AI Control Plane Market Map 2026-07 — After the H1 Consolidation

License: CC-BY 4.0 — reproduce freely with attribution to Visionality (visionality.ai).

Between January and May 2026, three of the four independent AI Control Plane vendors got acquired. If you're evaluating vendors right now, half the market map from six months ago is out of date. This article is what we've been telling investors, partners, and prospects since the Portkey acquisition closed: here's what the category actually looks like on 2026-07-02.

1. Category taxonomy — four columns, not one

The AI tooling category is often collapsed into "AI gateway" or "LLM observability" in press coverage. That collapse hides real differences in what each product does and who buys it. Four columns:

| Column | What it does | Vendors as of 2026-07 | |---|---|---| | Cost tracking | Reports what you already spent, retrospectively | Vantage · CloudZero · Amnic · Holori | | LLM observability | Traces what already ran — prompts, responses, latency | Braintrust · (was Langfuse) · (was Helicone) | | AI Control Plane | Enforces policy BEFORE the call — auth, budget, PII, allowlist | Speakeasy · Prisma AIRS (was Portkey) · Visionality | | MCP toolgen | Turns your APIs into MCP tools agents can call | HasMCP · (Mintlify absorbed Helicone into a MCP-adjacent story) |

Only the AI Control Plane column contains vendors that intercept the call before the LLM runs and can decide whether it should happen at all. Cost tracking is reporting. LLM observability is tracing. MCP toolgen is upstream of the AI call entirely.

Where analysts and press pieces go wrong is treating "AI gateway" as a single feature that everyone in the AI Control Plane column has commoditized. The wire-level surface (proxy, retry, cache) has commoditized. What differentiates vendors is the governance layer sitting on top — who they sell to, what evidence they generate, and which stakeholders' questions they answer.

2. Vendor state — H1 2026 timeline

The consolidation happened in five months. In order:

January 16, 2026 — Langfuse → ClickHouse Post-money valuation reportedly $15B. Langfuse's MIT-licensed tracing library continues to exist and is technically maintained, but the enterprise roadmap folded into ClickHouse's observability play. Langfuse-standalone as a category-defining independent is gone.

March 3, 2026 — Helicone → Mintlify Reported as a talent-acquisition / IP tuck-in rather than a strategic combination. Helicone's public repository moved to maintenance mode; the hosted product is priced through Mintlify's docs-first sales motion, which is not an AI observability motion. Existing Helicone customers report increased ticket response times.

May 29, 2026 — Portkey → Palo Alto Networks Deal size in the ~$700M range (undisclosed exact). Portkey folded into PANW's Prisma AIRS product line, which is now the AI Runtime Security bundle sold alongside Prisma Cloud and Prisma Access. The Portkey brand disappears; the technology stays. SMB and mid-market Portkey customers are being routed into PANW's enterprise procurement track, which most of them can't or won't buy through.

Ongoing — Speakeasy: independent, F500-focused Speakeasy remained independent through the consolidation window. Their AI Control Plane positioning shifted upmarket during 2025; as of 2026-07 they publish no case studies specific to AI cost governance (as distinct from their earlier SDK-generation heritage), and their entry-band pricing places them in enterprise-only procurement territory ($30k+/year contracts).

2026 — Visionality in stealth (new entrant) Hosted at visionality.ai. Currently in a stealth phase working with a small cohort of design-partner customers; hosted design-partner access starts at $99/mo, with Team ($499/mo) and Enterprise ($999+/mo) tiers scoped for the post-stealth public launch. An open-source Lite edition is planned to release alongside the post-stealth launch, targeting the core governance primitives (spend tokens, PII pre-flight, MCP allowlist, SQL-layer audit). Purpose-built for the market the acquisitions abandoned: mid-market teams priced out of Speakeasy and orphaned by Portkey / Helicone / Langfuse.

