SpaceX's record-breaking IPO was a watershed moment for capital markets. The offering became the reference case for how investors price a business with a complex, multi-layered thesis tied to the AI theme.
I recently joined an expert webinar hosted by Ted Seides, founder of Capital Allocators, to talk about SpaceX’s debut in the context of the broader IPO market. During the discussion, Marcelo Ballvé, head of research at Sacra, explored what’s behind the mega-cap tech IPO surge, what’s ahead for the upcoming slate of potential offerings, and what’s moving tech markets in 2026.
Watch the full webinar here.
SpaceX and the Reopening IPO Window
Because so many venture players held equity in SpaceX, the company’s IPO unfroze liquidity across Silicon Valley and gave Wall Street a working blueprint for valuing businesses that sell compute capacity, inference, and foundation models. SpaceX somewhat justified the Street’s optimism as it generated strong operational results in its first quarter as a public company, despite volatile stock performance since its IPO pricing.
Yet the broader deal environment paints a more mixed picture. Market liquidity remains unevenly distributed: Even as venture fundraising and deal-making have picked up after a rocky first half of the year, that capital is concentrated in a small handful of mega-deals rather than spread across the broader ecosystem. Company research shows that 43% of all venture capital in H1 2026 went toward just two names, OpenAI and Anthropic. Meanwhile, financing rounds for smaller, seed-stage companies have actually declined.
Amid this backdrop, late-stage private companies are exhibiting more caution before going public. Ballvé pointed out that these IPO hopefuls are adopting the habits of a public company to ensure a smooth eventual transition to public markets.
There definitely are more and more private companies that do a kind of baby version of being a public company before they go public [by externalizing] what they're planning, roughly where they are in terms of top-line operating metrics, and…letting people know where they are at least on an annual cadence…I think that’s where the difference in being a private company and a public company, though still really dramatic, [is] getting a little smoothed over.
Frontier vs. Open-Source Economics
Enterprise AI is transitioning from “tokenmaxxing” (focusing on volume) to “valuemaxxing” (focusing on ROI). Because agents rely heavily on input tokens, which are far cheaper to serve than output tokens, inference margins improve as agents scale. This dynamic means the rise of agentic AI workflows is margin-accretive for OpenAI and Anthropic.
None of that changes the sheer scale of what it costs to stay at the frontier, Ballvé pointed out.
When you get into the cost structure overall, and how much money these companies are burning, I think that would be my biggest question. I think OpenAI is burning something like $10 billion annually, and Anthropic, half of that. Those numbers are just massive, and it's going to raise a question…for anyone thinking about these companies.
Price cuts, like those for OpenAI's Luna model, help the revenue side of the equation. Lower prices provide just enough token consumption to support revenue generation for these firms. Yet smaller “token resellers” with razor-thin margins risk scaling their way straight into bankruptcy if they don't own the underlying workflow, Ballvé said.
Foundation Model Debate
At the bottom of the AI stack, a debate has grown around foundation models: closed source models such as OpenAI and Anthropic versus open-source alternatives. The rising crop of open-source alternatives is considered a potential disruptive threat to the more premium frontier models.
Can the two tiers coexist? Ballvé sees a place for both open-source and frontier models, and believes “OpenAI and Anthropic will continue commanding a premium as long as they can stay six, three, [however many] months ahead of the open-source models,” while open-source will win on efficiency. Early evidence suggests he’s right, as the frontier labs are still growing rapidly: Anthropic’s annualized revenue has surged into the billions, and OpenAI’s has more than tripled year over year.
Databricks and Stripe: Next in Line?
While frontier models are still proving their ROI, profitability is accruing elsewhere in the AI stack. Cloud data is one example: Databricks, the leading company in the space, now has more than $7 billion in run-rate revenue, with year-over-year growth accelerating to 80%. In payments, Stripe recently expanded its own AI footprint through a $7+ billion acquisition of OpenRouter and is building toward stablecoin-enabled micropayments. Both companies have enjoyed strong momentum and are building toward their own public-market debuts.
Some potential headwinds remain. Amid concern that a crowded slate of high-profile offerings could dampen investor enthusiasm, Databricks CEO Ali Ghodsi recently indicated that the company will likely wait until 2027 to go public. The company faces competition from developer-first tools like Supabase, a favorite of “vibe coders,” Ballvé pointed out. There are also open questions surrounding how well Databricks can integrate its recent acquisitions, and whether its small sample size of positive cash flow holds up over time.
Physical AI: The Next Frontier?
Investment in physical AI and robotics has surged, with major moves such as SoftBank’s acquisition of ABB’s robotics unit driving up valuations across the AI ecosystem. But deployment is proving slower than bulls had hoped for, with full-scale enterprise adoption of agentic AI still in the single digits as companies have struggled with integrating systems into their existing workflows.
I think a lot of venture capital is hoping that physical AI is where the next big Anthropic- [or] OpenAI-shaped opportunity will come out of, just because the TAM is so big for automating real world processes. But I think it's really further off than people assume.
Physical supply constraints are another barrier to broader implementation. The biggest immediate supply constraint is in hardware. High-bandwidth memory (HBM) and optical networking are considered the tightest bottlenecks right now, with Micron recently reporting that it’s sold out of memory capacity through 2028. To adapt, Nvidia is moving beyond its traditional hyperscaler customers to fund AI-native companies and sovereigns directly.
Looking Ahead
SpaceX has given investors an initial framework to evaluate a complex company tied to the AI thesis. This is the exact type of transaction that we are likely to see in Anthropic, Databricks, Stripe, and other mega-cap tech names.
Yet in this top-heavy capital environment, the math is much more complicated for smaller potential offerings. Because AI ROI remains an open question at the enterprise level, investors will be looking for signs of strong retention and pricing power for any company going public.
But which parts of the AI stack are likely to generate the most ROI? As in previous technology cycles, AI profits should eventually accrue at the application layer, Ballvé said. But he acknowledged it may take some time for that to happen.
Historically, profits pool at the app layer. That's where the most value can be added, and I don't think that'll be different here in the long term. But we're still kind of early in the cycle, and I think we'll see that profit bump migrate up the stack as we move through it.
Track Mega-Cap Tech Unicorns Through the IPO
The technology sector moves at a relentless pace, making it increasingly difficult to track which IPO candidates are gaining momentum. Staying ahead requires getting from input to insight quickly. AlphaSense is designed to ensure users never miss a key management call or data release with Tegus Expert Insights, which leverages the industry’s most extensive content library containing over 300,000+ expert interviews.
AlphaSense is constantly investing in its search stack and uses an LLM-agnostic approach to ensure outputs are richly sourced and grounded in the proper context. With our premium, proprietary content sets, you can better evaluate new opportunities, navigate market complexity, and generate decision-grade outputs.
Discover how AlphaSense can help you uncover alpha with confidence. Start your free trial today.
