Erica · PM Research MemoAll memos

PM RESEARCH MEMO

Title: Agent Traffic Crossed Human (May 2026) — Ads Break; Micropayments, Isolates, and Cloudflare as Machines-Internet Tollbooth
Author / source: Matthew Prince (CEO, Cloudflare), guest on Empire (@empirepod / Blockworks) with host Jason Yanowitz
Source title: Cloudflare CEO: The Man Deciding The Internet’s Next Business Model
Source URL: https://www.youtube.com/watch?v=f6GkamBBNbs
Video ID: f6GkamBBNbs
Channel: Empire (@empirepod / Blockworks) — not Forward Guidance (yt-dlp metadata confirms Empire)
Published: 2026-09-07 (yt-dlp upload_date=20260907; ~06:00 AM ET / America/Toronto)
Duration: 1:11:26 (~4,286 seconds)
Memo date: Tuesday, September 8, 2026 (America/Toronto)
Transcript: /workspace/youtube-transcripts/f6GkamBBNbs.md (~15,046 ASR words) · Brief: /workspace/youtube-transcripts/f6GkamBBNbs_brief.md
Caption source: YouTube English automatic captions (ASR) only. No official/manual English track. Name/number locks from brief applied below.
Source type: Infra / AI × internet business-model interview. Channel disclaimer at open: not a recommendation to buy or sell. No ticker pitch, price target, or portfolio expression from Prince.
Product: Research map of Cloudflare strategic positioning and internet monetization shift. Not advice. No buy/sell from this desk.
Source discipline: Primary source is this transcript only. Traffic multiples, TPS figures, revenue mix, and spend guesses are Prince/Cloudflare claims via ASR — flag for external verify before treating as fact. Companion memos on this desk are different sources — do not silently merge numbers.

ASR name / number locks (from brief):
| ASR heard | Likely intended |
|---|---|
| Cladflare / Cloudflow / Kle / Oak Clafler | Cloudflare |
| Bailey Gford | Baillie Gifford |
| Beni off / cloud force | Marc Benioff / Salesforce Agentforce |
| codeex | Codex (OpenAI) |
| replet / base 44 | Replit / Base44 |
| Okami (COVID call) | Akamai |
| Net net dollar | Net Dollar (Cloudflare stablecoin concept/name) |
| div force | divorce |
| feeasants | pheasants (“flush the pheasants”) |
| species specist | speciesist |
| halves and halves not | haves and have-nots |
| thatic chat | that AI chat |
| Piracy / privacy shield mix | Italy Piracy Shield |
| Hyperlid (ad read) | Hyperliquid |
| Larry Paige | Larry Page |


EXECUTIVE SUMMARY


SOURCE-ACCURATE SUMMARY

Chapter-ordered with approximate timestamps. Quotes ≤20 words where useful. Source = stated on tape; Inference labeled when used.

Cold open & internet evolution (0:00–5:33)

  1. (0:00) Disclaimer / setup. Empire disclaimer: nothing said is a recommendation to buy or sell. Host Jason Yanowitz; guest Matthew Prince, Cloudflare CEO (~16–17 years running the company). Prince in Park City, Utah.

  2. (0:00–1:51) Early-web vantage. College ~1993: helped create one of the first online-only news publications; Wired award; “first thousand” public sites era. Today: hundreds of millions of sites; remote production that once cost “tens if not hundreds of thousands” now cheap.

  3. (1:51–3:43) Ads as only scaled micropayments; Google as hero. Last ~20 years driven by advertising. High CPM for high-income users; even a “kid in Sri Lanka” still gets something. Ads = only micropayment system at real scale. Google = “hero of the last really 27 28 years” — search plus Analytics plus DoubleClick stack as economic engine of the web.

  4. (3:43–5:33) Agent traffic crossover timeline. Web Summit Lisbon (Nov): forecast automated > human in 2H 2027. SXSW Mar 2026: pulled to 1H 2027. May 2026: team said automated already passed human. Caveat: “wrong in every prediction so far.” If trends continue: agentic traffic “a thousand times as much as human traffic in five years.” Humans don’t decrease use; bots explode → humans ≈ rounding error. Core problem: “bots don't click on ads.” Business model change has implications for media, society, polarization. “Most interesting question… what's the future internet business model.”

