20VC: Sam Altman Offers Trump 5% of OpenAI: Fool or Genius? | Alex Karp Sounds the Alarm: Enterprises Fear Frontier Models & Questionable ROI of AI | The Rise of Chinese Open Source: Deepseek Building Own Chips

What It Covers

Harry Stebbings, Rory O'Driscoll, and Jason Lemkin analyze five major tech stories: the US government lifting the Claude 3.5 Opus export ban, Sam Altman's proposal to give the US government a 5% OpenAI stake, DeepSeek building its own chips, Meta launching a cloud compute business, and Alex Karp's warnings about enterprise AI skepticism and ROI concerns.

Key Questions Answered

• •Government Equity Stakes as Strategic Anchoring: Sam Altman's 5% government stake proposal functions as a deliberate anchoring tactic, not a naive giveaway. By publicly framing 5% as the appropriate number, OpenAI shapes the political conversation before regulators set their own terms. The risk: once you invite government ownership, the political logic can escalate from 5% to 50%, as Bernie Sanders has already suggested publicly.
• •Founder Dilution Tolerance Has Fundamentally Shifted: Modern AI founders accept dilution levels that would have been unthinkable five years ago. Dario Amodei owns roughly 1.7% of Anthropic after dozens of funding rounds. The practical implication for seed investors: model your real entry price at 4x the nominal valuation to account for cumulative dilution across 15-20+ rounds, not the historical 2x assumption.
• •Enterprise AI ROI Remains Unproven at Scale: Alex Karp identifies two concrete enterprise blockers — unclear return on AI investment and fear that frontier model providers are training on proprietary business data. HubSpot's forced rollback of its contact-sharing prospecting feature within one week of launch confirms that data privacy concerns are not theoretical. Enterprises in regulated industries are actively choosing on-premise or isolated deployments over frontier model APIs.
• •Chip Self-Sufficiency Follows Compute Ownership Logic: DeepSeek and Anthropic both pursuing custom silicon reflects a strategic principle: controlling compute infrastructure is as critical as controlling model weights. The counterargument — that NVIDIA will build custom chips for any customer generating sufficient revenue — underestimates the margin capture motivation. Companies spending billions annually on NVIDIA GPUs are reclaiming 40-60% gross margin by moving to proprietary silicon.
• •Chinese AI Dominance in Video Is Already Established: Kling holds the top position in AI video generation globally, reaching $500M ARR in Q1, while OpenAI shut down Sora citing cost inefficiency. The structural reason: OpenAI's GPUs generate higher returns in enterprise coding than consumer video at roughly $1.30-$2.00 per 30-second generation cost. Chinese developers built video leadership partly because US frontier models are inaccessible inside China's firewall.
• •Liquidity Programs Are Now a Hiring Prerequisite: Top engineering talent evaluates startups based on the probability of tender offers within 24 months, not just IPO potential. Companies like Eleven Labs ($22B secondary) and Clay ($5B tender) have normalized regular liquidity events. Founders below unicorn status who cannot credibly signal near-term secondary options face a structural disadvantage in recruiting against Anthropic and OpenAI, which offer their own liquidity mechanisms.