SpaceX enters its IPO week having transformed its financial profile in the span of roughly seven to ten calendar days. Two compute agreements, a $15 billion per year Anthropic deal and a newly closed $11 billion per year Google deal, add $26 billion of incremental annual revenue to a company that generated $18.6 billion in total revenue in 2025. On a pro forma basis that lifts total revenue to approximately $44.7 billion and swings reported income from negative $2.5 billion to positive $23.4 billion, because the data center capacity was already built and paid for and the new revenue drops almost entirely to the bottom line. The AI and data center division, once dismissed as a drag, is now the central revenue driver and operates with the economics of a hyperscaler comparable to AWS, Google Cloud, Azure, and CoreWeave, layered on top of Starlink and the launch business. The public market reaction has lagged the facts. Several equity research notes, including a Morningstar assessment valuing the company near $780 billion, were published without incorporating the Anthropic or Google agreements even though those deals more than doubled revenue. Prediction markets are more constructive, with Polymarket pricing a roughly 60 to 62 percent probability that SpaceX closes its first trading day above a $2 trillion market cap. The signal for investors is that consensus sell-side models have not yet caught up to the contracted revenue, which leaves an unusually wide gap between perception and materiality.
The more durable lesson from this debut applies to every IPO, not just this one. Newly listed stocks trade erratically in the days after pricing, driven by insider and institutional selling, employee liquidity, and portfolio rebalancing rather than by any clean read on fundamentals. Attempting to call the post-listing direction is closer to speculation than investing, and the more useful exercise is to ask whether the business will be worth more per share three to five years from now. That framing pushes attention back to business lines, addressable markets, and growth potential. For SpaceX specifically, the pending Cursor acquisition and a potential Tesla merger are exactly the kind of structural events that could reshape the financial profile inside that three-year window and belong in any serious underwriting. The house view is that high-quality pre-IPO and newly public franchises reward patience, and that buying with a multi-year holding period is a sounder approach than trading the opening print.
A quieter but equally important theme is the shift from prompting AI to designing loops that run AI continuously. A loop chains a series of tasks or skills and then instructs the system to keep iterating, correcting, and following up until a stated goal is met, which means the operator states an objective and the agent works out the execution and reports back when finished. Practitioners at the frontier, including the engineer Peter Steinberger and Claude Code lead Boris Cherny, now describe loops rather than one-off prompts as the default mode of work. The business implication is a move toward an AI-first operating model, the same kind of structural rethink that internet-first and mobile-first once forced, where teams, policies, and products are organized around autonomous agents rather than manual workflows. This is precisely where forward-deployed engineers become valuable, because most business owners recognize the opportunity but lack the in-house capability to implement loops, agent orchestration, and tools like Claude Code themselves. Anthropic, OpenAI, and Glean are already hiring aggressively for forward-deployed roles, and there is a clear opening for specialists who embed directly inside a business to operationalize AI. For investors, the key insight is that a loop is by definition inference compute, so the proliferation of always-on agents converts directly into demand for data centers, semiconductors, and power.
The throughline across the most compelling forward themes is that the largest opportunities are hiding in plain sight, visible to investors but underestimated because the growth is exponential rather than linear. AI infrastructure sits at the center of this view, with a credible case that the market will need ten to fifty times more AI data center capacity than exists today, pulling semiconductors and electricity along with it. AI application companies are a second underappreciated category, with names like Harvey in legal and Glean in enterprise search illustrating a segment most investors cannot yet name. Beyond software, humanoid robotics appears to be approaching an inflection point that could make it an obvious allocation within two to three years, while small modular fission reactors offer a longer-dated, proven-technology answer to data center power, distinct from fusion, which remains pre-commercial. Stablecoin and payments infrastructure, exemplified by Stripe's recent moves, looks strategically important even if the direct equity returns are harder to size, and genetic seed engineering of the kind David Friedberg's company - Ohalo - is pursuing points to a food technology frontier despite heavy regulatory friction. The longest-dated bet is a hardware shift away from screens toward augmented reality, eye-based contacts, earpieces, and ambient AI that travels with the user. Diversifying a pre-IPO portfolio across these distinct technology vectors, rather than concentrating in a single theme, is the prudent way to gain exposure to a future that is arriving faster than most models assume.