Centralization, societal critique, labs vs Google (5:33–16:59)

  1. (5:33–7:24) Centralization vs decentralization. Host: open web → AOL → social + Google centralization; Instagram/Facebook ads as business substrate. Prince: some parts centralized, some still decentralized. Early investor asked biggest competitor → Prince said Facebook (looked crazy): Facebook made owning a website hard (security/performance), so people gave up; Nigeria as example where internet ≈ Facebook. Cloudflare : Facebook :: Shopify : Amazon — make independent sites easier.

  2. (7:24–9:12) Web growth re-acceleration via AI. ~2012 through early 2025: web growth “minuscule”; for a period more sites taken offline than created. Late 2025: AI let non-developers create sites/apps → growth “closer to… early 2000s.” Provocation: “everything that's wrong with our world today… is Google's fault” — immediately softened as overstatement; Google “amazing company.”

  3. (9:12–11:04) Attention → dopamine/cortisol → polarization. Ad model = capture attention, monetize it; begat TikTok-class systems. Best attention capture: dopamine or cortisol — “irrationally happy or irrationally angry.” Optimistic alternative: reward knowledge creation instead of fight-or-flight.

  4. (11:04–14:11) Labs’ fear of Google; data asymmetry. All major labs share “incredible fear of Google.” Reported OpenAI origin: dinner Sam / Elon / Larry Page; Page allegedly “speciesist” if AI out-evolves humans. Google advantages: DeepMind talent, custom TPUs/silicon, “twice as much of the internet as open AI” (and more vs Anthropic / X / Meta). Claims Google “screwed it up so far” but “would… not count them out.” ~80% of major AI companies use Cloudflare; customers include OpenAI, Anthropic, X/Grok, Chinese labs. Policy aim: Google pays like everyone else — “shouldn't be able to leverage their monopoly of yesterday… monopoly tomorrow.” Want 500,000 AI companies, not 1 or 5.

  5. (14:11–16:59) Open weights + paid crawl; tragedy of the commons. Signed letter against regulating open-weight models; still wants creators compensated — not a contradiction. At 1000× agent traffic, who pays? Efficiency work (e.g. rewrite WordPress) + free tier help, but real bandwidth/server costs remain. Free scouring → bots crawl every page then recommend Wendy’s. Need model where beneficiaries of scanning compensate infrastructure and original content creators.

Cloudflare scale, isolates, AI adoption timeline (16:59–27:19)

  1. (16:59–20:41) State of the union. Network in 350+ cities, milliseconds of most population; security (cyber/DDoS) + performance + user protection; developer platform since ~2017. Vibe-coding platforms (Lovable, Replit, Base44, Codex, Anthropic tools) prefer Cloudflare runtime when agents build software. “A lot more than 20% of the internet sits behind Cloudflare”; “almost half of the Fortune 500”; major crypto companies building on Cloudflare developer platform. (Midroll: Token2049 Singapore Oct 7–8 + DAS Asia; Avalanche Summit NYC Sep 16–17; promo BLOCKWORKS15 — sponsored, not thesis.)

  2. (20:41–24:13) AI lean-in chronology. Developer platform started 2017 (pre-ChatGPT). Early VC pitch calling Cloudflare an “AI company” got laughed out of the room; used ML on traffic for security/performance. Happy selling picks and shovels; nervous about ROI and shipping internal data to external AI vendors → “slow adopters.” Mid-2025: “blindingly obvious”; turning point November 2025 — restructure how business is organized around AI. Product architecture “dumb luck” right; org rearchitecture was the deliberate move (late Oct / early Nov 2025 around Opus 4.6 era per host).