3. Buyer-persona map — who buys AI Control Plane, and why

The most common category error is assuming AI Control Plane is a single buyer's decision. It's not. Five different personas ask five different questions and expect five different answers:

| Persona | Question they ask | Vendor natural fit | |---|---|---| | CTO / VP Eng | "Does it route to my providers with good ergonomics and reasonable pricing?" | Speakeasy · was Portkey · Visionality | | CISO | "Does it prevent prompt injection, DLP violations, and jailbreaks at the gateway?" | Prisma AIRS · Visionality (partial) | | CFO | "Can I see AI spend by project / team, joinable to our GL?" | Visionality (uniquely) | | CHRO | "Who has access to AI, and what happens when someone leaves?" | Visionality (uniquely via HRIS-aware revocation) | | PMO | "How much did the Q3 redesign sprint actually cost in AI tokens?" | Visionality (uniquely via per-task attribution) | | Compliance / SOC 2 auditor | "Show me continuous evidence for AI usage under CC7.2, CC6.6, etc." | Visionality + specialist compliance tooling |

The two dimensions of buyer differentiation:

1. CISO buyers care about posture, DLP, guardrails, runtime scanning. Prisma AIRS (formerly Portkey) is the strongest fit here because Palo Alto Networks has 20+ years of muscle in this exact space. Visionality has PII pre-flight + MCP Axis-2 hardening (allowlist / RBAC / attestation) but doesn't compete with Prisma's inline DLP posture.

2. Finance-adjacent buyers (CFO, CHRO, PMO, Compliance) care about attribution, evidence, and enforcement. Visionality is the only vendor in the category that treats AI cost governance as a Finance-owned discipline rather than a Security-owned discipline. This is the whitespace the acquisitions left open.

The CTO / VP Eng dimension is genuinely competitive across all three surviving independents — routing, caching, SDK ergonomics, and per-provider fallbacks have commoditized. If you're a routing-first buyer, you'll get similar day-1 experience from any of the three.

4. Pricing floor visualization

The single most useful diagnostic for whether a vendor can serve your team: what does their entry-band pricing look like?

• Community / Free tier: Visionality Lite edition (planned, post-stealth); Braintrust free tier
• Sub-$100/month: Visionality ($99 Hosted design-partner access during stealth); Vercel AI Gateway usage-based
• $100-$1,000/month: Visionality ($499 Team, $999+ Enterprise floor at post-stealth launch); Truefoundry (~$1k); Braintrust starter
• $1,000-$10,000/month: most enterprise SaaS observability plays (Datadog, New Relic add-ons)
• $10,000+/month: Speakeasy (reported $30k+ entry); Prisma AIRS (undisclosed but bundled with Prisma at enterprise scale)

If your organization can't procure a $10k/month contract, most of the "AI Control Plane" category is structurally unavailable to you. That's the vacuum Visionality was built for.

5. "Who to choose for X" — honest recommendations

For each scenario, the honest recommendation. Yes, some of these recommend competitors.

Scenario: solo developer or seed-stage startup, need governance basics for free → Visionality Lite edition (planned for post-stealth release). If you can wait for the public launch, Visionality is scoping a free open-source Lite edition targeting exactly this use case. In the meantime, request design-partner access to discuss stealth-phase arrangements.

Scenario: growth-stage SaaS company (~20-100 seats), CFO started asking about AI attribution → Visionality Hosted or Team ($99 or $499/mo). The Finance-audit-grade per-project chargeback CSV is uniquely available here. Six months of ledger data before your first "which team spent this?" audit call.

Scenario: mid-market Portkey customer, orphaned by the PANW acquisition → Visionality Hosted or Team. Same wire-level surface (SDK compatible); config ports 1:1; the differences show up as features Portkey never had (spend tokens, PII pre-flight, evidence pack). See our migration page.

