The Twenty Minute VCThe Early Days of Anthropic & How 21 of 22 VCs Rejected It | The Four Bottlenecks in AI | Anj Midha
CHAPTERS
- 0:00 – 1:25
Why scaling laws still work—diminishing returns depend on the domain
Anj argues scaling laws are not “dead,” but some benchmarks (like coding evals) appear saturated, making gains look expensive. In less-explored domains like materials discovery, he claims additional compute still yields outsized improvements when paired with tight experimentation loops.
- •Diminishing returns show up in well-explored tasks (e.g., certain coding evals)
- •In materials science, more compute can still drive large capability jumps
- •Periodic Labs example: LLM proposes materials → robots synthesize → instruments validate → data feeds training
- •“Bitter lesson” framing: scale + data/feedback loops continue to pay off
- 1:25 – 2:55
The four bottlenecks holding AI back: context/feedback, compute, capital, and culture
Anj lays out a framework for what limits frontier progress today, emphasizing that algorithms are less of a bottleneck than the systems around them. He highlights culture as the meta-bottleneck that enables top talent and flexible research direction.
- •Four bottlenecks: context & feedback loops, compute infrastructure, capital, and culture
- •Algorithmic innovation is downstream of attracting great researchers (culture)
- •Context/feedback loops are where capability progress and commercial advantage become legible
- •Capital must fund iterative loops and infrastructure (land/power/shell, clusters, financing)
- 2:55 – 7:36
Why “AI for science” underperformed: missing data and weak real-world feedback loops
He describes benchmarking frontier models on physics/chemistry and finding them surprisingly weak relative to the hype. The core issue is lack of high-quality scientific data on the public internet and difficulty accessing lab/manufacturing datasets, motivating vertically integrated data generation.
- •Marketing narratives outpaced real model performance in scientific reasoning (at the time)
- •Scientific data is scarce online and often locked in national labs, academia, or industry
- •Progress requires building new context/verification loops, not just scraping the web
- •Periodic Labs’ physical lab setup is positioned as the solution to this bottleneck
- 7:36 – 9:36
Vertically integrated model companies and the “Claudification” question
Harry probes how to predict which vertical AI companies get commoditized by foundation models. Anj reframes moats: unique context/feedback access makes progress visible and can support superior economics, but it’s not automatically a permanent moat.
- •Vertical integration will proliferate across domains to create proprietary feedback loops
- •Not all “context” is a durable moat, but it signals where progress is possible
- •Companies tied to physical-world data generation are harder to commoditize
- •Moat analysis is less useful early than identifying bottlenecks and differentiated access
- 9:36 – 13:31
Sovereign data, the CLOUD Act, and why Europe wants local AI infrastructure
Anj explains how legal and geopolitical constraints make certain workloads impossible to run on US-managed clouds. He uses the CLOUD Act to motivate “sovereign” infrastructure and local providers that can serve sensitive enterprise and government workloads.
- •CLOUD Act: US-managed infrastructure may be subject to US government data access
- •Defense/logistics/manufacturing workloads often require local processing and control
- •Europe lacks sufficient trusted AI infrastructure providers at scale
- •Sovereignty creates an opening against hyperscaler dominance for startups
- 13:31 – 14:27
The investment thesis behind Mistral: full-stack European independence
He describes Mistral as a bet on European sovereignty across the AI stack—power, facilities, compute, and locally trained models—alongside open deployment. The goal is independence at scale rather than relying on US hyperscalers and labs.
- •Thesis: independent European stack—land/power/shell, compute, and model training
- •Open models enable customization and deployment while preserving sovereignty options
- •Government and enterprise demand can anchor large infrastructure buildouts
- •Positioning challenges the long-running hyperscaler lock-in dynamic
- 14:27 – 20:52
The brutal early days of Anthropic: 21 of 22 VCs said no
Anj recounts helping founders translate a scaling-law research hypothesis into a business plan, then facing broad investor skepticism. He describes how many VCs didn’t understand GPT-3 or compute-driven scaling economics, while strategics like Amazon immediately saw the alignment.
- •Early 2021: weekly sessions to turn “scale” into a repeatable product/business loop
- •Initial target raise was far larger; re-anchored to a $100M seed round
- •Many VCs didn’t know GPT-3 or grasp compute multipliers and scaling economics
- •Amazon partnership made sense strategically (cloud distribution + compute + capital)
- 20:52 – 23:06
Public Benefit Corporations (PBCs): resolving mission vs profit tensions
Anj defends PBC governance as a mechanism to make long-term decisions that aren’t maximally profit-seeking in the short run. He positions AMP as mission-aligned infrastructure + standards advocacy, including offering compute at cost to support frontier progress.
