Conference Presentation, Panel
Compute as Capital: Redefining AI's New Currency | WEKA, Akamai, TensorWave & More | RAISE 2026
- McKinsey projects approximately seven trillion dollars in AI infrastructure spending by 2030, with the sector anticipated to remain in early stages through at least 2029.
- Infrastructure expansion plans include scaling from 50 megawatts to gigawatts by the end of 2027, with a deployment requirement of 12 billion dollars in equity and debt over the next 12 to 18 months.
- White Fiber targets a six-month speed to market for data center construction, compared to the typical two years, utilizing a plan to convert old factories into Tier 3 facilities across North America.
- A predicted invention for White Fiber aims to link disparate locations within 83 kilometers via dark fiber to form a super virtual supercluster, with commercialization intended as a success marker.
- Sam Tabar identifies access to electricity and grid power as the primary constraint, requiring a minimum asset base of 50,000 chips and necessitating protection against future depreciation curves through insurance and residual value calculations.
- Sam Tabar predicts that near-term value remains inverted in the supply chain, shifting back to factory fabrication providers for DRAM, storage, GPUs, and CPUs around 2030 to 2031.
- Sam Tabar notes that developing reliable large-scale infrastructure takes decades, while predicting that power will remain the most sought-after commodity in the medium and long term.
- Val Bercovici anticipates tokens serving as the industry's ultimate currency due to transparency and diverse offerings, requiring continued guaranteed contracts to stabilize the spot-based market.
- Val Bercovici expects domain-specific models to emerge within two to three years due to economic and safety necessities, predicting the industry will view token maxing as obsolete practices by then.
- Val Bercovici suggests that banking participation could lower the industry's cost of capital, while engineering efficiency in the memory wall could generate $10 of value for every $1 of compute capital.
- Carmen Lee plans to launch future options at CME in the coming fall to hedge compute volatility, projecting three trillion dollars in diverted trading volume if the spot market cap of 300 billion grows tenfold.
- Carmen Lee forecasts a healthy compute index market contingent on a mix of short and long positions, noting that a flattened term structure for longer-term contracts indicates a current market equilibrium.
- Peter Tomasik forecasts powerful open-source models within three months, with Chinese export controls expected to cause a readjustment period regarding open model availability.
- Carmen Lee warns that restricting US data center access while limiting European, Southeast Asian, or South American access may harm the industry's overall economy.
- Risks include power availability constraints caused by community resistance and data center moratoriums, as well as potential scarcity of resources tripping the industry in the near term.
- Industry maturity is expected by the turn of the decade, with the panel host predicting the AI bubble narrative will be dismissed by the Ray Summit in 2029.
- Yotor and the panel host expect financiers to eventually recognize GPU residual value, criticizing current underwriting practices that focus on short time periods.
- Long-term success is viewed as structurally sound despite cyclical market gyrations and varying company valuations, driven by the need for proven, scalable, and repeatable ROI use cases.
- Critical success factors include the continued performance of capital markets with banks offering small spreads over SOFR and the proliferation of tokens as the ultimate indicator of industry health.