2026-03-02

Anthropic vs. The Pentagon, Claude Outpaces ChatGPT, and Consulting Gets Replaced | #234

The panel opens on the standoff between Anthropic and the Pentagon over autonomous-weapons and surveillance safeguards, then turns to India's AI Impact Summit and the New Delhi Declaration as evidence that AI has become full-blown geopolitical infrastructure. A large middle section covers Anthropic's enterprise-driven revenue overtaking OpenAI's consumer strategy, the arrival of 'recursive self-improvement' where models reportedly emit weights for successor models, and knock-on effects across cybersecurity stocks, hardware, and the 'Rent-A-Human' gig economy for AI agents. Energy and land-use friction (data centers vs. farmland), a big walk-back in OpenAI's projected compute spending, and biotech breakthroughs ($100 genome sequencing, cheap lab-grown meat) round out the news segment. The episode closes with a subscriber AMA covering AI job displacement and UBI, the future of universities, and whether AI agent instances could be considered conscious.

▶ Watch on YouTube

Topics

Anthropic vs. the Pentagon over autonomous weapons and surveillance Geopolitics ▶ 0:00

The Department of War pressured Anthropic to remove safeguards blocking use of its models for autonomous weapons and mass surveillance; Dario Amodei refused, risking $200M in contracts, while Secretary Hegseth floated invoking the Defense Production Act and branding Anthropic a supply-chain risk.

India's AI Impact Summit and the New Delhi Declaration Geopolitics ▶ 1:47

Top AI CEOs and Modi convened in India for a summit that produced $250B in committed AI investment and an 88-nation declaration (including the US, China, and Russia) on diffusion, transparency, and public-good use of AI; the panel debated India's neutrality strategy and whether Chinese open-weight models will win the region.

Frontier lab CEO clips and the shift from AI hype to inevitability AI ▶ 12:34

Clips from Sundar Pichai (data centers in India and space), Sam Altman (warning about dictator-aligned superintelligence and AI warfare), and Demis Hassabis (AGI as 10x the Industrial Revolution in 1/10th the time) prompted discussion of AI CEOs becoming de facto geopolitical and moral actors.

Anthropic's revenue overtakes OpenAI: enterprise beats consumer AI monetization Economy ▶ 35:34

A chart showed Anthropic revenue growing 10x/year versus OpenAI's 3.4x/year, crossing over in 2026; the panel argued this reflects Anthropic's enterprise/coding focus versus OpenAI's consumer-subscription strategy, since enterprises consume far more (and more expensive) reasoning tokens than consumers want.

Recursive self-improvement: models emitting weights for successor models AI ▶ 41:23

Prompted by an OpenAI Codex lead's claim of massive progress within 10 weeks, Alex Wissner-Gross argued the industry has moved past pretraining-era (yearly) and reasoning-era (quarterly) improvement cycles into a phase where models can emit weights directly for successor models, compressing capability jumps to weeks.

Claude Code security tool crashes cybersecurity stocks AI ▶ 46:04

Anthropic's unveiling of a Claude Code-based security tool sent cybersecurity stocks down sharply on announcement alone; the panel discussed AI-vs-AI cybersecurity, the NIST vulnerability database being overwhelmed by AI-discovered bugs, and the shifting bottleneck from vulnerability discovery to trustworthy remediation.

Rent-A-Human: humans as "meat puppets" for AI agents Economy ▶ 51:19

Rent-a-human.ai passed 500,000 humans registered to perform physical/sensory tasks for AI agents via MCP; the panel framed this as an inversion of Moravec's paradox and a possible bootstrap labor market until humanoid robots arrive.

Data centers vs. farmland: land, energy, and NIMBYism Energy ▶ 1:17:40

US farmers rejected multi-million-dollar bids to convert farmland into data centers; the panel argued the land/energy/water footprint needed for AI compute is tiny relative to the country, but warned rising NIMBY-style backlash could fuel civil unrest and cede ground to China.

OpenAI revises 2030 compute spending from $1.4T to $600B Compute ▶ 1:22:36

OpenAI cut its projected 2030 compute spend roughly in half; the panel debated whether the original $1.4T figure was fundraising theater or whether TSMC simply reallocated chip supply elsewhere, noting total industry CapEx and demand for chips hasn't actually declined.

