PAPER 2 OF 2 · CATEGORY FORECAST · APRIL 2026

The Absorption Calendar

What the Frontier Absorbs Between 2026 and 2030 — a Category-by-Category Forecast for CISOs, CAIOs, Founders, and Investors

STEVE ABBEY · CO-FOUNDER, TELEOX.AI · 2026-04-16 · v1.0

FOREWORD

The companion paper — The Middleware Collapse — argues a structural thesis. That thesis is useful if you are a frontier-lab chief scientist, a corp-dev team, or an investor thinking about portfolio posture.

It is less useful if you are a CISO doing vendor selection next quarter, a CAIO writing next year's AI architecture plan, a founder deciding whether to pivot, or an investor repricing a specific position. Those readers need a calendar.

This paper is that calendar. Each section is self-contained; skim the category that matches your decision.

— Steve Abbey, Teleox.ai

METHODOLOGY

How to Read This Calendar

Band 1 · 12–18 months (by Q4 2027)

Hyperscaler-native capability ships; procurement cycles begin moving.

Band 2 · 18–30 months (by H2 2028)

Category margin compression visible in revenue-per-customer and NRR.

Band 3 · 30–48 months (by 2029)

Category consolidates to 2–3 survivors per durable position.

Band 4 · 48–60 months (by 2030)

Residual absorption; survivors' moats provably durable or provably not.

Tier A

Structurally exposed — capability gap is central to moat.

Tier B

Partially defended — has a real durable-position asset alongside the exposed surface.

Tier C

Insulated or adjacent — primary business outside the absorption dynamic.

CATEGORY-BY-CATEGORY FORECAST

Four durable positions — strategic crossroads

Band 1 · 12–18 months

Voice Generation

BACKDROP

$2.4B → $47.5B by 2034. Contact-centre industry $14.7B. 30% of live agents being replaced in 2026 where AI voice is compliance-safe.

CAPABILITY GAP

Per-sentence SECS + Whisper WER + sanity-gate audit log that is a platform property, not a consultant deliverable.

TIER A — STRUCTURALLY EXPOSED

ElevenLabs — $11B / $330M ARR. Default voice layer. Structurally exposed because per-sentence verification is easier to ship natively than to retrofit.
Cartesia — $86–191M raised. Real-time TTS API with emotion.
Play.ht / Murf / Resemble / PlayHT — long-tail consolidation base case.

TIER B — PARTIALLY DEFENDED

Descript — $55M ARR. Podcasting workflow integration partially insulates.
Speechify — consumer/creator-economy niche.

TIMING

→ Q4 2027 — hyperscaler-native voice with per-sentence verification ships.

→ Q3 2028 — category ARR compression visible in filings.

→ 2029 — consolidation: 2–3 voice-AI survivors per durable position.

WHAT SURVIVES

Creator-economy specialists, sovereign/on-prem voice, or compliance-grade infrastructure builders. Generic unverified-voice APIs do not survive as venture-scale businesses.

Band 1–2 · 18–30 months

Avatar & Video Generation

BACKDROP

Enterprise avatar-video: $3.86B → $42.29B by 2033. Runway $5.3B. Luma AI $4B. Deepfake fraud: $3.5B US losses 2025, $58.3B projected.

CAPABILITY GAP

No per-frame identity verification, no cosine-signature audit trail, no provable-consent artifact.

TIER A — STRUCTURALLY EXPOSED

Synthesia — $4B / $146M ARR / 70% Fortune 100. Customer-depth moat is real but is not a capability defence.
HeyGen — ~$100M ARR / $500M. SMB-focused. Canva integration a defensive move.
Tavus (~$8M ARR), D-ID (280K devs), Colossyan, Hour One — long-tail consolidation.

TIER B — PARTIALLY DEFENDED

Runway — $5.3B. World-models posture (Position 2). Partially insulated.
Luma AI — $4B. Similar creative-generation posture.

TIMING

→ Q2 2028 — first hyperscaler ships native identity-locked avatar.

→ 2028–2029 — Synthesia/HeyGen reprice or pivot.

→ 2030 — 2–3 survivors (enterprise depth, sovereign, verification-infrastructure).

WHAT SURVIVES

Verification substrate owners, sovereign/compliance-gated verticals, or world-model/creative-generation model makers. Unverified corporate-training video is not durable.

Band 2 · 18–36 months

Brand Voice & Style Writing

BACKDROP

First category to partially collapse in public. Jasper pivoted twice, lost 60% subscribers. Copy.ai priced out by free-tier ChatGPT.

CAPABILITY GAP

Style alignment with verifiable consistency — a structural guarantee, not best-effort prompt-engineered approximation.

TIER A — STRUCTURALLY EXPOSED

Jasper — lost 60% subscribers. Peak $1.5B (2022); declining.
Copy.ai — absorbed by ChatGPT free tier.
Typeface, Tome, Anyword, Rytr, Sudowrite — writing-AI long tail.

