The Conclave — Gallery, page 2

Who area the most bill of material, pure-play suppliers for on-chip direct-to-silicon cooling?

No mainstream, pure-play, bill-of-materials supplier currently exists for on-chip direct-to-silicon cooling; the closest BOM-level suppliers in the adjacent direct-to-chip market are CoolIT Systems, ZutaCore, and Accelsius.

Panel:GPT-5, Claude, Gemini, and 3 more

Provide a best in class estimate of the EBITDA and EBITDA margins for TMT Finance (https://www.tmtfinance.com/), decomposed by top line revenue, operating expenses with technology expense, headcount, depreciation, and other expenses broken out.

No verifiable, fact-based EBITDA estimate or decomposed P&L can be produced for TMT Finance from public information.

Panel:Magistral, Grok, DeepSeek, and 3 more

Will AI be able to write material amounts of complex cross compiler code in the next three years?

YES — AI will write material amounts of complex cross-compiler code for mainstream ISAs within three years, but strictly as a supervised tool relying on dense oracle verification rather than autonomous authorship.

Panel:Claude, Grok, GPT-5, and 3 more

How is the CIA's analytical organization organized from a taxonomy point of view? In other words, I assume they have the Asia desk and the EMEA desk or whatever, and within those there are specialists in certain countries. But then there may be other specialists that focus on things that are more horizontal in nature, like terrorism or natural resources or something along those lines. I'm curious to learn how they scaffold all of this, both organizationally and from a data point of view. They must have a current file on Uzbekistan, for example. How is that file maintained? What's in it? Where does it source from? How is it updated? How is it revised with accurate information that replaces what was inaccurate information, etc.?

The CIA’s analytical organization is a partial matrix where the Directorate of Analysis organizes analysts by region/country and functional specialties, overlaid by Mission Centers that integrate analysis with operations.

Panel:Magistral, GPT-5, DeepSeek, and 3 more

Provide a best in class estimate of the EBITDA and EBITDA margins for techinsights.com, decomposed by top line revenue, operating expenses with technology expense, headcount, depreciation, and other expenses broken out.

TechInsights generates approximately $120 million in revenue with an EBITDA margin of 20–25%, resulting in an EBITDA of $27–30 million.

Panel:Gemini, GPT-5, Magistral, and 3 more

Is Oracle really still a software company in its classic, historical sense, or has it morphed into something far more complex, and if so, what?

NO—Oracle is no longer a classic software company. It has morphed into a capital-intensive AI-compute infrastructure operator (effectively an "AI landlord") whose growth, valuation, and risk profile are now dictated by GPU-heavy dat…

Panel:Claude, Gemini, Magistral, and 3 more

Is there a material difference in supported use cases between the following: using Claude code with an MCP to Snowflake vs just using Snowflake Cortex?

Yes, there is a material difference, but it is strictly non-functional: the distinction lies in inference location (governance boundary) and default user persona, not in functional capability.

Panel:DeepSeek, Magistral, Grok, and 3 more

Why hasn't Nvidia lost more share to AMD given that AMD is already way ahead in FP16 and far ahead in FP64 (but not FP32).

Nvidia has not lost more share to AMD because the market rewards realized throughput and system-level scaling rather than peak FP16/FP64 FLOPS.

Panel:GPT-5, Claude, Magistral, and 3 more

Provide a best in class estimate of the EBITDA and EBITDA margins for NewtonX (https://www.newtonx.com/), decomposed by top line revenue, operating expenses with technology expense, headcount, depreciation, and other expenses broken out.

NewtonX is currently operating at a loss, with a best-estimate EBITDA margin of -7% to -10% on a revenue base of $35M–$38M.

Panel:Claude, Grok, Gemini, and 3 more

Provide a best in class estimate of the EBITDA and EBITDA margins for Valona Intelligence (valonaintelligence.com), decomposed by top line revenue, operating expenses with technology expense, headcount, depreciation, and other expenses broken out.

No credible numerical EBITDA or margin estimate can be produced for Valona Intelligence due to a complete absence of verifiable public financial data.

Panel:GPT-5, Claude, Magistral, and 3 more

Provide a best in class estimate of the EBITDA and EBITDA margins for LinqAlpha, decomposed by top line revenue, operating expenses with technology expense, headcount, depreciation, and other expenses broken out.

No credible, best-in-class company-specific estimate of LinqAlpha’s EBITDA, EBITDA margin, or decomposed P&L components is possible with publicly available information.

Panel:Gemini, Grok, GPT-5, and 3 more

Is "10x Compute" required for each doubling of model intelligence?

No. The claim that "10x compute" is required for each doubling of model intelligence is false; it is a capital budgeting heuristic, not a physical law.

Panel:DeepSeek, Gemini, Claude, and 3 more