Biotech breakthroughs: $100 genome sequencing and lab-grown meat economics Biotech ▶ 1:25:39

Element Biosciences' Vitari device promises $100-per-genome sequencing, reviving talk of sequencing every newborn and hospital patient and of environmental DNA sequencing; separately, lab-grown meat costs have collapsed from $330,000/lb in 2013 to $10/lb in 2025, spurring an ethics discussion including off-world food production.

Tesla FSD safety data and robo-insurance economics Transportation ▶ 1:36:38

Tesla logged over 8 million supervised FSD miles with roughly 9x the safety of the US driving average; Lemonade now offers FSD-mileage insurance discounts, and the panel discussed how autonomy could shrink the US car fleet from ~140M to ~20M vehicles and hollow out the auto-insurance industry.

AI-driven white-collar job displacement, UBI/UHI, and looming social unrest Economy ▶ 1:42:18

Reacting to Andrew Yang's prediction of mass white-collar layoffs, the panel split between skepticism (citing IBM's increased entry-level hiring) and alarm about an unprepared government, a collapsing payroll tax base for Medicare/Social Security, and the risk of near-term social unrest and dysfunctional UBI politics.

Predictions made

open Alex Wissner-Gross: Countries outside the US will look back and question why AI model training remained so centralized in the US while inference deployment became so decentralized globally.

EP #? · · due: 1-2 years · ▶ watch

“I suspect a year from now, 2 years from now, we'll look back and we'll wonder why...training was so centralized all the while inference time was so decentralized.”

Your call:

open Alex Wissner-Gross: The Pentagon and Anthropic (and other frontier labs with similar stakes) will find a way to resolve the standoff over military use of AI models amicably.

EP #? · · due: unspecified · ▶ watch

“My expectation is that the Pentagon and Anthropic and also the other frontier labs...will find a way to resolve this amicably.”

Your call:

open Peter Diamandis: If Anthropic's revenue growth rate (10x/year) holds, it will reach roughly $1 trillion in annual revenue by around 2029.

EP #? · · due: 2029 · ▶ watch

“If you extrapolate that growth rate for Anthropic, you hit a trillion dollars of revenue in like 2029.”

Your call:

open Dave Blundin: Anthropic, OpenAI, and SpaceX will go public amid a hot IPO market, potentially within about six months, at extraordinary valuations.

EP #? · · due: ~6 months (mid-2026) · ▶ watch

“Anthropic going public, OpenAI going public, SpaceX going public. These are going to be insane numbers in the next...6 months?”

Your call:

open OpenAI Codex lead (clip): The current state of AI coding agents will look primitive compared to what arrives within 10 weeks.

EP #? · · due: ~10 weeks from March 2026 · ▶ watch

“I'm beyond excited for the next 10 weeks will bring. I think the current state of coding agents will be remembered as being so primitive, it'll be funny in comparison.”

Your call:

open Alex Wissner-Gross: Expect major AI capability jumps within weeks (not quarters) as models begin generating weights for successor models via recursive self-improvement.

EP #? · · due: next few weeks from March 2026 · ▶ watch

“So expect big things over the next few weeks.”

Your call:

open Alex Wissner-Gross: The next major foundation-model breakthroughs will come from small language model research rather than frontier large models.

EP #? · · due: unspecified · ▶ watch

“I strongly suspect that the next major revolutions...in foundation models will come from the small side.”

Your call:

open Alex Wissner-Gross: Andrew Yang's predicted mass white-collar job losses from AI will not materialize; instead more hiring/work will occur, similar to IBM increasing entry-level hires.

EP #? · · due: unspecified/near-term · ▶ watch

“Are they likely to? No. I think we're going to see the opposite.”

Your call:

open Dave Blundin: Massive social unrest driven by AI-related economic disruption is imminent, arriving by the end of 2026 and before the next US presidential election.

EP #? · · due: end of 2026 / before next presidential election · ▶ watch

“There's going to be massive social unrest, huge social unrest, and it's imminent. It's coming, you know, toward the end of this year, and certainly before the next presidential election.”