TIER B — PARTIALLY DEFENDED

Writer — $1.9B / $200M Series C. Best-positioned: enterprise brand-voice workflow depth. Still exposed when hyperscalers ship TCT-conditioned LoRA fine-tuning.
Grammarly — browser-extension distribution partially insulates.

TIMING

→ H2 2027 — hyperscaler-native LoRA fine-tuning APIs ship.

→ 2028 — Writer's enterprise customers face per-output style-verification procurement question.

→ 2029 — category consolidation. Durable form: workflow-depth + compliance substrate.

WHAT SURVIVES

Enterprise writing-AI with deep regulated-vertical workflow integration plus a style-verification substrate. Generic AI writing assistants do not survive.

Band 1–2 · 12–30 months

AI Governance, Trust & Verification

BACKDROP

~46 AI-governance funding deals as of Feb 2026. Robust Intelligence → Cisco. Lakera → acquired (CB Insights, Sep 2025).

CAPABILITY GAP

Bolt-on heuristics vs. per-output cosine scores against a geometric manifold. A classifier can be circumvented; an arithmetic constraint cannot.

TIER A — STRUCTURALLY EXPOSED

Credo AI — launched GAIA governance agent. Microsoft/IBM/Databricks partnerships.
Fiddler AI, Arthur AI, Galileo, Patronus AI — observability/evaluation cohort.
HiddenLayer, Protect AI — model-security cohort.
Sycamore — $65M raise April 2026. New governance substrate entrant.

TIER B — PARTIALLY DEFENDED

Weights & Biases — acquired by CoreWeave. Infrastructure-layer inside a hyperscaler parent.
Arize AI, Truera — ML observability; adjacent.

TIMING

→ H2 2027 — hyperscaler-native Constellation-Guard equivalent ships.

→ 2028 — Credo-class vendors make fork-in-road: go deep on substrate or become Frontier connector.

→ 2029 — 46 → ~6–10 survivors (Position 2 substrate + Position 4 verification).

WHAT SURVIVES

Governance businesses that own a substrate — prompt-injection resistance, per-output cosine verification, or regulatory-certification paths the hyperscaler cannot vertically integrate.

Band 3 · 24–48 months

Data Labeling & RLHF Services

BACKDROP

Scale AI 49% Meta-owned at $29B. Meta has publicly criticised Scale data quality. Surge AI $1.2B rev. Mercor $10B.

CAPABILITY GAP

Human annotation ceiling. Scaling laws require labeled signal, not raw tokens. TCT bypasses the human-hour scaling curve.

TIER A — STRUCTURALLY EXPOSED

Scale AI — strategic slow-absorb base case. Remaining 51% is a run-off unless it pivots.
Mercor — $10B priced on continued expert-labour scaling.
Labelbox, Snorkel AI, Appen — absorbed/declined base case.

TIER B — PARTIALLY DEFENDED

Surge AI — bootstrapped, profitable, specialist-RLHF. Survives for RLHF-appropriate problems where TCT is not the correct frame.

TIMING

→ 2027–2028 — first frontier lab adopts TCT at scale.

→ 2029 — training-signal-as-asset-class market emerges.

→ 2030 — 2–3 human-annotation specialists plus labs' own pipelines.

WHAT SURVIVES

Human-annotation services that pivot into problem classes TCT does not cover — open-ended preference elicitation, domain-expert evaluation. Meaningfully smaller than current valuations.

Band 2–3 · 24–48 months

Vector DBs & RAG Infrastructure

BACKDROP

RAG market $9.86B → $64.5B by 2035. Voyage AI → MongoDB. Jina AI → Elastic.

CAPABILITY GAP

Single-embedder RAG ceiling. 13-embedder RAG with RRF and cross-correlation anomaly detection is categorically better.

TIER A — STRUCTURALLY EXPOSED

Pinecone, Weaviate, Qdrant, Chroma — standalone vector-DB cohort. API commoditising.

TIER B — PARTIALLY DEFENDED

LlamaIndex, LangChain — RAG orchestration. Position 2 candidates if they become infrastructure.

TIMING

→ 2027 — first hyperscaler ships 13-embedder-class retrieval.

→ 2028–2029 — cohort consolidation. 2–3 survivors with multi-embedder differentiation.

WHAT SURVIVES

Retrieval infrastructure with multi-embedder decomposition, per-customer embedder-ensemble configs, and retrieval-quality audit trails.

Band 2–3 · 18–48 months

Regulated-Vertical AI Agents

BACKDROP

Frontier's six launch partners. Harvey $11B. Abridge $5.3B. Ambience ~$1B.

CAPABILITY GAP

Citation-grade output guarantees, per-output provenance chains for FDA/IRB/EMA, per-response compliance auditability.

TIER A — STRUCTURALLY EXPOSED

Sierra, Decagon — customer-support AI. Hardest path: Frontier is designed to own this workflow natively.
Clay — GTM data. Defensibility depends on becoming picks-and-shovels.

TIER B — PARTIALLY DEFENDED

Harvey — $11B. Real BigLaw depth. Exposed because citation-grade verification is platform-reachable.
Abridge — $5.3B. Deep Epic integration; 250+ health systems. Strongest customer-depth moat in the Frontier Partner cohort.
Ambience — ~$1B. Similar to Abridge, less penetration, higher absorption risk.

