Quant AiQ

Quant AiQ

Utilities

Energy AI at scale

About us

Industry
Utilities
Company size
2-10 employees
Type
Self-Owned

Updates

  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    AI Data Centers: Have we thought this through? The Grid Can't Deliver at this speed. 𝗧𝗵𝗲 𝗨𝗦 𝗶𝘀 𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝘁𝗼 𝗳𝗮𝗰𝗲 𝗮𝗻 𝟴𝟬𝗚𝗪 𝗳𝗶𝗿𝗺 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝘀𝗵𝗼𝗿𝘁𝗳𝗮𝗹𝗹 𝗮𝘁 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝘁𝗿𝗮𝗷𝗲𝗰𝘁𝗼𝗿𝘆. Boston Consulting Group (BCG) Big tech and "The Grid" evolve at different speeds. 𝗧𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝘀𝗲𝗰𝘁𝗼𝗿 𝗱𝗲𝗮𝗹𝘀 𝗶𝗻 𝗿𝗲𝗮𝗹 𝗮𝘀𝘀𝗲𝘁𝘀, 𝗽𝗵𝘆𝘀𝗶𝗰𝘀, 𝗮𝗻𝗱 𝗲𝗹𝗲𝗰𝘁𝗿𝗼𝗻𝘀—𝘀𝗹𝗼𝘄 𝗮𝗻𝗱 𝗵𝗲𝗮𝘃𝗶𝗹𝘆 𝗿𝗲𝗴𝘂𝗹𝗮𝘁𝗲𝗱. 𝗧𝗲𝗰𝗵 𝗰𝘆𝗰𝗹𝗲𝘀 𝗱𝗲𝗮𝗹 𝗶𝗻 𝗯𝗶𝘁𝘀 (𝟭𝘀 𝗮𝗻𝗱 𝟬𝘀). 𝗜𝘁 𝗶𝘀 𝗱𝗶𝘀𝗿𝘂𝗽𝘁𝗶𝘃𝗲 𝗮𝗻𝗱 𝗳𝗮𝘀𝘁. Consider this: Microsoft, Google, and Meta Facebook CAPEX surpassed $100 billion in 2023, driven largely by data center infrastructure. 𝗧𝗵𝗲 𝗹𝗮𝘀𝘁 𝘁𝗶𝗺𝗲 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝗴𝗿𝗶𝗱 𝘀𝗮𝘄 𝘁𝗵𝗶𝘀 𝘁𝘆𝗽𝗲 𝗼𝗳 𝗶𝗻𝘃𝗲𝘀𝘁𝗺𝗲𝗻𝘁 𝘄𝗮𝘀 𝗱𝘂𝗿𝗶𝗻𝗴 𝗪𝗼𝗿𝗹𝗱 𝗪𝗮𝗿 𝗜𝗜. 𝗪𝗵𝘆 𝗶𝘀 𝗶𝘁 𝘀𝗼 𝗵𝗮𝗿𝗱? Building new electricity infrastructure—generation, poles, wires, transformers—requires significant capital and can take 10 to 20 years. 𝟭. 𝗔𝗜 𝗗𝗮𝘁𝗮 𝗖𝗲𝗻𝘁𝗲𝗿𝘀 𝗮𝗿𝗲 𝘀𝗽𝗲𝗰𝗶𝗮𝗹: • The nature of AI computation means instantaneous power demands. • 𝗔𝗜 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗰𝗮𝗻 𝗿𝗲𝗮𝗰𝗵 𝗳𝘂𝗹𝗹 𝗽𝗼𝘄𝗲𝗿 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗮𝗻𝘆𝘁𝗶𝗺𝗲 - 𝗟𝗹𝗮𝗺𝗮 𝟯 𝘁𝗼𝗼𝗸 𝟱𝟰 𝗱𝗮𝘆𝘀 𝘄𝗶𝘁𝗵 𝗶𝗻𝘁𝗲𝗿𝗿𝘂𝗽𝘁𝗶𝗼𝗻𝘀. • Demand response and Peak Shaving models can only help so much. 𝟮. 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗻𝗴 𝘁𝗼 𝗮 𝗿𝗲𝗴𝗶𝗼𝗻𝗮𝗹 𝘁𝗿𝗮𝗻𝘀𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗴𝗿𝗶𝗱 𝗰𝗮𝗻 𝘁𝗮𝗸𝗲 𝘆𝗲𝗮𝗿𝘀: • In ERCOT (Texas), a new resource interconnection takes an average of 3.5 years. • In other regions, it can take six years or more.. • ERCOT uses a “connect and manage” system—other regions don’t. 𝟯. 𝗧𝗵𝗲 𝘀𝘂𝗽𝗽𝗹𝘆 𝗰𝗵𝗮𝗶𝗻 𝗶𝘀𝗻'𝘁 𝗷𝘂𝘀𝘁 𝘁𝗶𝗴𝗵𝘁—𝗶𝘁'𝘀 𝗰𝗵𝗼𝗸𝗶𝗻𝗴: • Manufacturers report lead times as high as 4 years for substation power and generator step-up transformers Wood Mackenzie Power & Renewables. • High-voltage circuit breakers average lead times of 3 years to 5 years - a 130% increase year-over-year. • Medium-voltage switchgear is also impacted, with lead times more than doubling. 𝟰. 𝗠𝗼𝗿𝗲 𝗖𝗼𝗻𝘀𝘁𝗿𝗮𝗶𝗻𝘁𝘀 𝗛𝗼𝗹𝗱𝗶𝗻𝗴 𝗨𝘀 𝗕𝗮𝗰𝗸: • 𝗧𝗿𝗮𝗻𝘀𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗖𝗮𝗽𝗮𝗰𝗶𝘁𝘆: There’s a lack of available transmission capacity to accommodate new energy resources. • 𝗥𝗲𝘁𝗶𝗿𝗶𝗻𝗴 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗼𝗿𝘀: Over 100 GW of aging generation capacity is set to retire in the next decade U.S. Energy Information Administration 𝗪𝗵𝗶𝗹𝗲 𝘁𝗵𝗲 𝘁𝗲𝗰𝗵 𝗴𝗶𝗮𝗻𝘁𝘀 𝗮𝗿𝗲 𝗺𝗼𝘃𝗶𝗻𝗴 𝗳𝗮𝘀𝘁, 𝘁𝗵𝗲 𝗲𝗹𝗲𝗰𝘁𝗿𝗶𝗰𝗶𝘁𝘆 𝘀𝗲𝗰𝘁𝗼𝗿—𝗴𝗿𝗼𝘂𝗻𝗱𝗲𝗱 𝗶𝗻 𝗽𝗵𝘆𝘀𝗶𝗰𝗮𝗹 𝗿𝗲𝗮𝗹𝗶𝘁𝘆—𝗶𝘀 𝘀𝘁𝗿𝘂𝗴𝗴𝗹𝗶𝗻𝗴 𝘁𝗼 𝗸𝗲𝗲𝗽 𝘂𝗽. 𝗦𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝗵𝗮𝘀 𝘁𝗼 𝗴𝗶𝘃𝗲. What did I miss? #datacenter #ai #energytransition Arcus Power Corp Boston Consulting Group (BCG)