Your call:

open Alex Wissner-Gross: There is roughly a 50% chance humanity discovers/exhausts fundamental physics (e.g., a unified field theory) within the next few years with the help of superintelligence.

EP #? · · due: next few years · ▶ watch

“I assign it maybe 50% probability that we run out of fundamental physics at some point, maybe even in the next few years.”

Your call:

Numbers that matter

$250 billion in combined AI investment committed at India's AI Impact Summit Includes a joint $210B commitment from Reliance and Adani.
88 nations signed the New Delhi Declaration First global AI agreement including the US, China, and Russia; covers diffusion, frontier transparency, and AI for public good.
$15 billion Google AI infrastructure investment in India Announced via Sundar Pichai's clip for a full-stack AI hub in Visakhapatnam.
$200 million in Pentagon contracts at risk for Anthropic Amount Dario Amodei is reportedly risking by refusing to remove AI safeguards for the Department of War.
~$4 billion launch valuation for Brett Adcock's new company Hark Adcock (Figure CEO) launched a separate physical-AI lab; he also has a personal net worth pegged by Forbes at $19.1 billion.
Anthropic revenue growing 10x/year vs. OpenAI's 3.4x/year Chart showing the two companies' revenue growth slopes crossing over in 2026.
Extrapolated ~$1 trillion Anthropic revenue by 2029 Based on extrapolating the 10x/year growth line; implied valuation cited at $20-30 trillion.
500,000+ humans registered on Rent-A-Human.ai to serve AI agents Framed as 'meat puppetry' — humans hired by AI agents via MCP for physical/sensory tasks.
OpenAI hardware team scaling to 200 people Building smart speakers and glasses with built-in cameras, targeting a 2027 launch to rival Amazon Alexa and Google Home; Jony Ive involved.
nanoGPT speedrun time fell from 48 minutes to 90 seconds in one year Community optimization of Andrej Karpathy's open-source small-model training repos.
Tech companies offered $33-$80 (thousand, per acre) for US farmland for data centers Farmers refused the offers, prompting a land-use and food-vs-compute debate.
40,000 acres of proposed data center land, about half the size of Washington, D.C. Used to argue the land footprint for AI compute is small relative to the whole country.
OpenAI cut projected 2030 compute spending from $1.4 trillion to $600 billion Panel debated whether the original figure was inflated for fundraising or reflects real efficiency/TSMC allocation shifts.
AI industry spend currently ~$2 billion/day, projected to reach $3-5 billion/day by 2030 Cited as evidence CapEx and revenue growth are racing each other.
$100 genome sequencing via Element Biosciences' Vitari device Machine priced around $600,000+, expected to ship in the second half of 2026.

Worth digging into

🕳️ The Anthropic-Pentagon DPA/SIPRNet standoff

Sets a precedent for how much control government can exert over frontier AI values and deployment in defense contexts, and whether a 'supply chain risk' designation can be weaponized against an otherwise compliant vendor.

🕳️ Models emitting weights for successor models (recursive self-improvement)

If accurate, this marks a qualitatively new phase of AI development compressing capability jumps to weeks, one of the most consequential technical claims in the episode.

🕳️ Anthropic vs. OpenAI revenue growth chart (10x vs. 3.4x/year)

Underpins the episode's central economic claim (trillion-dollar revenue by 2029) and the consumer-vs-enterprise AI monetization thesis.

🕳️ OpenAI's compute spending revision ($1.4T to $600B by 2030)

A large, specific walk-back in projected CapEx that could indicate overhyped fundraising numbers, or real efficiency gains and TSMC chip-allocation shifts toward other customers.

🕳️ Element Biosciences' Vitari $100 genome sequencer

Would mark a major inflection in genomics if verified, since genome sequencing costs plateaued for years despite the Illumina/Element rivalry.

🕳️ Rent-A-Human and the 'meat puppet' economy

A concrete, emerging labor market where AI agents hire humans for physical/sensory tasks via MCP — an early real-world instance of Moravec's paradox playing out commercially.