  3. (24:13–27:19) Isolates vs containers. Hyperscalers (AWS, Google Cloud, Azure) built for mobile/container era. Evolution: physical servers → VMs → Docker containers. Problem: if every knowledge worker has ≥1 agent in containers → need “40 times the number of CPUs that are produced in the world” just for those agents. Beyond containers = isolates (browser-tab model) — lightweight, fast deploy; “exactly what agents need.” Many agents already on Cloudflare isolates or generating code that runs in isolates.

Org redesign, layoffs, public markets (27:19–44:16)

  1. (27:19–28:50) WSJ op-ed framing. Host recounts: monster Q1; late Apr / early May announcement; then ~1,000–1,100 people (~~20%) cut. Builders / sellers / measurers frame — AI does audits, finance, legal, compliance, middle management.

  2. (28:50–32:43) What AI is / isn’t good at. Not “everyone loses jobs.” Developers much more productive → hire as many developers as possible (even if 10× productive, still more work). Sellers: pre-AI ~40% time with customers / ~60% presentations + Salesforce; now ~90% with customers. Measuring: humans have shared team biases; AI biases “uncorrelated” → fresher outside eyes. Harvard-style span ~6 direct reports → can do 12 with tools → org flattens (larger average span = flatter). Cut middle managers; helped place them; “great severance” and continued vesting. Emotional; dear friends let go.

  3. (32:43–36:01) Peer dynamics & kindness-of-speed. Peers at other large publics: “we got to do that too” but fear stock reaction (Cloudflare dipped then recovered); Prince got “death threats.” Leadership: once you know the change is needed, delay is “disservice to the employees… eventually going to lay off” — kinder when labor market still absorbing (early-year easier than later). Open-sourced harness as Cloudflare OS. From position of strength vs peers restructuring from threat.

  4. (36:01–39:07) IC vs manager blur; intern zag. Many managers want to return to IC; senior people going to Anthropic wanting IC roles. Hiring 1,111 interns over 2026 — AI-native; reverse-mentor seniors (“train us”). Not doomer; accounting-degree nervousness; builders + sellers = “most fun parts”; enterprise $100M deals still human; agent-to-agent commerce emerging. Explicitly not in Dario / lab “take everyone's jobs” camp — “refocus how we run businesses.”

  5. (39:07–42:48) Public markets preference. Prefers public: PE/VC lack “no-fault divorce” → nobody tells truth. Public investors call after stupid podcast comments — honest feedback. Counter to short-termism myth: Baillie Gifford ~16-year average hold vs VC fund life 10–12 years. Earnings calls: strategy questions dominate; CFO gets few questions. Host jab: Benioff “Agentforce” as marketing for public markets as much as customers.

  6. (42:48–44:16) Most misunderstood. Customers still see “security CDN” — “such a small part.” Investors once thought too SMB; SMB “less than… 10% of revenue today” — mostly large enterprise.

Machines-internet stack & micropayments TPS (44:16–55:54)

  1. (44:16–45:33) Full-stack vision / Dark Knight. Host: tollbooth (pay-per-crawl / pay-per-use), wallet, browser, identity, Net Dollar. Prince: Dark Knight / Joker “everyone thinks they have a plan” empathy — no grand unifying plan; build what internal teams need; respond to tragedy-of-commons agent crawl incentives by flipping so agent pays.

  2. (45:33–48:49) Scale math. ~0.5 billion web transactions/sec published; 1–10% monetizable → ~10M financial TPS day-one; need path to ~100M TPS. Visa peak ~80k TPS (three orders short); equities busiest day ~2M TPS. Crypto space “shooting to be the next Visa”; opportunity is “white space that Visa can't even handle.” Host’s 2018 Blockworks blog predicted browser crypto-wallet micropayments — reminder things take longer.

  3. (48:49–52:33) Crypto role; Net Dollar as bait; build vs buy. Role for blockchains getting funds in and out. Consensus speed question. Wallet + Net Dollar as name/idea to “flush the pheasants” — force partners to come talk. Largest potential partners ~2M TPS (hybrid/batching) — still 1–2 orders short of need. Build team hiring “like crazy” and happy to partner/acquire if solution appears. Not religious about blockchain for high-velocity core.

