Marc Andreessen - The Battle For Tech Supremacy - [Invest Like the Best, EP.410]

At a Glance

Marc Andreessen (a16z co-founder, with both engineering background and historical strategic vision) in this interview systematically articulates his assessment of the AI competitive landscape, geopolitics, venture capital industry transformation, and power structure evolution, centered on the DeepSeek R1 open-source model event. Core thesis: DeepSeek R1 achieves top-tier AI reasoning capability at a 30x cost reduction and is fully open-sourced. This is not a threat from China, but a "wake-up call" to the West's "closed, controlled" AI path, forcing the US to return to its core values of "openness and freedom," or else lose this "Cold War 2.0."

Theme 1: DeepSeek R1 – The "Sputnik Moment" of AI and Paradigm Shift

Marc Andreessen believes that the release of DeepSeek R1 is a "earth-shattering" event in the history of AI, fundamentally changing the competitive landscape and adoption path of AI.

• Technical breakthrough and open-source strategy: DeepSeek is not an original breakthrough but a "very impressive implementation" of existing ideas from the US and Europe (e.g., neural networks can be traced back to 1943). Its key value lies in full open-source (code and technical papers), which is completely different from the previous practices of Chinese companies (e.g., Huawei), and instead resembles the original intent of Western "open AI." Users can download and run it themselves, requiring only $6,000 worth of hardware to obtain capabilities close to the top-tier systems of OpenAI/Anthropic, and can run it on a MacBook or iPhone through distillation technology.
• Cost and demand elasticity: DeepSeek achieves a 30x cost reduction. Marc cites "Jevons paradox" (i.e., falling costs stimulate demand explosion) and points out that a sharp drop in AI prices will lead to a surge in usage (possibly 100x or 1000x), thereby driving explosive growth across the entire industry. This is a huge boon for all AI application companies (e.g., AI lawyers) because their key input costs have plummeted.
• Confirmation of reasoning capability: DeepSeek proves that reasoning is feasible and will reach human or superhuman levels in fields such as programming, mathematics, physics, chemistry, biology, economics, finance, law, and medicine. He predicts: "Within five years, every person on Earth will have a superhuman AI lawyer and AI doctor available anytime, anywhere."

Theme 2: Competitive Landscape – The "Judo" of Open Source and America's Strategic Choice

Marc believes that DeepSeek's open-source creates a "paradoxical" situation: China's open system is challenging the US's closed AI companies, forcing the US to choose between "returning to openness" and "moving toward closure," and this will determine the outcome of "Cold War 2.0."

• Winners and losers (static perspective):
• Winners: All users (consumers and companies), and startups that rely on AI.
• Losers: Closed proprietary model companies (e.g., OpenAI, Anthropic), and NVIDIA (because DeepSeek uses cheaper chips and is more efficient, and may also use Chinese domestic chips).
• Dynamic perspective (elasticity effect): If demand explodes due to price declines, all companies (including OpenAI, Anthropic, NVIDIA) could benefit, creating a "rising tide lifts all boats" effect.
• US-China competition and "Sputnik moment": Marc compares the release of DeepSeek to a "Sputnik moment", i.e., a "sudden blow" to the US in the technology race. He points out that the US is in a "chaotic" state in AI policy: on one hand, it wants to beat China; on the other, it wants to "lock down and control" AI domestically like China. China's open-source strategy is a kind of "judo" , forcing the US to return to its core principles of "freedom, democracy, and openness."

Dimension	US (Current)	China (Current)	Author's Judgment
AI Model	Mainstream is closed (OpenAI, Anthropic)	Open (DeepSeek)	China is "leading" in openness, which ironically challenges Western values
Competition Strategy	Restrict chip exports (e.g., NVIDIA high-end chips)	Incentivize R&D for more efficient chips, build domestic supply chain	Export restrictions may be "counterproductive," accelerating China's technological self-sufficiency
Geopolitical Goal	Promote US tech stack globally	Promote China's tech stack globally (e.g., "Digital Silk Road")	AI will become the core of the next round of global tech stack competition

Theme 3: Evolution of Power Structures – The "Civil War" Between Elites and Anti-Elites

Marc interprets the current global political turmoil as a struggle between "incumbent elites" and "anti-elites," and technological change (including AI) is accelerating this process.

• "Iron Law" and elite circulation: Citing the "Iron Law" from The Machiavellians, he argues that all societies are ruled by "small organized elites" over "large unorganized masses." When elites become "corrupt and dysfunctional," anti-elites (e.g., Trump and Sanders in 2016) rise to challenge the incumbents under the banner of "better representing the masses."
• Current situation in the US and globally: Globally, trust in incumbent institutions is plummeting (e.g., Gallup surveys show a continuous decline since the 1970s, with a "cliff-like drop" after 2020). He believes this is not "the masses being misled," but rather the elites having "rotted" (e.g., Congress wasting taxes, media lying, university professors teaching "racial communism," etc.).
• His own experience: Marc uses his own experience as an example, describing how he went from being accepted by the elite (invited to Davos, etc.) to eventually becoming "mortal enemies." He observes that what elites need is not "new ideas," but "obedience." He ultimately chose to become an "anti-elite" and believes this was the right choice.

