Financial Reports: Anthropic vs. OpenAI

Summary of Financial Reports: Anthropic vs. OpenAI

by Candace Fan

14m•April 20, 2026
Listen to Episode

Overview of Financial Reports: Anthropic vs. OpenAI

Host Candace Fan breaks down major recent shifts across the AI ecosystem: a surprising rebound in app launches driven by AI tools, OpenAI’s shutdown of its Sora video product and simultaneous executive departures, Cerebras’ IPO filing and claims of taking inference business from NVIDIA, Stanford’s new AI Index findings (notably US vs China model parity and a big jump in agent capabilities), and the startling news that Anthropic’s annualized revenue run-rate has passed OpenAI’s — while using far less compute.

Key topics covered

  • App Store and Play Store activity: app launches rising sharply (AI-driven “vibe coding” enabling non-developers).
  • OpenAI: Sora shutdown, executive exits, likely strategic pivot to coding & enterprise.
  • Cerebras: IPO filing, large compute deals (OpenAI, AWS), CEO claim about stealing inference work from NVIDIA.
  • Stanford AI Index: US lead over Chinese models has narrowed to ~2.7%; AI agents’ real-world task success jumped from ~12% to ~66%.
  • Anthropic vs OpenAI: Anthropic reported a higher annualized revenue run-rate ($30B vs $25B) and superior unit economics (less training compute), with strong enterprise traction.

Data & metrics (highlights)

  • App launches: +60% YoY in Q1 2026 globally; iOS app releases up ~80%; April releases up ~104% year-over-year (so far).
  • Sora: peaked ~1M users, later <500k; app to close late April, API to shut down in September; reported massive daily compute burn (estimates mentioned between $1M–$15M/day).
  • Cerebras (recent financials & fundraising): $510M revenue in 2025, ~$237M net income (noted possibly with one-time items); Series H valuation ~$23B; reportedly a >$10B deal with OpenAI; AWS partnership; mid-May IPO target.
  • Stanford AI Index: US top models lead Chinese top models by ~2.7% (gap collapsed from ~17–32 points in 2023).
  • AI agents: success on real computer tasks rose from ~12% a year ago to ~66% now.
  • Anthropic vs OpenAI revenue: Anthropic run-rate ~$30B; OpenAI run-rate ~$25B. Anthropic growth: ~$9B (end 2025) → $20B (early March) → $30B (early April).
  • Customer mix & compute: ~80% of Anthropic revenue from businesses; >1,000 customers spending >$1M/year (up from 500 in Feb); Anthropic reportedly spends ~4× less on training than OpenAI; Anthropic has ~1 GW of Google compute for 2026 and ~3.5 GW next-gen TPU capacity starting 2027.

Company breakdowns

OpenAI

  • Action: Shutting down Sora (text-to-video) late April; API closure scheduled for September.
  • Internal turbulence: product chief, head of Sora, and enterprise CTO departed in a short window.
  • Strategic shift: reported move to prioritize coding and enterprise offerings; shedding “side quests.”
  • Scale: still consumer-heavy with ~900M+ weekly active ChatGPT users (large distribution moat).

Anthropic

  • Financials: reported $30B annualized run-rate — now ahead of OpenAI.
  • Growth: explosive run in early 2026 (9→20→30B in months).
  • Unit economics: claims of much lower training compute spend (≈4× less) while taking more revenue — implying better margins/efficiency.
  • Customer base: enterprise-heavy (~80%), rapid increase in large enterprise customers (500→1,000+ spending >$1M/year).
  • Infrastructure: major compute commitments with Google (1 GW in 2026; 3.5 GW TPUs in 2027) and Broadcom.

Cerebras

  • Positioning: AI chip maker pitching itself as high-performance training/inference alternative to NVIDIA.
  • Corporate moves: previously pulled an IPO in 2024 over an investment review; returned with big funding rounds and is pursuing an IPO.
  • Claims & partnerships: CEO asserts Cerebras took fast inference business from NVIDIA at OpenAI; reports of a >$10B OpenAI deal and AWS integration.
  • Why it matters: any material shift of OpenAI inference workloads off NVIDIA is the first crack in NVIDIA’s frontier-market dominance.

Market implications & interpretation

  • App ecosystem is not dead — AI tools (low-code/no-code & agent builders) are accelerating app creation rather than replacing apps.
  • Enterprise demand and utility-focused AI (agents, coding assistants) are driving real dollars — Anthropic’s enterprise-first approach appears to be paying off.
  • Efficiency matters: Anthropic’s ability to grow revenue while spending significantly less on training suggests important differences in model and infra efficiency that could reshape economics at the frontier.
  • Hardware competition: Cerebras’ wins could loosen NVIDIA’s dominance if partnerships and claims hold up — watch compute-vendor dynamics closely.
  • China vs US: performance parity is tightening despite a large investment gap — a strategic concern for US policymakers and companies.

Notable quotes / insights

  • “The rumors of the Apple Store’s death have been greatly exaggerated.” — Greg Joswiak (on app growth).
  • Cerebras CEO (Andrew Feldman, quoted): “NVIDIA didn’t want to lose the fast inference business at OpenAI and we took that from them.”
  • Host’s practical point: agents have gone from experimental to useful — try giving an agent bounded real tasks and verify outputs.

Practical takeaways & recommended actions

  • If you build or manage workflows, re-test AI agents now — agent reliability and usefulness have improved markedly.
  • Enterprises should evaluate Anthropic’s offerings for coding, automation, and agent workflows (appears to be enterprise-leading today).
  • Developers and non-technical founders: the rise in app launches suggests a golden moment to ship via low-code/no-code tools and “vibe coding.”
  • Investors and infra teams: monitor Cerebras’ IPO, OpenAI’s compute choices, and large announced compute commitments (Anthropic + Google TPUs).
  • Consumers: expect more AI-powered productivity and utility apps — the app ecosystem appears to be expanding, not contracting.

Closing / host notes

  • Candace emphasizes the practical value she’s derived using Anthropic agents and automations (scheduling, scraping, dashboards) and encourages listeners to experiment with agents and low-code AI tooling. She also mentions her product, AIbox.ai, as an automation platform (80+ models, $8.99/month) and a community/course called AI Hustle School for learning “vibe coding” and growth strategies.