  4. (52:33–55:54) Hybrid settlement; universal translator. High-velocity processing = hard CS problem; play to Cloudflare strengths (commodity servers worldwide, precise time sync to avoid double-spend) vs beefy centralized servers. May use internal ledgers; chains useful across orgs but consensus slow → likely hybrid. Partner PayPal / Visa / Amex / banks + crypto rails — “universal translator.” Interest: content/app customers thrive. Crypto ecosystem sometimes “religious”; Cloudflare “not very religious.”

Regulation, Italy, capacity, inequality, interface (55:54–1:11:26)

  1. (55:54–59:03) Money/regulation adjacency. Neutral plumbing today; money is regulated/custodial/adversarial. Already deep in content regulation (Spanish copyright, Indonesian politics). Don’t need to hold money — help websites and agents transact. Regulators constructive so far because framing is “pay creators,” not “be a bank.” Naive about future fraud/problems at high volume of tiny txs. Historical content moderation: Daily Stormer, 8chan, Kiwi Farms — comply with regional court orders.

  2. (59:03–1:00:32) Italy Piracy Shield. Did not sign up. Scheme: media companies put sites on list; global takedown; no judicial oversight. Attempted block of US State Department site. Would comply with Italy-only court order; push back on unelected cabal blocking everywhere. White-collar joke: sentence could be “stay in Italy” / Tuscany.

  3. (1:00:32–1:03:23) COVID 2× vs agent 1000×. COVID: traffic doubled in two weeks; almost collapsed at European exchange points. Call with Netflix, Amazon, YouTube, Akamai — “Mexican standoff”; forced bitrate cuts (no 4K). Ted Stevens “series of tubes” had finite capacity. Agents may not stream video much today, but without efficiency or payment, break risk. World “hasn’t woken up.”

  4. (1:03:23–1:07:04) New-entrant / consolidation worry; spend split. Agents ignore convenience/emotion; trust data-rich incumbents → consolidation pressure; hard for new small businesses. Prince wants 500,000 AI companies and room for new entrants — no clear solution yet. Spend guesses: self ~$1M/yr on agents; average US consumer ~$1,200/yr from entertainment displacement. Anthropic ads critique: ads with agents feel creepy; users switch if agent corrupted. Google ~$750/US user/yr; Meta ~$250 — fight for $1,200. Sri Lanka kid / haves-and-have-nots: ad-subsidized free web may vanish.

  5. (1:07:04–1:11:26) Interface shift; fame vs money. Host bear case: Chat/Anthropic become the internet. Prince: 14-year-old’s internet is TikTok; dad (85) lives in chat. Prior shifts (web → social → mobile) kept ads as model. Next interface = AI chat/agent — Jetsons / “Rosie the helpful robot.” Can path money to creators; fame harder when consumption is AI “cliffnotes.” Met Spotify’s Daniel Ek: musicians motivated more by fame than riches (irrationally). Proposal: LLM-era Nobel/Academy awards measuring who contributed most to knowledge domains — recognition to keep feeding the systems.


SYSTEMS MAP

[Human browsers]──ads/CPM──►[Publishers / creators]
       │                            ▲
       │                            │ pay-per-crawl / pay-per-use (proposed)
[Agent / bot traffic]──►[Crawl / use requests]──►[Content + infra costs]
       │                         │
       │                    (today: often free / robots.txt weak)
       ▼                         ▼
[Cloudflare edge: 350+ cities | >20% internet | ~½ F500]
       │
       ├── Security / CDN / performance (legacy perception)
       ├── Workers / Isolates runtime (agent & vibe-code preferred)
       ├── Identity / browser / wallet (building)
       ├── Tollbooth: pay-per-crawl / monetization gateway
       └── Settlement: hybrid high-TPS ledger + crypto in/out
              │                    │
              │              [Net Dollar name/idea — partner bait]
              ▼
     Need ~10M → ~100M financial TPS
     vs Visa ~80k | equities ~2M | best crypto hybrid ~2M