- •PBCs can legitimize long-horizon trade-offs between mission and profits
- •AMP provides compute at cost to expand access for real frontier innovators
- •He argues PBCs help avoid harmful short-term shareholder pressure
- •Frames AMP as both infrastructure operator and standards evangelist
- 23:06 – 25:21
The AMP Grid: building an electricity-grid analog for compute
Anj describes AMP as an “independent system operator” coordinating compute capacity rather than owning datacenters like a cloud provider. The thesis is that pooling and dispatching capacity improves utilization, reduces overprovisioning, and accelerates frontier output—similar to early electricity markets.
- •AMP aims to coordinate compute supply so teams provision baseload, not peak
- •Analogy: 1885 electricity era—factories running private generators at low utilization
- •Grid-like pooling lets different workloads spike at different times for higher utilization
- •Compute procurement and partnerships built from early relationships (e.g., a16z Oxygen program)
- 25:21 – 35:30
Back-to-the-future venture: co-founding and incubation like the early Valley
He argues frontier companies require deep operational partnership, not just check-writing. Drawing on Intel/Genentech/Apple examples, he suggests value accrues to investors who help build companies hands-on—especially where CapEx, infrastructure, and scientific execution dominate.
- •Historical model: Arthur Rock, Genentech-at-Kleiner, Mike Markkula at Apple
- •Incubation requires time-on-site and close partnership with technical founders
- •Hard for “spray-and-pray” VC to coexist with deep incubation in one person/firm
- •Frontier industries shift venture from finance toward company-building
- 35:30 – 37:49
GPU wastage bubble, not an AI bubble: stranded compute and poor utilization
Anj claims the core problem is infrastructure inefficiency rather than capability hype. Large pockets of compute sit idle due to fragmentation, mismatched needs, and lack of coordination—creating the appearance of a bubble even as capability demand remains real.
- •He rejects the notion of an AI capability bubble; focuses on infrastructure waste
- •Stranded compute exists across the ecosystem despite high demand
- •Pooling/coordination is presented as the remedy (grid model)
- •Inefficiency distorts perceptions of overinvestment
- 37:49 – 42:16
Why compute isn’t fungible: chip heterogeneity and missing standards
He explains that unlike electricity, compute capacity can’t be easily swapped across chip types or clusters, even within Nvidia generations. This prevents workloads from moving efficiently, creates stranded assets, and amplifies boom/bust dynamics typical of pre-standardization infrastructure eras.
- •Compute isn’t interchangeable across H100/GB200/GB300 clusters (memory/architecture constraints)
- •FLOPS are not “equal” in practice due to system-level dependencies
- •Lack of open protocols/standards prevents efficient transfer of capacity
- •Standardization could reduce waste and stabilize infrastructure cycles
- 42:16 – 45:15
China’s systems co-design advantage and distillation as a catch-up engine
Anj argues China is competing via full-stack optimization rather than leading-edge chips alone—co-designing chips, infrastructure, and training to boost efficiency. He highlights adversarial distillation from Western endpoints as a mechanism to accelerate iteration and close the gap.
- •Race framed as systems co-design, not just a chip race
- •Huawei + stack integration can rival frontier performance through efficiency gains
- •Adversarial distillation: extract capability signals from Western models at scale
- •Open releases can be a strategic bootstrapping phase until domestic sufficiency
- 45:15 – 49:07
Coordinating defense: an “Iron Dome” for inference and frontier security
He warns that distillation and insider threats exploit fragmented defenses across Western labs. His proposal is a shared proxy/coordination layer across inference providers so attacks seen by one actor can trigger rapid collective response.
- •Threats: distillation attacks, insider risks, vulnerable inference endpoints
- •Current defense is informal (founder group chats) and doesn’t scale
- •Proposal: shared inference proxy to detect/coordinate responses across labs
- •Without coordinated security, staying at the frontier becomes unsustainable
- 49:07 – 1:01:43
Perfect competition is for losers: aiming for “optimal competition” in AI markets
Updating Thiel’s “competition is for losers,” Anj argues the real enemy is perfect competition that commoditizes everyone and wastes scarce resources. He advocates an “optimal competition” structure with a small number of strong players per layer to maintain innovation without monopoly stagnation.
- •Perfect competition leads to thin margins and weak defensibility (restaurant analogy)
- •Monopolies become “mafias” that hoard resources and slow innovation
- •Optimal competition: ~3–4 serious teams per frontier to sustain pressure and progress
- •Compute scarcity makes over-funded duplication especially destructive
- 1:01:43 – 1:15:18
Quick-fire: LP advice, building to learn, legacy, and personal reflections
In the closing segment, Anj emphasizes that LPs and investors must do the work—read, build, and understand bottlenecks—rather than outsource judgment. The conversation shifts to leadership qualities, health and time, independence as motivation, and what he wants to be remembered for.
- •LPs should educate themselves and invest around bottlenecks, not narratives
- •GPs should build with AI tools to understand constraints firsthand
- •Dario’s strengths: scientific brilliance, truth-seeking empiricism, mission focus/culture
- •Personal reflections: health/time, family, independence, and desired legacy (“He was right”)
