Who he is
Affiliation: Founder & Managing Partner, Link Ventures; General Partner of Link-XPV, the AI seed fund he co-leads with Peter Diamandis (KKCG LP investment announced 2025-12-16); founder of multiple AI startups (EverQuote co-founder; Cyberian Outpost / Cogito background [TBD: precise founding history]); MIT-trained AI researcher; co-host, Moonshots podcast (with Diamandis, Ismail, Wissner-Gross).
One-line position: Scaling laws are the dominant force in AI; agents are the deployment unlock; most enterprises are dramatically underestimating the speed and scope of disruption.
Discipline & technical bet
MIT AI background gives him more technical depth than the typical VC commentator; the operating mode is deployment economics, not research. The concrete wager now: a 100x jump in largest-model scale via quantization breakthroughs and Chinese research in 2026 (his named Moonshots 2026 prediction), plus the claim that recursive self-improvement needs faster inference and better chips, not Einstein-level AI (Moonshots, Jun 2026). Both are unusually falsifiable for him — hold him to them.
Key claims (Says)
- Empirical: scaling laws (compute × data × algorithmic gains) continue to compound; no capability plateau through 2025–26. Held — frontier cadence unbroken through mid-2026 (new top-tier model lines shipped); he doubled down with the 100x-scale prediction for 2026.
- Empirical: agents (autonomous multi-step task execution) are the deployment unlock — copilot → autopilot → autonomous. Held — directionally consistent with 2025–26 enterprise agent rollouts; specifics vary by vertical.
- Empirical: most software categories will be rewritten by AI-native entrants within 3–5 years (by ~2028); vertical SaaS especially exposed. Live prediction — early signs in coding, support, legal; grade at 2028.
- Empirical: enterprise underestimates speed (benchmarking current model quality, not 12-month-ahead trajectory). Drifted — his own June 2026 position is now "legacy enterprises are adopting AI faster than expected" (Moonshots summary, 2026-06-06); the wait-and-see foil he was reacting against is dissolving, by his own account.
- Empirical: AI-native delivery economics (marginal cost ~zero per inference) restructure SaaS gross margins and pricing. Held — consistent with continuing API price deflation; he now frames it as "40x deflation" (Moonshots episode, Aug 2026).
- Mixed: VC should concentrate on agent platforms, vertical AI applications, and infrastructure picks-and-shovels. Held — and enacted: Link-XPV raised institutional money (KKCG) to do exactly this at seed stage, ~75% from the MIT/Harvard ecosystem.
- Empirical: the bottleneck shifts from model quality to integration, eval, and trust infrastructure as agents take on autonomy. Held — his own 99.9%-accurate-financial-advice example (can't deploy in the US "for fear of lawsuits over the 0.1%") is this claim in anecdote form.
- Normative: founders should target replacement, not augmentation, in defensible verticals.
- NEW (Jun 2026): only three countries (US, France, Argentina) are fully self-sufficient in the energy and materials needed for large-scale AI deployment; Argentina could serve as a regulatory-arbitrage deployment zone.
- NEW (Dec 2025, Moonshots 2026 predictions): a new industry acronym emerges in 2026 and mints multiple billionaires, especially young founders ("nearly certain," 95%).
Notable predictions — with falsifiable checks
- (2024) Most software categories rewritten by AI-native entrants by ~2028. Check: by 2028-12-31, have ≥3 major vertical-SaaS categories seen an AI-native entrant take #1 or #2 revenue share? Early signs (coding, support, legal) logged; not yet decided.
- (2024) Agents mainstream in enterprise within 12–24 months (by 2025–26). Drifted — agent products shipped widely but enterprise-wide autonomous deployment remains uneven; "mainstream" held only under a generous reading. Re-check with 2026 enterprise adoption surveys at next review.
- (Ongoing) Scaling laws hold through 2026; no plateau. Held so far — 2026 frontier releases consistent.
- (Multiple, Moonshots) Inference cost drops ~10x/year on equivalent capability. Directionally correct on API pricing curves; benchmark-dependent — he now quotes cumulative "40x deflation." Check: like-for-like frontier-class token pricing, next review.
