20VC: Anthropic vs The Pentagon: Who Wins | The Ultimate Stock Picks: What to Buy | The Data Centre Arms Race: Is the Capex War Stalling | The Era of Public Company Deceleration is Dead

What It Covers

Harry Stebbings, Rory O'Driscoll, and Jason Lemkin analyze four major stories: Anthropic's lawsuit against the Pentagon over a supply chain risk designation threatening billions in contracts, Oracle and OpenAI scaling back Stargate data center expansion, Meta absorbing surplus AI capacity, and CrowdStrike beating earnings while trading down, plus stock picks across six public companies.

Key Questions Answered

• •Anthropic vs. DOD blast radius: The Pentagon's supply chain risk designation costs Anthropic roughly $200M in direct government revenue — approximately 1% of their run rate at $1.5B annually — but the real damage is B2B sales friction. Prospects with any federal exposure are cutting deals in half or switching to OpenAI and xAI, which carry no equivalent designation risk. Legal consensus favors Anthropic on the merits, but winning in court won't end the political pressure.
• •CapEx cycle reality check: $600B in annual AI infrastructure spending across hyperscalers equates to roughly $4,000 per US worker. Oracle's Stargate pullback from 2GW to 1.2GW reflects a weak balance sheet, not demand collapse — Meta immediately absorbed the surplus capacity. The companies that can sustain spending (Meta, Google, Microsoft) are betting on 24/7 persistent AI agents requiring orders-of-magnitude more compute than current episodic usage patterns generate.
• •Junior role elimination as CapEx funding mechanism: Enterprise demand for AI agents is accelerating the elimination of entry-level positions in software development, legal, customer support, and sales. At Penn State, only six students in an entire CS cohort received tech offers. The budget freed from not hiring and training juniors is being redirected toward AI tooling and compute — making this a self-reinforcing cycle that accelerates data center investment.
• •Reacceleration as the only viable public market thesis: The era of "gentle deceleration" — where SaaS companies managed gradual growth slowdowns while expanding margins — ended in 2025. Public markets now reprice decelerating companies to 8-9x EBITDA with no premium. CloudFlare accelerated from 27% to 34% revenue growth with 40% net new customer growth year-over-year. Founders and portfolio managers should treat any company not actively reaccelerating as a terminal value story requiring immediate intervention.
• •Agentic B2B demand outpaces supply of deployment talent: The binding constraint for AI B2B companies like Intercom, Sierra, Harvey, and Legora is not product quality but forward-deployed engineers (FDEs) capable of onboarding enterprise customers within 30 days. Vendors without sufficient FDE capacity are losing deals regardless of agent performance. Founders building in this space should treat FDE hiring and training as a primary growth lever, not a post-sales afterthought.
• •Stock picks framework — growth tier segmentation: The panel segments public market bets into three tiers: value plays at 8-9x EBITDA (Salesforce, Intuit, Toast); GARP at early-teens EBITDA (CrowdStrike, Atlassian); and story-priced momentum stocks above 30x EBITDA (Palantir, CloudFlare, Shopify). Consensus picks include CrowdStrike for cybersecurity durability, Palantir for two-year administration tailwinds, CloudFlare for AI infrastructure positioning, and Nubank as an underappreciated high-growth fintech entering the US market.