Buying a communications platform this year means buying several different products at once. The RCS rollout stopped being a rumor, WhatsApp opened its call path, the EU started enforcing AI-disclosure rules with real fines, and a shortlist that says "SMS plus a voice SDK" answers a question nobody asks anymore. This post synthesizes the four category shifts our news explainers covered one at a time — RCS carrier enablement, WhatsApp Business Platform expansion, EU AI Act applicability, and APAC's regional super-apps — into the single "state of multichannel in H2 2026" read a buyer or builder can plan against.
Quick answer: the category moved from "connect channels" to "manage a per-destination channel mosaic with a compliance surface." RCS is a real business channel whose reach is carrier-by-carrier, WhatsApp now carries voice as well as messaging, AI agents that talk to people are regulated deployer-side in the EU, and APAC reach requires regional apps integrated per-market. Vendor evaluation now has to weigh fallbacks, tenant-owned compliance controls, and per-channel cost visibility, not headline channel counts.
(1) Four shifts that define H2 2026
RCS moved from Android-bubble to both-OS business channel
The RCS carrier-state explainer breaks down what 2026 actually delivered: Apple shipped RCS in iOS 18, carriers publicly enabled RCS on iPhone subscribers across the US, UK, and major European markets through 2025–2026, and Apple Messages for Business opened RCS delivery for registered brand agents beyond iMessage. Google pushed Universal Profile 3.0 on Android. The operating rule that survives all of it: the destination carrier decides. OS support sets the ceiling; the recipient's carrier sets the floor. The planning question changed from "is RCS real yet?" to "which slice of my audience can actually receive it, and what falls back?" — which is why per-recipient capability checking and an org-level RCS → SMS fallback chain, not headline reach, are the surfaces that matter. The launch sequencing (agent registration, template approval, staged rollout, measurement) is in the RCS launch checklist post, and the channel-selection trade-off against WhatsApp is in RCS vs WhatsApp for Business Messaging.
WhatsApp Business expanded from templates into voice
Meta made the WhatsApp Business Calling API generally available in July 2025, adding user-initiated and business-initiated VoIP calls to the same Cloud API a business uses for messaging — with WebRTC media, calls webhooks, and SIP interworking for contact-center estates, as the WhatsApp calling GA analysis documents. The move retires the "WhatsApp is text-only" objection that used to break thread continuity at the moment a customer wanted to talk. Coupled with the RCS-vs-WhatsApp trade-off above, the H2 2026 answer for most businesses is still "both, routed per market" — but the shortlist math changed, because a single account can now carry the customer-facing number for chat and voice together.
The EU AI Act's transparency tier became enforceable on August 2, 2026
For deployers of human-facing AI — which includes every AI voice agent and chat agent touching EU contacts — Article 50's disclosure and content-marking duties moved from preparation to operation: the person must be informed they are interacting with an AI, and AI-generated output must be machine-readably marked. The duty lands on the deployer (the tenant running the agent), not the platform. Devotel Orbit ships the control surface — configurable voice intro, chat notice, metadata.ai_generated stamping, jurisdiction-rulable settings, evidence exports — as tenant-owned configuration. Fines run into the tens of millions of euros, so "is the disclosure on?" is now a procurement and audit question, not a roadmap item.
APAC messaging is a regional mosaic, and SMS plus WhatsApp undercovers it
The APAC channel guide maps the four region defaults: LINE for Japan, Thailand, and Taiwan; WeChat for China; KakaoTalk for Korea; Zalo for Vietnam. Three of the four are template-only (WeChat, KakaoTalk Alimtalk, Zalo ZNS); LINE and KakaoTalk Friendtalk carry free-form. All four are bring-your-own-credential on Orbit behind one messaging API, one contact record, one webhook surface, with org-level fallback to SMS for the unreachable recipient. A global CPaaS claim that stops at "SMS + WhatsApp" leaves those markets structurally uncovered, and the beyond-SMS OTT channel review covers the wider OTT landscape the same way.
(2) What changed versus 2025 — the rolling-news record
Treat this section as the shopping list of shifts that solidified between the 2025 cycle and now, each one already documented in a published post rather than asserted here:
- RCS on iPhone moved from P2P to business delivery. The iOS-18 RCS base became Apple Messages for Business delivery for verified agents on publicly-enabled carriers — coverage is carrier-gated rather than universal. (carrier-state post)
- WhatsApp added a native call path. Business-initiated and user-initiated VoIP went GA in July 2025, with per-user-per-day limits on business-initiated calls and a short exclusion list of country codes. (WhatsApp calling GA)
- EU AI Act Article 50 became enforceable. August 2, 2026 moved AI-disclosure from preparation to operation, with penalties attached. (EU AI Act explainer)
- SMS pumping moved from ops nuisance to procurement criterion. Regulators named artificial-traffic inflation, carriers published AIT cost figures, and pumping tooling commoditized — so pre-send fraud gates and conversion-anomaly reporting became buying criteria. (SMS-pumping explainer)
- Agent-side economics got a control surface. The token ledger — not SMS/voice minutes — is now the line that can double overnight, and Orbit shipped a cost governor, LLM spend dashboard, and Agent ROI attribution to keep it accounted. (LLM FinOps post)
- The category's comparison matrixes updated. WhatsApp calling landed as shipped parity for the Cloud-API providers and absent from several UCaaS incumbents' telephony suites — the ranked guide CPaaS alternatives and competitors 2026 reflects the current matrix.