- (2025-12, for 2026) 100x increase in largest AI models via quantization + Chinese research. Check: by 2026-12-31, does any credible disclosure show ~100x effective-scale jump vs 2025's largest? (External assessment at prediction time: 60% for major scaling, 20% for literal 100x.)
- (2025-12, for 2026) New acronym mints multiple billionaires. Check: identifiable 2026-coined category label + ≥2 new billionaire founders attributable to it by 2027-06.
- (2026-06) Recursive self-improvement threshold needs chips/inference speed, not Einstein-level AI. Watch-only: no date attached; falsifier would be a stalled capability curve despite ~100x chip-level gains.
Revealed behavior (Does)
- Raised and deployed Link-XPV with Diamandis as co-GP (KKCG investment announced 2025-12-16) — seed/incubation AI fund, ~75% MIT/Harvard ecosystem deal flow, offering data, compute, and office space. The podcast and the fund are now one funnel: media surfaces deal flow; deal flow feeds the media narrative.
- Doubled down publicly on maximal scaling bets for 2026 (100x prediction) rather than hedging into the "bubble" discourse — he keeps taking the over.
- Time allocation: weekly Moonshots co-hosting through 2026 plus fund management — he has become one of the most public voices of the exponentialist camp, a change from his thin pre-2024 footprint.
Feels
Energized by being early and right — the MIT-operator identity matters to him (technical credibility distinguishes him from "pure VCs"). Little visible fear of downside scenarios; disruption reads to him as opportunity, not loss.
Hears
MIT/Harvard founder ecosystem and Link portfolio companies first; the Diamandis abundance network; frontier-lab release notes and chip roadmaps. Notably absent from his input diet: labor economists and safety researchers.
Sees
Seed-stage deal flow from the MIT/Harvard ecosystem before it's public — a genuine early-warning channel on what technical founders are building — plus portfolio-company revenue curves that outsiders can't see.
Incentive map
Paid via Link Ventures / Link-XPV carry; selling (implicitly) the AI-goldrush thesis that makes his fund's vintage attractive. The Link-XPV/Diamandis tie-up deepens CONVENTIONS rule 25 exposure: bullish forecasts are now institutional marketing as well as opinion. Cannot say: that AI seed valuations are unsustainable; that the acronym-billionaire machine is froth; that enterprise adoption might stall. Weight his technical observations (chips, quantization, inference costs) above his market-timing claims.
Theories aligned with
- Techno-optimism — scaling-laws / agents variant
- Adjacent to post-labor economics — implicit in "agents do the work" framing
- "Exponential Organizations" (with Salim Ismail)
What he's reacting against
- "Wait and see" enterprise IT posture toward AI deployment (a foil he himself now says is dissolving — Jun 2026)
- SaaS incumbents who treat AI as a feature add rather than an existential reframing of their category
- Academic / regulatory framings that treat capability gains as decoupled from deployment economics — sharpened in 2026 into the Argentina regulatory-arbitrage argument
- Linear cost-of-software thinking — argues marginal cost trends to near-zero with AI-native delivery
- Investor caution on "AI bubble" framing — argues capability and revenue curves are both real
Where he overlaps / splits (with Rich)
- Overlaps tightly with Diamandis, Ismail, Mostaque, Wissner-Gross — Moonshots crew shares the exponential lens; Blundin is the deepest on scaling-laws specifics, and is now formally in business with Diamandis (Link-XPV)
- Overlaps with Andreessen, Altman on speed and agents as the unlock; splits on policy framing (less politically combative than Andreessen, less institutional than Altman)
- Overlaps with Amodei, Hassabis on capability trajectory; splits on safety-as-binding-constraint — Blundin treats it as solvable in stride, not gating
- Splits with Brynjolfsson — expects displacement faster, less J-curve patience
- Splits with Acemoglu, Cowen on diffusion speed — vertical-by-vertical capture in 3–5 years, not 10–20
- Less engaged with Yudkowsky, Bengio safety arguments — operationally background, not action item
- For Rich's bench: his picks-and-shovels concentration matches the infra thesis; his inference-deflation numbers are a useful input to the AI-native conversion scorecard — but discount his agent-adoption timelines by the rule-25 haircut.
