5%, early majority 34%, late majority 34%, laggards 16%) — the inflection sits around the early-majority crossing. A striking regularity: time-to-50%-penetration has compressed across eras — telegraph 56 years, radio 22, PCs 16, internet 7, smartphones 5, AI tools ~3 (projected) — i.e. modern curves are steeper (higher k). The same source fits AI adoption with a Bass model at innovation coefficient p ≈ 0.01 and imitation q ≈ 0.8 (very high network effect), projecting AI reaching ~77% of internet users by 2030 — figures that "exceed historical technology adoption speeds" and should be treated as illustrative, not precise.

The track record — the honest part (cost curves right, timing wrong)

This is the load-bearing caveat. Seba's cost-trajectory forecasts have been notably accurate: as early as 2010 he forecast solar becoming the cheapest electricity in most of the world by the 2020s — "widely dismissed at the time as overly optimistic" — and his solar and battery cost calls came true (Seba track-record synthesis via WebSearch). But his adoption-timing forecasts have missed badly: the 2017 prediction that 95% of US vehicle-miles would be autonomous Transportation-as-a-Service by 2030 (at <18¢/mile, 10× cheaper than ownership), the 2018 claim of a production-ready fully-autonomous vehicle "within ~1,000 days," cheap robotaxis, and used-ICE prices going to ~zero "by around now" are all well behind (energi.media; Wikipedia: RethinkX). Critics' framing is consistent: the direction is "about the right spot," but "his time frame and some key details are wrong," and his entire transportation scenario was conditional on full autonomy arriving in the early 2020s — which it didn't. Wikipedia's editors note no evidence the earlier (2010–2017) timed predictions materialized on schedule, and skeptics flag a profit-motive / attention incentive (dramatic claims attract investors). Food disruption (manufactured protein 5× cheaper by 2030, 10× by 2035; US cattle −50% by 2030) remains contested and unproven.

Where things sit on the curve right now (2026)

• Solar: reached a record ~11% monthly share of global electricity (May 2025) and ~8.7% of generation for the year — still a small share but "the dominant driver of change," meeting roughly three-quarters of last year's demand growth (Ember Global Electricity Review 2026; Axios). US utility solar additions: 27.2 GW in 2025 → 43.4 GW planned 2026 (+60%), a third straight record year (ess-news).
• Battery storage: 15 GW (2025) → 24.3 GW (2026, +~62%), five years of "exponential growth," >40 GW installed (ess-news).
• EVs: ~24.7% of global new-car sales in 2026 (~22.7M units), China >50%, Europe ~1-in-3, North America lagging (IEA Global EV Outlook 2026). Solar/wind/storage are described as climbing "phase 4" of the S-curve (the steep part) (RMI 2026 trends).

The investing translation: curve position → remaining runway

The practitioner investing frame (ARK, RMI, the TALC literature) converts the curve into an upside estimate: growth decelerates as a technology penetrates its addressable market, so the remaining runway depends on where on the curve you buy (WebSearch: investing/S-curve summary; RMI). Three usable ideas:

1. Position determines magnitude. Buying near the inflection (early-majority crossing, accelerating) captures the steep part; buying near saturation captures little even if the company is excellent — the literal "are we capturing enough growth" test.
2. Wright's Law links cost to demand. ARK's stated method: cost falls per Wright's Law → "positive deflationary forces drive down prices and drive up demand," and for some technologies (their EV example) growth can outperform the traditional S-curve by expanding the addressable market rather than just penetrating an existing one (ARK page returned 403; claim from the search snippet — verify before relying on it).
3. Speed favors fast movers. RMI: "the fastest movers will stand to capture the biggest rewards, and the slowest movers will be left with the biggest losses" — steepness (k) compresses the window to take a position.

The timing caveat from the track-record section is the counterweight: a steep cost curve does not guarantee a near-term adoption re-rate (autonomy is the cautionary tale), and the hype-cycle risk (paying near peak expectations) is real.

