Flexible data centers can ease, not erase, power limits
- Flexible workloads could free existing grid capacity, but Pat Gelsinger warns electricity supply still threatens data center projects.
- The two shows describe complementary tools: use power more flexibly now, while building the capacity AI demand will need.
AI data centers may be able to make better use of the grid they have - but flexibility is not the same as new power. In its Oct 8, 2026, coverage, This Week in AI argued that software can shift or briefly pause workloads to ease peak demand. The next day, the a16z Show featured Pat Gelsinger warning that slow-growing energy supply could leave planned data centers without enough electricity to operate.
Varun Sivaram of Emerald AI, speaking on This Week in AI, said studies by Duke University and McKinsey estimate that flexible data-center use could unlock 100 gigawatts of U.S. grid capacity, with curtailment needed for less than 2% of the year. The show also described trials in London and Phoenix, where automated load reductions were used to respond to grid needs. Those claims point to a way to relieve bottlenecks without building lines for rare peaks.
Gelsinger’s warning, on the a16z Show’s Oct 9, 2026, episode, is about the wider supply problem. He said energy capacity has grown barely 4% annually while data-center demand surges, and predicted defaults on construction projects if utility power arrives too late. Faster chip design cannot solve the lag: the episode said fabrication and packaging take months after a design is ready.
The Frontier had already covered both sides of this squeeze: its Oct. 9 story examined workload throttling, and on Oct. 12 it reported Gelsinger’s warning that grid delays could put projects at risk. Taken together, the shows suggest flexibility could make existing capacity go further, while Gelsinger’s forecast underlines that it cannot ensure enough electricity reaches every planned site. The reports do not establish how much of the estimated capacity is available where and when new data centers need it.
The practical question is therefore not whether data centers can adjust - they can, according to the trials described by This Week in AI - but whether utilities and operators can coordinate that flexibility at scale while power supply expands. The shows offer a near-term tool, not proof that the larger shortage is solved.
Source Intelligence
- Deep dive into what was said in the episodes
Chips, Memory, and Power | Pat Gelsinger • Oct 9
- Pat Gelsinger warns that fixed energy grids will cap economic growth in an AI economy. Pat Gelsinger predicts mounting default rates on data center construction projects because utility power capacity cannot meet planned computing demand.
- AI design tools allow engineers to complete complex logic architectures in three months, but physical fabrication and 3D packaging add a nine month delay. Pat Gelsinger warns this delay risks rendering targeted chip designs obsolete before deployment.
Also discussed on this episode: (6)
Chips (2)
- Engineering the Intel 486 required Pat Gelsinger and Intel engineers to build early design automation tools. Intel created its own hardware description language and collaborated with UC Berkeley to pioneer automated placement and routing.
- Pat Gelsinger argues memory innovation will finally occur after 30 stagnant years, driven by AI memory bandwidth constraints. Memory chip manufacturers added $2.5 trillion in market capitalization over four years, fueling new venture investments in 3D stacked logic and ferroelectric materials.
Startups (1)
- Pat Gelsinger expects the market containing over 100 specialized AI accelerator startups to consolidate rapidly. Evolving reasoning workloads, high capital requirements, and buyer selection by major labs like OpenAI will force hardware convergence toward standard platforms.
AI Infrastructure (2)
- Pat Gelsinger anticipates broad commercial adoption of co-packaged optics by 2028 or 2029 as copper interconnects reach physical limits. Moving to optical switching and predictable AI traffic flows will erase traditional boundaries between scale-up and scale-out cluster architectures.
- Stalled power efficiency in silicon across five GPU generations is pushing data center operators toward radical power delivery redesigns. Pat Gelsinger highlights adoption of 800-volt DC architectures, vertical GAN converters, and superconducting chips to handle severe thermal limits.
Agents (1)
- Pat Gelsinger and Guido Appenzeller argue that traditional virtualization infrastructure built at VMware must be redesigned for autonomous AI agents. Management software must adapt to serve agent security, policy guardrails, and migration rather than human operators.
Data Centers Aren't Killing the Grid. They Could SAVE It. • Oct 8
- Varun Sivaram argues that data centers must become flexible grid assets to accommodate rising energy demands. By dynamically throttling workloads or shifting compute to other regions during peak demand, data centers can integrate into the existing grid without raising community rates.
- The US electric grid lacks a standardized digital communication system across its 3,000 distinct utilities. Many operators still call data centers on the phone to request power reductions during peak load emergencies.
- Emerald AI demonstrated dynamic power throttling in London by coordinating with the national grid to reduce data center power during a sports halftime. The halftime pause mitigated a sudden one-gigawatt power spike caused by citizens turning on electric tea kettles.
- Studies from Duke University and McKinsey show the US grid holds 100 gigawatts of available power that could be unlocked with minimal curtailment. Data centers would only need to reduce power usage for less than two percent of the year.
- The United States aims to add 70 gigawatts of AI data center capacity by 2028 but is projected to deploy only 30 gigawatts. Meanwhile, China added 500 gigawatts of power capacity last year compared to 50 gigawatts in the United States.
- Varun Sivaram projects China will possess 400 gigawatts of peak spare capacity by 2030 to fuel its artificial intelligence infrastructure. This growth is driven by massive state investments in nuclear, solar, wind, coal, and battery power.
- The United States power grid operates at an average utilization rate of 50 percent or less. Increasing overall grid utilization by a stretch target of 10 percent would unlock significant capacity without building new transmission lines.
- Emerald AI is partnering with NVIDIA and Digital Realty to launch a 100-megawatt flexible data center in Virginia. The project will serve as a reference design for future power-flexible AI factories.
- Texas Governor Greg Abbott has paused the state's batch zero process for connecting new data centers to the grid. Varun Sivaram notes this pause is likely to persist through the midterm election season.
Also discussed on this episode: (5)
Chips (1)
- NVIDIA hardware efficiency has increased exponentially over the last decade. The Rubin GPU generation produces one million times more tokens per watt of input energy than the Kepler generation released in 2012.
Agents (1)
- Vibhav Viswanathan co-founded Pascal AI to build context-aware AI agents for hedge funds and asset managers. The startup's current customer base represents an accumulated half-billion dollars in assets under management.
Models (1)
- Pascal AI integrates proprietary internal data, licensed market feeds, and web data to analyze investment theses. Viswanathan explains that accessing a fund's internal memos and portfolio history provides a proprietary edge that frontier AI models cannot replicate.
Enterprise (1)
- Pascal AI sells its software to secretive financial institutions, including a top-ten global hedge fund and a 40-billion-dollar Tiger Cup fund. These institutions use the platform to maintain a real-time, context-aware view of their portfolios.
Labor (1)
- Vibhav Viswanathan claims AI adoption in hedge funds is shifting analyst roles toward manual, boots-on-the-ground research. Because software automates document analysis, analysts are traveling more and conducting physical channel checks to discover investment alpha.