TIMING

→ 2027 — Frontier native verification ships. Partners evaluate deepening vs. parallel substrates.

→ 2028 — partnership renegotiations. Leverage only if they own proprietary data or regulatory certification.

→ 2029–2030 — vertical consolidation. Survivors: deep-enough integrations or proprietary-enough domain data.

WHAT SURVIVES

Agents that own proprietary context (Position 1) + workflow depth (Position 3) measured in years. Abridge's Epic integration is the archetype.

Band 1–3 · 12–48 months

Enterprise Search & Workflow

BACKDROP

Glean $200M ARR (doubled in 9 months), $7.2B valuation. Moveworks → ServiceNow (already absorbed).

CAPABILITY GAP

Competing with Copilot default-install distribution. Knowledge-graph-as-proprietary-substrate is the credible Position 1 play.

TIER A — STRUCTURALLY EXPOSED

Cresta — contact-centre AI. Same voice/contact-centre absorption dynamic.

TIER B — PARTIALLY DEFENDED

Glean — $7.2B / $200M ARR. Strongest enterprise-AI-search posture. Three-year test: can the knowledge-graph substrate produce enough incremental value over Copilot?
Gong — revenue-workflow depth partially insulates.

TIMING

→ 2027–2028 — Copilot/Frontier ship knowledge-graph-equivalent defaults. Glean tested at renewal.

→ 2028–2029 — 2–3 enterprise-AI-search survivors with proprietary-substrate moats.

WHAT SURVIVES

Proprietary-substrate moats (knowledge-graph ontologies the enterprise's processes depend on). Generic enterprise-AI chatbots do not survive.

ONE-PAGE CALENDAR VIEW

CATEGORY	PUBLIC COHORT LEADERS	BAND	TIER MIX
Voice generation	ElevenLabs ($11B), Cartesia, Play.ht, Murf	1 · 12–18 mo	Tier A heavy
Avatar & video (enterprise)	Synthesia ($4B), HeyGen ($500M), Tavus, D-ID	1–2 · 18–30 mo	Tier A heavy
Creative video (world models)	Runway ($5.3B), Luma AI ($4B), Pika	3–4 · 30–60 mo	Tier B
Brand voice / writing	Writer ($1.9B), Jasper, Copy.ai	2 · 18–36 mo	Writer Tier B
AI governance / trust	Credo AI, Sycamore ($65M); Lakera, Robust Intel absorbed	1–2 · 12–30 mo	Tier A majority
Data labeling / RLHF	Scale AI ($29B), Surge ($1.2B), Mercor ($10B)	3 · 24–48 mo	Surge Tier B
Vector DB / RAG infra	Pinecone, Weaviate, Qdrant; Voyage/Jina absorbed	2–3 · 24–48 mo	Tier A majority
Regulated-vertical agents	Harvey ($11B), Abridge ($5.3B), Ambience (~$1B)	2–3 · 18–48 mo	A/B mix
Enterprise search / workflow	Glean ($7.2B); Moveworks absorbed	1–3 · 12–48 mo	Tier B
Thin-wrapper app layer	Hundreds of cohort	4 · 36–60 mo	Tier A default

WHAT TO DO

Actionable Framework, by Role

If You Are a CISO or CAIO

Which durable position does each vendor in your shortlist hold?
If the answer is none, what happens when the platform ships this vendor's capability natively in 18–30 months?
Does your 2028 switching cost exceed the 2026 capability premium?
For Band 1 categories: does the vendor have a credible per-output verification roadmap?

If You Are a Founder

Which of the four durable positions does your product hold?
What is your specific 12-month repositioning move?
If you cannot name the move, the 2026 round closes; the 2028 round is a down round or a strategic sale.
The time to reposition is while you still have revenue, customers, and optionality.

If You Are an Investor

Long the frontier labs with the clearest model-maker posture.
Long pick-and-shovel infrastructure the agents depend on.
Neutral-to-negative on middleware companies whose moat is customer-depth only.
Watch Teleox deal flow — a single acquisition reprices every category simultaneously.

“The direction is the load-bearing claim, not the date. Everything else is renting.”

— Steve Abbey · Teleox.ai · 2026-04-16

Book 20-Min Briefing →Schedule 48-Hour POC →

← Paper 1: Middleware Collapse All Research →

How to Read This Calendar

Band 1 · 12–18 months (by Q4 2027)

Hyperscaler-native capability ships; procurement cycles begin moving.

Band 2 · 18–30 months (by H2 2028)

Category margin compression visible in revenue-per-customer and NRR.

Band 3 · 30–48 months (by 2029)

Category consolidates to 2–3 survivors per durable position.

Band 4 · 48–60 months (by 2030)

Residual absorption; survivors' moats provably durable or provably not.

Tier A

Structurally exposed — capability gap is central to moat.

Tier B

Partially defended — has a real durable-position asset alongside the exposed surface.

Tier C

Insulated or adjacent — primary business outside the absorption dynamic.