Track record
- Link Ventures portfolio: long deep-tech / AI history; EverQuote co-founder (public company) [TBD: catalog notable wins/losses]
- Cogito: emotional-intelligence AI for call centers — pre-LLM AI, mixed but durable
- MIT background: technical credibility relative to pure-VC commentators
- 2026 predictions are his first crisply falsifiable public calls (100x scale; acronym billionaires) — a scoreability upgrade from the v1 pattern of unfalsifiable directionality
- Pre-2024 he was less public-facing; the podcast (and now the fund with Diamandis) made him a leading exponentialist voice
Empirical vs normative
- Empirical: scaling laws continue; agents are the unlock; software categories will be rewritten; 100x scale in 2026; national self-sufficiency triad (US/France/Argentina)
- Normative: founders should build replacement-grade AI products in defensible verticals; investors should concentrate on agents + infrastructure; regulation-constrained deployment should route to friendlier jurisdictions
- Conflict of interest: substantial and growing — Link Ventures AI exposure plus Link-XPV institutional fund co-led with a Moonshots co-host (CONVENTIONS rule 25)
Weak spots / open questions
- Public written record still thin relative to podcast output — triangulation against written sources remains hard
- "3–5 years to rewrite software categories" is scoreable but vertical results will vary — track which categories actually flip
- Less engaged with Acemoglu-style welfare arguments about displacement-mode AI
- Underweights enterprise integration friction, change management, and security/eval gating — consistent with VC bias (though his own 0.1%-lawsuit anecdote concedes the friction is real)
- Doesn't directly address safety arguments — treats them as separable from deployment forecasting
- "Inference cost down 10x/year" / "40x deflation" is benchmark-dependent; needs like-for-like comparison
- The Argentina self-sufficiency triad is a strong claim sourced (so far) only to his own podcast remarks — verify the energy/materials analysis before reusing it
- Podcast + fund is now one commercial machine — separating analysis from deal-flow marketing gets harder each quarter
Rich's take
- (your synthesis here)
Delta log
2026-09-04 — v1→v2 migration + validation (weekly batch)
- Grades: 5 Held / 2 Drifted / 0 Wrong.
- Held Scaling-laws continuation; agents-as-unlock; AI-native economics; VC concentration thesis (enacted via Link-XPV); integration/eval bottleneck.
- Drifted "Enterprise underestimates speed / wait-and-see" — he himself now reports legacy enterprises "adopting AI faster than expected" (Moonshots summary, 2026-06-06); the foil dissolved.
- Drifted "Agents mainstream in enterprise by 2025–26" — shipped widely, deployed unevenly; "mainstream" only under a generous reading.
- New behavior logged: Link-XPV AI seed fund co-led with Diamandis, KKCG LP investment (KKCG, 2025-12-16) — podcast and fund now one funnel; rule-25 discount deepened.
- New falsifiable 2026 predictions logged with external probability assessments (ImpactLab, 2025-12-21): 100x model scale via quantization/Chinese research; acronym-minted billionaires. Plus Jun 2026 claims: recursive self-improvement needs chips not genius; US/France/Argentina self-sufficiency triad.
- Most surprising delta: he formalized the Diamandis relationship into an institutional fund — the "independent technical voice" and the abundance-marketing machine are now the same balance sheet.
- Tier: quarterly (active fund deployment + dated 2026 predictions to score). Wake triggers: fund/exit news (moves incentive map), portfolio unicorn events (tests thesis), public revision of the 100x or 3–5-year calls (grade immediately), job change (canonical miss).
Sources
- Podcast: Moonshots (primary current source) — Jun 6 2026 summary (recursive self-improvement, Argentina, enterprise adoption) · "The AI Wealth Gap: Why 40x Deflation Changes Everything" (2026-08)
- Fund: KKCG invests in Link-XPV (2025-12-16) · Link Ventures · Crunchbase
- Predictions: Assessing the Moonshots 2026 Predictions (ImpactLab, 2025-12-21)
- Profiles: Crunchbase (EverQuote co-founder) · XPRIZE · LinkedIn
- Social: X/Twitter and LinkedIn posts (lower tier per CONVENTIONS rule 4)
migrated v1→v2 2026-09-04 (weekly validation batch) · backup: _archives/dave-blundin.html.bak-20260904 · AI & Society domain