Analyst: Six AGI 'Neolabs' Are Implicitly Betting Against Superintelligence, With Compute 10× Below OpenAI
Summary
- • Six AGI-focused neolabs are implicitly betting LLMs will plateau, argues a veteran AI forecaster with data.
- • Discovery Loop, Jeff Dean's new lab just founded August 2026, enters the field with seed round still open.
- • Best neolab compute forecast (0.75 GW) is roughly 10× below OpenAI's ~10 GW Stargate buildout direction.
- • Three neolabs target frontier model status by 2029–2030; three others show median timelines stretching to 2035–2037.
Details
SSI is forecasted to raise $21B by end-2028 with a frontier model expected January 2029
Ilya Sutskever's SSI has raised ~$8B so far. Frontier model range is March 2027–June 2034. Has 7 senior big-3 hires. Compute median: 0.65 GW.
Thinking Machines leads all neolabs in compute at 0.75 GW and 23 senior big-3 hires
Mira Murati's lab has $2B+ raised; forecasted $15B by 2028; frontier model December 2030 (range: Feb 2028–2043). Highest compute and hire count of any neolab.
Reflection AI has the most senior hires (32) and targets frontier status by June 2029
~$4.6B raised; $13.5B by 2028. Open-source approach with 32 senior big-3 hires. Compute median: 0.21 GW.
Ineffable Intelligence (David Silver) has a frontier model median not until June 2035
$1.1B raised; $6.8B by 2028. Mysterious research agenda. 15 senior hires. Compute median: 0.28 GW.
AMI Labs (LeCun) bets against LLMs entirely with a frontier model median of December 2037
~$1B raised; $4.2B by 2028. Only 6 senior big-3 hires and the lowest compute median (0.04 GW)—reflecting a fundamentally different architectural bet.
Discovery Loop is the newest neolab, founded August 2026 by Jeff Dean with seed round still open
Seed round unclosed at publication. Forecasted $6.5B by 2028; frontier model June 2036. Compute median: 0.18 GW with 10 senior hires projected.
Best neolab compute (0.75 GW) is roughly 10× below OpenAI's Stargate buildout trajectory of ~10 GW
The compute gap between even the best-positioned neolabs and OpenAI's Stargate infrastructure is an order of magnitude—and widening as Stargate scales.
Best neolab capital median (~$15B by 2028) equals only one-sixth of OpenAI's latest fundraising round
Even the most capitalised neolab is a fraction of OpenAI's war chest, making direct compute competition implausible without a paradigm shift.
Neolabs can only win if the current AI scaling paradigm stalls or shifts to different architectures
The author argues the implicit investment thesis is that AI will be disruptive but not via near-term recursive self-improvement—a contrarian but internally coherent position.
Forecasts are probabilistic medians with wide confidence intervals. Compute = projected GW at end-2028. Capital = projected total raised by end-2028. Frontier model = top-5 on a recognized index or credibly established as frontier-class by independent assessment.
What This Means
A veteran AI forecaster argues the current wave of 'neolab' AGI startups—despite commanding billions in funding—are implicitly making a contrarian bet: that large language models will hit a ceiling before recursive self-improvement takes hold. This reveals a philosophical divide in the AI investment community, where VCs are simultaneously bullish on AI disruption while quietly skeptical of near-term superintelligence timelines. With compute and capital positions dwarfed by OpenAI and Anthropic, these neolabs' viability hinges on a world where the current scaling paradigm falters—a scenario most forecasters still consider unlikely in the near term.
Sources
- The Neolabs Are a Bet Against SuperintelligenceFuturesearch