[Google crawl asymmetry ~2× OpenAI view]
       └── Level-field push: same pay rules; many AI labs (target “500k”)

[Org system]
  Builders ──AI tools──► hire more / higher output
  Sellers ──AI admin──► ~40%→~90% customer time
  Measurers ──AI audits──► shrink middle layers; span 6→12

[Interface stack]
  Web → Social → Mobile (ads constant) → AI agent interface (ads break)

Source nodes: traffic crossover, ad break, micropayment TPS, isolates, org triad, Google data asymmetry, COVID capacity analog, equity/access worry — all from Prince on tape.
Inference: Cloudflare’s economic moat thesis = control of (or partnership around) the payment + identity + isolate runtime layer when human-ad economics no longer clear.


SECOND / THIRD-ORDER EFFECTS

Chain 1 — Traffic mix → ad break → creator funding crisis → micropayment rails

Chain 2 — Container-era CapEx assumptions → isolate/runtime re-architecture

Chain 3 — Org flatten → labor market & software spend mix

Chain 4 — Level playing field crawl → AI competitive structure

Chain 5 — Capacity stress (COVID 2× analog) → efficiency + payment or breakage


SCENARIOS

Bull (machines-internet rails clear; Cloudflare central)

Base (partial monetization; hybrid messy; gradual org diffusion)

Bear (ad model limps on; rails don’t scale; backlash)


WATCHLIST

Item Why it matters Source hook
Cloudflare published bot vs human traffic series Validates May 2026 crossover & 1000× path 3:43–5:33
Pay-per-crawl / AI gateway product metrics Tollbooth thesis live or vapor 44:16–47:24
Net Dollar / wallet status (product vs name) Settlement stack seriousness 48:49–52:33
Partner or acquire high-TPS settlement tech Build-vs-buy resolution 50:40–54:04
Workers / isolate usage & vibe-platform design-wins Runtime moat vs hyperscalers 18:51–27:19
Earnings: enterprise vs SMB mix; strategy Q&A Misunderstood CDN vs platform 42:48–44:16
Peer public AI restructuring / layoff waves Org thesis diffusion 32:43–36:01
Google AI crawl / unbundle / paid-access disputes Level playing field 11:04–14:11
Italy Piracy Shield / extraterritorial takedown cases Content-regulation risk template 59:03–1:00:32
IXP / CDN capacity stress commentary COVID 2× → agent 1000× analog 1:00:32–1:03:23
Publisher / creator micropayment experiments Demand side of rails 14:11–16:59
Intern / AI-native hiring trends at large tech Labor zag 36:01–39:07

DILIGENCE QUESTIONS (5–10)

  1. Traffic audit: Can Cloudflare’s May 2026 “automated > human” claim be reconciled with public blog methodology (definitions of automated, crawlers vs agents, customer mix bias)? External check needed.
  2. Monetizable %: What evidence supports 1–10% of 0.5B tx/sec becoming paid microtransactions rather than blocked/cached/free?
  3. TPS roadmap: Concrete architecture for 10M→100M financial TPS on commodity global servers — latency, double-spend, dispute, KYC/AML at micropayment dust sizes?
  4. Net Dollar: Legal/status — named concept, partner vehicle, or Cloudflare-issued? Custody model if “don’t need to hold the money”?
  5. Isolate cost curve: Independent benchmark of isolate vs container $/agent-hour and cold-start; do Lovable/Replit/Base44/Codex preferences show in contracts or only anecdotes?
  6. Google parity: Any signed agreements or technical controls that actually bill Google crawlers equivalently to other labs?
  7. Layoff outcomes: Retention of builders/sellers post-cut; measured span-of-control change; stock/talent reaction vs peers who waited.
  8. Revenue concentration: Confirm SMB <10% and “almost half of Fortune 500” with public disclosures; crypto/AI developer revenue share.
  9. Regulatory perimeter: How wallet/identity/micropayments interact with money-transmitter, sanctions, and fraud regimes across jurisdictions already litigating content (Italy, Spain, etc.)?
  10. Access equity: Any Cloudflare or industry proposal that preserves free or subsidized access for low-income users if ads die — or is this acknowledged open gap only?

