Virtual Conference
When AI Rings the IPO Bell
Anthropic vs. OpenAI AI IPO Virtual Conference
Two frontier labs. Anticipated. $1.8 trillion awaiting the public markets.
The largest pure-play AI listings ever attempted
Six Sessions Across Two Days
Two virtual half-days of industry-anchored debate on whether public markets will validate the largest pure-play AI listings ever attempted.
Format: Virtual with live Q&A. Six sessions of 45 minutes across two mornings, with 5-minute breaks between sessions. Both days begin at 9:00 AM ET.
Tuesday, September 1
Starting at 9:00 AM ET · 3 sessionsThe AI IPO Supercycle & the Bubble Question
Against more than $700B of AI infrastructure spend, anticipation is building for the IPOs of Anthropic and OpenAI. On the heels of SpaceX's record-breaking, near-$2 trillion debut in June, the 2026-2027 period emerging as the most market-formative IPO window in history. This framing session tests whether public markets can absorb near-trillion-dollar debuts without forced repricing, setting the working macro view every subsequent session builds
- Whether the AI capex cycle reflects genuine end demand or self-reinforcing capital among a small set of counterparties
- Market capacity to absorb $200B+ of combined IPO proceeds in one window
- Stress scenarios: what breaks first if rates back up or enterprise AI budgets pause
- Defensibility of the ~$965B and ~$852B private marks under public scrutiny
- Index inclusion timelines and the scale of forced passive inflows post-listing
- Lessons from the SpaceX debut on pricing, allocation, and aftermarket behavior
- The signal a failed or repriced AI listing would send across private AI valuations
Too Powerful to Ship: Government Oversight & the Regulatory Unknown
The unpriceable session. Government has become the swing variable neither S-1 can model — Washington can restrict what ships, who is allowed to buy it, and which rivals get to compete. The Fable/Mythos export-control suspension pulled a flagship offline overnight; the Pentagon's first-ever "supply chain risk" designation of a US company cut Anthropic off from the federal market over a policy dispute, and is now in the courts; and any move on Chinese open-weight models reshapes the price war in a single stroke. With no settled regulatory regime, a single executive action can reprice these companies between filing and lockup.
- Government's power to restrict releases — what the Fable/Mythos export-control suspension proved about how fast and how completely a flagship can be pulled
- Pre-release review: the White House–Anthropic friction over shipping a powerful model, and whether frontier launches now run through Washington
- The Pentagon's "supply chain risk" designation of Anthropic — the first against a US company, the litigation, and what losing the federal and defense channel does to the equity story
- Chinese open-weight models (DeepSeek, Qwen) as a policy lever: would US restrictions on hosting or deploying them protect frontier pricing — a national-security win that doubles as a commercial subsidy?
- Export controls in both directions — chip controls constraining compute supply, model-export controls constraining which markets the labs can sell into
- The emerging regime — deployment licensing, capability thresholds, the EU AI Act — and whether oversight entrenches incumbents or caps their upside
- The unpriceable tail: underwriting equity when policy is unsettled and reversible, and what these S-1 risk factors will actually have to disclose
How Durable Is the Frontier Premium?
Low-cost open-source and equal-weight models, including Llama, GLM, DeepSeek, and Qwen, have closed the capability gap on inference, and third-party hosts serve them at a fraction of frontier API prices. What remains of Anthropic's and OpenAI's price leverage sits in three places: frontier-tier releases, reasoning models, and enterprise contracts.
- The current price gap between frontier APIs and hosted open-weight equivalents at comparable benchmark scores
- Which of the three defensible zones carries pricing power today and which is already eroding
- Share of current revenue sitting in commodity inference an open-weight host can undercut immediately
- The blended revenue-per-token floor the labs must defend given committed capital
- What keeps enterprise buyers paying the premium after a bake-off against a cheaper open-weight equivalent
- The China vector: DeepSeek and Qwen as a distinct price-pressure lane versus Meta's strategic loss-leader
- Whether the premium has anywhere to retreat if open-weight quality reaches reasoning parity by 2027
Wednesday, September 2
Starting at 9:00 AM ET · 3 sessionsWhat Enterprises Are Actually Investing In
The primary-research session. CIOs, AI platform leads, and procurement heads report from inside live deployments: which model families won recent bake-offs, how renewal and expansion behavior is trending, and where agentic tools sit in real budgets. This is the demand-side ground truth for every valuation debate across the two days.
- Bake-off outcomes: which model families are winning enterprise evaluations and on what criteria
- Renewal, expansion, and churn behavior across frontier API contracts
- Claude Code versus Codex in actual engineering budgets and seat counts
- Whether agentic AI spend is incremental or cannibalizing existing software line items
- Multi-model architectures: how many enterprises are dual-sourcing and what that does to pricing
- Switching costs in practice: what it takes to move a production workload between model providers
- Procurement's view of the IPOs: does public-company status change vendor risk assessments
Compute, Chips and Outlook for Capex Commits
The supply-side session. Both equity stories rest on access to compute at a price that permits margin, and both companies depend on partners who are simultaneously suppliers, investors, and competitors. The panel examines Nvidia dependence and custom silicon timelines, hyperscaler cloud commitments as both lifeline and leash, and the financing structures behind the next wave of data-center buildout.
- Nvidia dependence today: allocation, pricing leverage, and the realistic timeline for custom silicon relief
- The Microsoft, Amazon, and Google commitments: contract terms, exclusivity, and exit costs
- Data-center buildout economics: power, land, and lead times as the real constraints
- Financing structures behind the next $200B: debt capacity, vendor financing, and equity dilution
- The related-party problem: pricing compute fairly when your supplier is also your investor
- Neocloud capacity as an escape valve: can CoreWeave-class providers change the labs' negotiating position
- What a compute supply shock, in either direction, does to each company's margin path
What Do You Want to Own for the Long Term?
The panel debates where a long-term investment is best deployed. Anthropic is one option among many, and its listing will set a benchmark multiple for AI. But the same exposure is available today through the hyperscalers that hold AI Lab equity, through semi providers, through power and data centers, through the data layer vendors, and through the software companies that package intelligence into products people actually buy. Panelists argue for a layer and defend it.
- Which layer of the stack keeps the profit: where are there scarce assets and defensible market positions, and where will commoditization drive down pricing?
- Owning the labs without buying the labs: Microsoft, Alphabet, and Amazon as lab exposure wrapped in profitable businesses at a fraction of the multiple
- Anthropic's market position versus OpenAI, XAI, and the open-weight models. Anthropic is the de facto market leader today, but how difficult is it to displace and how likely is it that it will maintain its leadership position for many years?
- What stays scarce and what gets cheap: power, packaging, and memory against model weights, inference, and applications
- The buyer's perspective: which of these vendors would an enterprise commit to as a five-year dependency, and what would make it hedge
- How will the AI Labs grow, and will they displace the software companies that are leveraging the service?
- The duration mismatch: models that turn over in a year, GPUs on depreciation schedules people disagree about, and power contracts that run twenty
- Polls: where the long-term dollar goes across labs, hyperscalers, silicon, power, data, applications, or none of the above