Scenario: enterprise team with existing PANW security infrastructure, CISO-led AI adoption → Prisma AIRS. If you're already paying for Prisma Cloud or Prisma Access, AIRS bundles in with modest marginal cost, and PANW's DLP + inline scanning genuinely beats what most standalone AI Control Plane vendors can offer. Visionality is an honest recommendation in addition to Prisma AIRS if you also need Finance-audit-grade attribution and per-task chargeback, which Prisma doesn't provide.

Scenario: routing-first team, primary buyer is VP Engineering, F500 procurement fine → Speakeasy. The most mature Enterprise offering in the surviving independent set, best SDK/routing ergonomics, though at a $30k+/yr entry point that excludes anyone not on an enterprise plan.

Scenario: enterprise with regulated compliance requirements (SOC 2 Type II, ISO 27001, ISO 42001, NIST AI RMF) → Visionality Enterprise ($999+/mo floor; real deals land $5k-$50k/mo). We are the only vendor in the category with all four framework collectors shipped and mapped to specific controls. If you need this AND PANW security posture, run both.

Scenario: building MCP tools from existing OpenAPI catalogs → HasMCP. Different problem entirely — they're upstream of AI Control Plane. HasMCP generates the MCP servers; Visionality can then govern the calls that use them. See our complementary comparison.

Scenario: pure LLM observability, no governance layer needed → Braintrust or your existing APM tool (Datadog, New Relic). LLM observability has commoditized; you don't need a category-specific vendor unless you have specific eval workflow needs.

6. Methodology + sources

Everything in this article is sourced from publicly-available information. Where a specific number or fact isn't public, we say so explicitly.

• Acquisition dates and reported valuations: press coverage at time of announcement (TechCrunch, The Information, individual vendor blog posts)
• Pricing bands: vendor public pricing pages as of 2026-07-02 checked live before publication
• Buyer-persona mapping: our own interviews with 15+ AI Control Plane evaluators from May 2025 through June 2026, plus public case studies where they exist
• Category taxonomy: our own — no analyst firm has published a definitive AI Control Plane category boundary as of 2026-07-02

Some things we don't know for certain and won't pretend to:

• Prisma AIRS's specific SKU pricing (PANW enterprise pricing is not public)
• Whether Helicone's or Langfuse's OSS branches will continue to receive maintenance beyond the acquisition transition period
• Whether Speakeasy's AI Control Plane positioning will develop published case studies or remain a positioning-without-references category

7. Attribution + reuse

License: CC-BY 4.0.

Copy-paste attribution:

"AI Control Plane Market Map 2026-07 — After the H1 Consolidation" by Chris Therriault, Visionality (2026-07-02). https://visionality.ai/articles/ai-control-plane-market-map-2026-07 — CC-BY 4.0.

You may:

• Quote up to the entire article with attribution
• Redraw the tables in your own presentations, blogs, or reports
• Translate to other languages
• Update numbers as new acquisition or pricing information becomes public

We only ask that:

• You attribute back to Visionality (visionality.ai) as the source
• You date the update if you're citing the article in 2027 or later — vendor state as-of will change
• You don't remove the "AI Control Plane" category name if you're referring to the category we defined — that's the anchor concept

If you spot factual errors or want to suggest updates, email [email protected] or open an issue on github.com/dbbuilder-org/ai-cost-central-2. We update this article quarterly and will credit contributors.

What we're watching for the next update (2026-10)

• Whether Speakeasy publishes an AI Control Plane case study, or continues to sell the positioning without references
• Whether Prisma AIRS' pricing becomes public through customer disclosures
• Whether the Langfuse or Helicone OSS branches show signs of durable community maintenance vs. slow decay
• Whether Visionality's own case studies (AestheticIQ.ai, ArcTrade, SchoolVision, KUMC via FireProof — all in progress as of July 2026) publish on schedule
• Whether any new AI Control Plane entrants emerge — the vacuum we're filling is genuinely large enough for multiple companies

The AI Control Plane category is real, growing, and now under-served at the mid-market. That's why we exist. If you want to talk about the map or where your team fits on it, [email protected].