RISK ANALYSIS

Risk Mechanism (from source) Severity if wrong Mitigant / open
Traffic thesis error Prince wrong “in every prediction so far”; crossover timing already moved twice High for urgency of micropayment build Continuous Cloudflare data; still directionally rising bots
TPS unsolved Best partners ~2M vs need 10M–100M High — rails don’t clear; tragedy of commons persists Hybrid ledger; build in parallel; partner/acquire
Religious crypto mismatch L1 narrative = “next Visa” not micropayment white space Medium — wrong partner set, wasted cycles Explicit non-religious stance; crypto for in/out only
Regulatory / money adjacency Fraud at high volume of tiny txs; custody expectations Medium-High “Don’t hold money”; constructive regulator talks so far (claim)
Content politics (Italy-class) Extraterritorial blacklist without due process Medium — brand/legal distraction Comply with territorial court orders; refuse global cabal
Walled-garden bear Chat becomes the internet; open web hollows High for tollbooth on open web Level-field crawl; many AI companies thesis
New-entrant / inequality Agents favor data-rich incumbents; paid access locks out poor users High societal; medium direct to Cloudflare near-term Prince: no good solution yet — watch political backlash
Isolate advantage erodes Hyperscalers ship equivalent light runtimes; container efficiency improves Medium 2017 head start; design-wins with vibe platforms
Org-cut talent brand Death threats / peer fear; stock dip optics Low-Medium if execution strong Placement help; severance; hire builders/interns
ASR / claim unverifiability Auto-captions; round numbers; personal spend guesses Process risk for memo users Treat as source claims; diligence list above

Explicit non-risks for this memo: No valuation, multiple, or buy/sell conclusion. Cloudflare (NET) discussed operationally only.


APPENDIX — KEY NUMBERS (SOURCE CLAIMS; VERIFY)

Claim Approx. tape locus Status
Automated > human traffic May 2026 (team to Prince) Unverified externally
Prior forecasts 2H27 → 1H27 Web Summit; SXSW Mar 2026 Source narrative
Agentic ~1000× human in 5 years Forward look + caveat Unverified; Prince caveat
~0.5B web tx/sec Published Cloudflare figure (as stated) Verify vs blog/earnings
1–10% monetizable → ~10M TPS; path ~100M Internal estimate Unverified
Visa peak ~80k TPS Compare Spot-check
Equities busiest day ~2M TPS Compare Spot-check
Crypto hybrid partners ~2M TPS Partner survey claim Unverified
>20% of internet behind Cloudflare Scale claim Verify
~Half of Fortune 500 Scale claim Verify
350+ cities Network claim Verify
~80% major AI cos use Cloudflare Relationship claim Unverified
Google sees ~2× OpenAI internet; more vs Anthropic/X/Meta Data asymmetry claim Unverified
Layoffs ~1,000–1,100 (~20%) post strong Q1 Apr/May window (host/Prince) Verify vs 8-K/WSJ
Sellers 40%→90% customer time; span 6→12 Org metrics Internal claim
1,111 interns in 2026 Hiring claim Verify
SMB <~10% revenue Mix claim Verify vs filings
Prince ~$1M/yr agents; US consumer ~$1,200/yr Personal guesses Speculative
Google ~$750 / Meta ~$250 per US user/yr ads Compare Spot-check
COVID traffic ~2× in ~2 weeks Capacity anecdote Historical claim

MEMO DISCIPLINE CLOSE

This memo is a source-faithful research extract of the 2026-09-07 Empire × Matthew Prince interview. All strategic theses (agent traffic crossover, ad-model break, micropayment tollbooth, isolate runtime, builders/sellers/measurers, Google level field, capacity and equity worries) are Prince’s framing on tape, not desk endorsements. ASR-only captions; locks applied from the companion brief. No buy/sell. No price target. Hypotheses and watchlist only.

End of memo.

Desk copy · not a trade recommendation · Erica · 8 Sep 2026 · HIGH batch