Theme 4: Capital Allocation Transformation – The "Eternal" and the "Present" of Early-Stage VC

Marc believes that AI will profoundly change investment analysis (especially late-stage and public markets), but the core of early-stage venture capital—judgment of people and deep collaboration—may be difficult to replace by AI, and this serves as the cornerstone for a16z's "franchise" building.

• Impact of AI on investment analysis: AI (especially reasoning models) will be able to analyze annual reports better than most analysts, but this will quickly become a "standard configuration," leading to an "arms race." For early-stage VC, he believes most of the work involves deep assessment of individuals and long-term deep collaborative relationships (e.g., starting a company with founders), which is akin to "project picking" or "talent scouting," a human activity thousands of years old, and may not be easily replaced by AI.
• a16z's "franchise" strategy: Marc distinguishes a16z from traditional "partnership" VCs, which typically "eat their own kill" and lack sustained value. a16z's goal is to build a company with "franchise value" , similar to large investment banks like Goldman Sachs. The core is to always maintain the central position of early-stage venture capital, and even as scale expands, it will never sacrifice the quality of early-stage investing for scale.
• Adapting to the "present": a16z continuously adjusts to adapt to new technologies and new environments. For example, in the past five years, political factors have become a non-negligible new dimension in company operations, so they have built a large political operations department, which was zero four years ago. He also notes that in the AI field, the previous industry iron law of "not investing in researchers" has been broken, and now the best AI companies are often founded by researchers.

Position Moves

Position	Guest Attitude	Key Data
DeepSeek	Positive (recognizes its open-source strategy and cost efficiency)	Achieves 30x cost reduction; can run on $6,000 hardware
OpenAI	Risk Warning (closed model suffers, "outmaneuvered")	Theoretically "open AI," actually "closed," accused of needing to rename to "ClosedAI"
Anthropic	Risk Warning (same ailment as OpenAI)	Closed model, does not publish research
NVIDIA	Risk Warning (faces challenges)	Chip profit margin as high as 90%; DeepSeek requires fewer chips and may use Chinese domestic chips
Meta (LLAMA)	Positive (open-source path)	Team is "dissecting" DeepSeek to improve next-generation LLAMA
Perplexity	Positive (as a representative of US companies leveraging DeepSeek)	Allows users to use DeepSeek R1 on US servers
Unitree	Positive (favorable view of its robotics business)	Robot dog (Unitree Dog) starts at $1,500, more than 30x cheaper than Boston Dynamics' $50,000–$100,000
Tesla	Neutral (mentioned as a representative of US manufacturing/robotics)	Service workshops are like "operating rooms," "computers + batteries"
Physical Intelligence	Not specified (mentioned as a US robotics "brain" company)	Building datasets required for robots
DJI	Risk Warning (geopolitical risk)	Global drone market share exceeds 99%, also heavily used by the US military, posing "backdoor" risk
Huawei	Risk Warning (geopolitical risk)	Previously sold telecom equipment worldwide at one-third the cost, constituting a "Chinese tech stack"
Boston Dynamics	Neutral (as a comparison object)	Robot dog price $50,000–$100,000

Judgments Worth Remembering

1. "DeepSeek is a 'wake-up call' for Western values." (Marc Andreessen) – China's open-source strategy forces the West to return to its core principles of "freedom, democracy, and openness," rather than imitating China's "control" model, or else it will lose this "Cold War 2.0."

2. "The cost reduction of AI will trigger the 'Jevons paradox,' leading to demand explosion." (Marc Andreessen) – If AI prices drop by 30x, usage may increase by 100x or even 1000x, ultimately benefiting all participants (including OpenAI, NVIDIA).

3. "Venture capital is an 'anti-elite' activity." (Marc Andreessen) – The core of early-stage VC is assessing and collaborating with "maverick" founders (anti-elites), which inherently conflicts with mainstream elite culture.

4. "Separating the 'eternal' (e.g., respecting founders) from the 'present' (e.g., adapting to AI, politics) is key to managing a VC firm well." (Marc Andreessen) – Corresponds to a16z's "franchise" strategy, with the core being to always maintain the central position of early-stage VC while quickly adapting to changes in technology, politics, and other external environments.

5. "NVIDIA's chip monopoly may be shaken by DeepSeek." (Marc Andreessen) – DeepSeek proves that top-tier AI performance can be achieved on cheaper, more efficient chips, while China is building a domestic chip supply chain, posing a long-term threat to NVIDIA.

6. "Weaponized drones are the greatest military innovation since the 'stirrup'." (Marc Andreessen, quoting a senior military planner) – A 20-person team equipped with drones could theoretically resist thousands or tens of thousands of conventional troops, fundamentally changing the economics of warfare.

7. "Historically, victory in Cold War 1 came from the economic 'supermarket' comparison, not guns." (Marc Andreessen) – The US, by demonstrating its economic vitality (e.g., a supermarket in New Jersey vs. a supermarket in Moscow) and technological superiority (e.g., personal computers), made the Soviet Union realize that "the Cold War cannot be won."