Contradictions and open questions

• Outpacing vs. undershooting — both happen, by segment. Solar and grid storage are beating mainstream forecasts (ess-news), yet 2026 road-transport battery demand is growing slower than previously expected because of weak EV sales in some markets and a shift to smaller-battery plug-in hybrids (IEA Global EV Outlook 2026) — and robotaxi/TaaS is far behind Seba's timeline. The S-curve lens must be applied per-segment, not to a whole theme.
• Cost forecastable, timing not. The framework's own track record shows the cost curve is the reliable part and the adoption date is the unreliable part — so an evaluation lens should weight cost-curve evidence heavily and treat inflection-timing as a wide distribution.
• Ceiling (L) and real-time inflection detection are unsolved. Knowing prospectively "we are at the knee" — the most valuable signal — is exactly what's hardest; the parameter estimates above (e.g. AI p/q, 77% by 2030) are illustrative and should not be treated as calibrated.
• Source incentives. RethinkX/ARK are advocates with commercial incentives; their "expanding addressable market" and "trillions of upside" framings are directional arguments, not neutral base cases. The ARK EV claim here is unverified (fetch failed).

Provenance

Rounds run: 3 (full)

Sub-questions by round:

Round 1 (broad survey):

1. The Seba/RethinkX framework (cost curves, convergence, S-curve, tipping point)
2. The "god of the gaps" / linear-extrapolation error analysts make
3. Recent solar/battery/EV adoption vs. forecasts (2025-2026)
4. AI/compute cost-decline curve
5. Using S-curve position for investing (penetration, growth remaining)

Round 2 (drill-down):

1. ARK's investing application of the S-curve / Wright's Law — targeted the "growth remaining" mechanism
2. AI inference cost-decline specifics (LLMflation) — targeted the steepest live cost curve
3. Seba's actual forecast track record and criticism — targeted honesty/timing risk

Round 3 (resolve remaining uncertainty):

1. The math of reading curve position & steepness (logistic L/k/x₀) — targeted operationalization
2. Where solar/EV sit on the curve right now (2026) — targeted current position

Anchor source (Grokipedia, attempted before round 1):

• No Grokipedia entry found for "Tony Seba" / "RethinkX" (search returned only unrelated "Seba" articles).

URLs fetched (6 successful, 1 failed):

Round 1:

• Disruption in depth — RethinkX — official/primary — the framework: cost curves, convergence, S-curve, rupture point, linear-extrapolation critique.
• RethinkX — Wikipedia — encyclopedic — named predictions (TaaS/energy/food), criticism, track-record skepticism.
• New US battery capacity in 2026 — ess-news — industry/news — 2025-26 solar & storage deployment figures.

Round 2:

• Harnessing the Power of S-Curves — RMI — institutional — 5 phases, 4 feedback mechanisms, fastest-mover argument.
• LLMflation — a16z — industry/VC — 10×/yr inference cost decline, 1000×/3yr MMLU-42, experience-curve framing.
• [EV Growth Outperforming the Traditional S-Curve — ARK] (https://www.ark-invest.com/articles/analyst-research/ev-growth-outperforming-the-traditional-s-curve-dynamics) — fetch failed: HTTP 403; "expanding addressable market" claim taken from search snippet, flagged unverified.

Round 3:

• Technology Adoption & Innovation S-Curves — Dr Li — blog/analysis — logistic math (L/k/x₀), Rogers segmentation, accelerating time-to-50%, AI Bass p/q.

Sources cited from WebSearch snippets (not individually fetched; cited where load-bearing): IEA Global EV Outlook 2026, IEA EV batteries, Ember Global Electricity Review 2026, Axios solar inflection, RMI 2026 trends, energi.media on Seba/autonomy timing, Warp News on Seba's record.

Tools used: WebSearch, WebFetch, grokipedia search (no entry). Generated: 2026-06-16 13:45 UTC