Bitcoin (BTC) Steady Near $86K as Alibaba Cloud Doubles Down on AI Compute

Alibaba Cloud Sets Its Sights on 20 GW

Alibaba set one of the most aggressive infrastructure targets of the current AI buildout at its annual Yunqi conference: Alibaba Cloud is to exceed 20 GW of global data center capacity by 2032. The announcement arrived with new silicon attached — T-Head, Alibaba's chip subsidiary, unveiled the Zhenwu V900 AI accelerator, slated for mass production and commercial release in the first quarter of 2027. Per the company's own disclosure, the V900 delivers roughly three times the performance of its predecessor, the M890, and scales to clusters of up to 500,000 chips for training and running large AI models. Chief executive Eddie Wu said AI customer demand is strong and pushing cloud revenue higher, though supply-chain constraints still cap how fast compute capacity can be added. Alibaba's own framing is blunt: a target above 20 GW only becomes operating capacity when electricity, construction and funding land in sync.

The financing side is tightening. SB Energy, the SoftBank-backed US data center developer, has slowed its IPO despite seeking a valuation near $50 billion, according to WuBlockchain's daily briefing; the company has no operating data centers yet, and a potential debt raise of close to $5 billion has drawn weak investor interest. Power is the second gate: Georgia Power signed Google to a nuclear uprate agreement expected to add about 96 MW at the Vogtle and Hatch plants by upgrading turbines, pumps, motors and cooling systems, pending approval by the Georgia Public Service Commission. California's governor, meanwhile, signed seven bills requiring data centers to shoulder grid and water upgrade costs rather than pass them to other customers.

Huang's Per-Gigawatt Price Tag

NVIDIA CEO Jensen Huang put a number on the cost curve behind these announcements: building one gigawatt of AI infrastructure requires roughly $50 billion to $60 billion. The estimate came at the G20 innovation ministerial in Chapel Hill, North Carolina, during a conversation moderated by US Commerce Secretary Howard Lutnick, with senior technology executives in attendance. Footage published by CNBC and in NVIDIA's official event recording shows Huang stating the same range, though no complete official transcript — and no G20-side validation of the figure — has surfaced.

The G20 joint statement limited itself to cooperation themes: innovation policy, public-sector AI adoption, workforce development, AI intellectual property, standards and supply-chain investment. The accompanying Carolina Emerging Technologies Principles recommend basic-research investment and commercialization but are explicitly non-binding. The documents contain no per-gigawatt cost standard, no member-state investment totals and no equipment purchase commitments.

Huang's figure fits the AI factory framing: a gigawatt-scale facility is a full industrial system — high-performance chips, memory and networking from suppliers such as Western Digital, plus land, power and data-center plant — rather than a server order. GPUs handle training and serving workloads, but a facility is complete only with power supply, cooling, network, servers and building systems in place. Actual costs shift with power procurement, GPU configuration and financing terms. Some analysts read the $50-60 billion range as a valuation yardstick for AI cloud operators; NVIDIA and the G20 have issued no official projection, and any national buildout still hinges on follow-up policy and project-level budgets. Readers tracking the market in real time can follow live spot and futures prices on Bybit.

Power, Capital and Bitcoin Near $86K

The through-line across these items is that compute ambitions now face two hard gates: capital and power. A stalled IPO for a pre-revenue developer shows public markets discounting announced capacity, while the Georgia uprate and California's cost-allocation laws price electricity and water directly into project budgets — pressures reaching the broader energy sector ETF complex and AI-adjacent equities, from compute suppliers to software names like Adobe. Bitcoin (BTC), the most liquid proxy for the same risk-liquidity cycle that funds AI capex, holds near $86,000 at the time of writing, per live spot data. Our desk's read: announced gigawatts convert into returns only where power, financing and contracted demand arrive together — the same discipline crypto markets apply to any treasury.