TL;DR: On June 25, 2026, Alex Mashrabov (@alexmashrabov, CEO of Higgsfield) introduced Supercomputer 2.0 — Higgsfield's first autonomous marketing agent, accelerated by NVIDIA technology, with enterprise trust, safety, and permissioning. Inc. describes it as ads created, launched, and optimized without you — with a twist: agents do nothing until you grant credentials. A PSA Skincare campaign reportedly hit 29× views and 37× likes. Available on Team & Enterprise plans.
From media supercomputer to marketing supercomputer
In May 2026, we covered the original Higgsfield Supercomputer — a cloud-native stack pairing Seedance 2.0 video with the Hermes Agent for productions like Hell Grind (23-minute pilot in ~4 days).
Supercomputer 2.0 is the enterprise marketing chapter of the same brand:
| Supercomputer (v1) | Supercomputer 2.0 | |
|---|---|---|
| Focus | Long-form generative video / Original Series | End-to-end marketing campaigns |
| Stack | Hermes + Seedance 2.0 + 40 tools | NVIDIA Agent Toolkit + Nemotron subagents + 35+ models |
| Buyer | Creators, studios, viral pilots | Team & Enterprise marketing orgs |
| Trust model | Cloud execution, episodic memory | Permission-first credentials, policy guardrails |
| Launch window | May 14, 2026 | June 19–25, 2026 (Cannes / X wave) |
Same underlying bet: orchestration beats prompting. Different GTM: Fortune 500 CMOs, not only indie filmmakers.
What Alex Mashrabov announced on X
Mashrabov's June 25 thread (10K+ views in hours, reposted by Robert Scoble):
Introducing Higgsfield Supercomputer 2.0. Our first autonomous AI agent, accelerated by NVIDIA technology, to run your marketing end-to-end with enterprise-level trust, safety, and permissioning.
Follow-ups in the same thread:
- PSA Skincare (Sephora label) — 29× views, 37× likes on one Supercomputer campaign
- 390 Fortune 500 companies cited as relying on Higgsfield (aligns with ~78% claims in press)
- Supercomputer 2.0 on Team & Enterprise plans
- Link to Inc. explainer — "Read how the agent works"
Community reactions ranged from "Goodbye marketing agencies" to skepticism (@zcserei: "after taking away the buzzwords, I'm left with 'introducing our first run: run marketing'"). Both are fair — the product is agentic marketing ops under a dramatic name.
NVIDIA partnership: Agent Toolkit + Nemotron
Inc. and The Next Web report Supercomputer 2.0 is built on:
- NVIDIA Agent Toolkit — orchestration layer for enterprise agents
- Nemotron models — specialized subagents for continuous campaign reasoning
- Soul models on Blackwell — Higgsfield's proprietary creative models in the 35+ model mesh
- 20+ production pipelines — TV spots, product reels, AI podcasts (AlexTech summary)
Jamie Allan, NVIDIA Director of AdTech, Sports, and Enterprise Media, quoted in Inc.:
Agentic AI is becoming the operating layer for enterprise marketing… By building Supercomputer 2.0 using the NVIDIA Agent Toolkit, Higgsfield brings the efficiency of small, fast models to the work that runs continuously inside every campaign.
Translation for builders: this is not one giant LLM writing ad copy — it is many small agents doing continuous optimization loops inside a campaign lifecycle.
Enterprise trust: permission-first, not move-fast-break-glass
The "twist" in Inc.'s headline is governance:
Permission-first credentials
Agents do nothing until you explicitly grant access — ad accounts, brand assets, distribution channels. That addresses the #1 enterprise blocker on agentic marketing: "What if it posts without legal review?"
Policy guardrails
Every action screened for data leaks and off-brand outputs before execution (TNW).
Auditability (roadmap)
Compliance audit trails are planned, not fully shipped at launch — TNW flags this as a gap for regulated industries evaluating today.