(3) What platform builders should operationalize in the next six months
Grounded in shipped Orbit surfaces — nothing here is roadmap:
- QA the AI voice agent like you would a human agent, except at 100% coverage. The Quality area (quality hub, per-agent scorecards, Evaluations, Leaderboard, Practice) scores every conversation against your rubrics and turns failures into replayable scenarios. The AI voice-agent QA guide walks rubric design and the practice loop.
- Put a budget on the model side of every agent. The
/insights/cost-governorsurface caps per-tenant LLM tokens (and ASR), downshifts routing near the cap, and reports cost-per-resolution margin per agent. The LLM spend FinOps post shows the three-surface loop. - Refuse fraud before the money moves, then watch conversion per destination. On the Verify API, the pre-send Fraud Guard returns a synchronous 403 on policy-blocked destinations, and
GET /api/v1/verify/conversion-anomalyflags prefixes converting far below your tenant baseline with an estimated spend-at-risk figure. The SMS-pumping explainer covers both policy and report. - Ship with a fallback chain, and measure the downgrade rate per market. Configure the RCS → SMS or OTT → SMS chain once per organization; silent fallbacks are the 2026 version of the old "RCS is almost here" mistake. Read rate and rich interaction split by destination market, never as a blended delivered count. The RCS carrier-state explainer lists the five checks.
- Turn the EU AI Act disclosure surface on for EU-facing traffic, and export the evidence. Voice intro, chat notice,
metadata.ai_generatedmarking, and the conformity/evidence exports are tenant-assembled postures; the deployer checklist has the four worked configurations.
(4) A buyer's checklist for evaluating a CPaaS vendor in this cycle
Score each vendor against these questions; a "channel count" answer to any of them is a red flag:
- Per-destination channel truth, not channel count. Does the platform expose a capability check per recipient, and does its reporting split RCS/WhatsApp/OTT reach by destination market rather than blending delivered counts? (RCS vs WhatsApp comparison)
- Configured, measurable fallbacks. Are fallback chains org-level and visible in analytics, so a downgrade shows as a downgrade instead of disappearing? (RCS launch checklist)
- Tenant-owned compliance controls for AI and consent. AI-disclosure (voice intro / chat notice / marking), two-axis recording consent, suppression, and evidence export are surfaces the vendor ships and you turn; a vendor "handling compliance for you" without controls is not a posture. (EU AI Act explainer)
- Pre-send fraud gates on verification traffic. A synchronous pre-send policy refusal, plus a conversion-anomaly report per destination, is the bar for OTP/verify endpoints in 2026. The ranked guide Best SMS-Pumping Anti-Fraud Verification APIs 2026 scores providers on exactly this.
- Cost surfaces for the AI side of the pipeline. A cost governor, LLM spend dashboard, and agent-ROI attribution — the token ledger is the line that can double overnight, so ask whether it is bounded and attributed. (LLM FinOps post)
- APAC and regional coverage as integrated channels, not a partnership slide. LINE / WeChat / KakaoTalk / Zalo should be connectable bring-your-own-credential channels on the same API and webhook surface, with SMS fallback. The APAC channel guide names the per-region defaults.
- Compliance evidence that exports. When procurement or an auditor asks for the posture, the vendor should be able to generate a signed pack from what your workspace actually configured — the empty-posture export should be empty. (EU AI Act explainer)
Cross-reference the ranked comparison pages as you run the checklist: CPaaS alternatives and competitors 2026, Best SMS API providers 2026, Best AI voice agent platforms 2026, and the omnichannel communications platforms 2026 guide.
Frequently asked questions
What does "state of CPaaS H2 2026" actually mean?
It means the four shifts this post synthesizes — RCS carrier rollout, WhatsApp Business expansion into voice, EU AI Act transparency enforcement, and APAC regional-app fragmentation — combined into a single buying and building picture. The synthesis is the point: the individual shifts are each covered in a dedicated explainer; this post is where they are treated as one market.
Is RCS ready to plan against in H2 2026?
Yes, with the floor/ceiling caveat: both major OSes support RCS and the business layer exists on both, but delivery depends on each destination carrier's enablement. Plan markets against the recipient's carrier, probe recipients with a capability check before scaling, and ship with a configured fallback. (RCS carrier state)
Does the EU AI Act apply outside the EU?
It applies to deployers of AI systems that interact with people in the EU, wherever the deployer is based. If your AI voice agent or chat agent touches EU contacts, Article 50's disclosure and marking duties have been enforceable since August 2, 2026 — the tenant-owned controls (voice intro, chat notice, content marking) exist to satisfy it. (EU AI Act explainer)
How should a buyer weigh WhatsApp vs RCS vs APAC regional apps?
Per market, not globally. RCS now reaches the native inbox on both OSes where carriers enable it; WhatsApp wins where its installed base is habitual; LINE, WeChat, KakaoTalk, and Zalo hold their respective APAC defaults. Most businesses run both and route per message, with SMS as the universal floor. (RCS vs WhatsApp and the APAC guide)
What is the single biggest procurement change versus 2025?
Fraud and AI-lane controls moved from ops notes to buying criteria. SMS-pumping defense (pre-send gates plus conversion-anomaly reporting) and AI-agent cost/disclosure surfaces are now shortlist rows, not nice-to-haves. The SMS-pumping explainer and the LLM FinOps post cover the two classes.
The takeaway
H2 2026 is the cycle where multichannel stopped being a channel list and became a discipline: per-destination reach proof, configured fallbacks, tenant-owned compliance, and bounded AI spend. Each of those disciplines is already shipped somewhere on Devotel Orbit and documented in the linked explainers. Run the checklist, read the ranked comparisons, and plan against destination-carrier floors rather than OS-supported ceilings.
Published 28 August 2026.