Roundtable
OpenAI & SpaceX S1 Drops | Layoffs at Cloudflare & ClickUp | OpenRouter & Polsia Raise Mega Rounds
- Anthropic is projected to surpass OpenAI in profitability, growth speed, and size within a couple of quarters, potentially making OpenAI a visible "number two" losing money if it delays an IPO.
- OpenAI faces strategic pressure to go public immediately to avoid a deteriorating position, whereas Anthropic may remain private if it sustains profitability with high demand, potentially staying like Stripe unless capital needs exceed hundreds of billions.
- Global AI capital expenditure is projected to reach $3 trillion by 2030, a scale potentially requiring public market access for the necessary funding, though current AI CapEx spend is expected to hit $100 billion this year with a run rate of $300–320 billion for NVIDIA.
- SpaceX's future valuation is heavily tied to AI; if it secures $15 billion in annual revenue from Anthropic, it could cover capex with high returns, but the "data centers in space" narrative is required for the merger to succeed, or it risks being a "bridge too far."
- By 2030, Starlink is expected to constitute the vast majority of SpaceX's value, while data centers in space would likely contribute a low percentage of total revenue and be a high-revenue, relatively low-return-on-equity business.
- If the $3 trillion global CapEx scenario materializes, the SpaceX AI business could be valued at $2 to $3 trillion, but if the "data centers in space" narrative fails, this valuation is considered a "rocket too far."
- The market may currently view 100x trailing sales valuations as "madness" or a "GeoCities deal" that investors will later ridicule, similar to how NVIDIA's stock growth has slowed to 20% as the market priced in 3x to 5x growth from a year and a half ago.
- As AI spending consumes significant portions of corporate wages, the narrative will shift from "vibe-based" spending to quantitative ROI demands, likely causing "token maxing" to cease and leading to layoffs or a halt in spending for firms lacking proof of value.
- Agentic expertise is expected to become as ubiquitous as Excel usage within 12 to 24 months, creating a bifurcation where "agentic experts" or "10x employees" command $1–2 million in salaries, while others risk becoming unemployable or facing increased "idle idleness."
- Companies with $2 million in revenue per employee can absorb higher token costs, but a primary constraint will shift from token prices to the human capacity to process AI output, necessitating high-value talent to filter ideas.
- Anthropic's premium pricing strategy may not be sustainable if the market determines ROI is absent for high-cost products, potentially forcing a price cut to match competitors or leading to a scenario where only high-ROI users increase usage even if prices double.
- The next phase for agent tools includes "picks and shovels" like structured web search and model switching becoming essential primitives, with agent search tools like Exa expected to become vital for app companies within five years.
- OpenAI may eventually replicate SpaceX's strategy of allocating massive share portions to retail investors if IPO demand is high, while the S1 filing suggests 90% of SpaceX's Total Addressable Market is in AI rather than traditional launch sectors.
- Stock market reactions to news like NVIDIA earnings are driven by the delta between actual results and expectations rather than the news itself, with the market currently happy but not anticipating 80% annual growth despite revenue trends.
- A potential risk exists that if no one else can build data center capacity efficiently, a single entity could potentially own every data center in the galaxy within five years, provided the $3 trillion CapEx reality holds.
- SpaceX's AI business has a 50% chance of failing if the "data centers in space" narrative is false, yet the market anticipates it could persist for five years similar to public storage if current revenue projections are met.