For teams comparing Supercomputer 2.0 to Claude Code hooks or OpenAI agent SDKs, the pitch is marketing-native permissioning, not generic tool calling.
By the numbers (company claims — verify independently)
| Metric | Claim | Caveat |
|---|---|---|
| Fortune 500 adoption | ~390 companies (~78%) | Not independently audited (TNW, Inc.) |
| Global businesses | 12,000 across 6 continents | Company-reported |
| Revenue growth | Net revenue ~4× in first 5 months of 2026 | Company-reported |
| MoM growth | ~30% | Company-reported |
| Valuation | ~$1.3B | Press estimates |
| Total funding | ~$138M ($80M extension Jan 2026) | Crunchbase / press |
| Campaign speed | Concept → live video ads in under 15 min | Marketing claim |
| PSA Skincare | 29× views, 37× likes | Single campaign; Mashrabov X post |
Higgsfield was founded 2023 by Mashrabov — former Snap generative AI head, co-founder of AI Factory (Snapchat Lenses). Credibility in consumer-scale creative AI is real; Fortune 500 penetration needs third-party validation before you cite it in board decks.
What the agent actually runs (marketing lifecycle)
Supercomputer 2.0 orchestrates 35+ models across:
- Creative ideation — brief → concepts → scripts
- Asset production — image, audio, video (Seedance / Soul stack)
- Campaign assembly — formats for social, CTV, retail
- Launch & distribution — where permissions allow
- Optimization loops — continuous subagent reasoning on performance
Inc. frames it as create, launch, and optimize ads without you — with human approval gates on credentials rather than every pixel.
Context: the same week, New York State began requiring synthetic actor labeling on AI-generated commercial talent — enterprise marketing agents now operate inside tightening disclosure law, not a Wild West.
Pricing and access
Per Mashrabov's June 25 thread:
- Supercomputer 2.0 → Team & Enterprise plans (not consumer free tier)
- API → not announced in the launch thread; enterprise sales likely required (@philkyprianou asked on X)
Consumer creators still use Higgsfield's video GenAI surface; 2.0 is the autonomous marketing ops upsell.
Who wins and who should wait
Strong fit:
- Enterprise marketing teams already on Higgsfield for video GenAI
- Brands running high-volume social creative (Sephora-tier retail)
- Orgs that can tolerate roadmap auditability with strict permission gates today
Wait or stay skeptical:
- Teams needing verified Fortune 500 case studies (get references, not press claims)
- Regulated industries requiring full audit trails on day one
- Agencies whose value is strategy and relationships, not asset throughput alone — @DonioVEditor's "goodbye agencies" take overstates near-term displacement
Agentic marketing vs agentic coding
Supercomputer 2.0 is the marketing vertical of the same macro trend as Claude Code and Cursor agents — but with brand safety as the primary constraint.
Eric Xing's agentic vs agentive framing applies here: Supercomputer 2.0 is deeply agentic (35-model harness, NVIDIA toolkit, permission scaffolding). Whether any "agency" is internalized inside the model versus engineered in the platform is an open research question — buyers should care about outputs and controls, not vocabulary.
Timeline
| Date | Event |
|---|---|
| May 14, 2026 | Original Higgsfield Supercomputer architecture |
| Jun 18–19, 2026 | Supercomputer 2.0 press cycle; Cannes Lions presentations |
| Jun 25, 2026 | @alexmashrabov viral X launch + PSA Skincare stats |
| Ongoing | Auditability layer on roadmap |
Related reading
- Higgsfield Supercomputer architecture (v1)
- Hell Grind — Seedance 2.0 Original Series
- What is AI slop? — quality vs volume in GenAI
- AI agents for influencer marketing automation
- Build AI marketing agents with Claude — tutorial
Primary sources: Alex Mashrabov on X · Inc. — Supercomputer 2.0 · The Next Web · Higgsfield on X
Fortune 500, revenue, and PSA Skincare metrics are company-reported unless independently verified. Product availability and audit features may change — confirm on higgsfield.ai before procurement.