    • No alternative text description for this image
    • No alternative text description for this image
  • Quant AiQ reposted this

    We’re excited to introduce our upcoming guest speaker: Wish Bakshi, an expert in AI Systems Engineering who has made significant contributions to the energy and utilities industry. Wish has applied 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 to optimize 𝘁𝗵𝗲𝗿𝗺𝗮𝗹 𝗽𝗼𝘄𝗲𝗿 𝗽𝗹𝗮𝗻𝘁 operations, improve 𝗲𝗻𝗲𝗿𝗴𝘆 𝘀𝘁𝗼𝗿𝗮𝗴𝗲 𝘀𝘆𝘀𝘁𝗲𝗺𝘀, and enhance the performance of 𝘀𝗼𝗹𝗮𝗿 and 𝘄𝗶𝗻𝗱 assets. His work has helped some of the largest companies in energy and trading achieve measurable results. In addition, Wish has developed 𝗔𝗜 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝗳𝗼𝗿 𝗻𝗲𝗮𝗿 𝗿𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝗰𝗼𝗺𝗺𝗼𝗱𝗶𝘁𝗶𝗲𝘀 𝘁𝗿𝗮𝗱𝗶𝗻𝗴, which have reshaped how the industry approaches efficiency and decision-making in commodities markets. Throughout his career, Wish has combined business insight with technological expertise, ensuring that machine learning is not just a theoretical tool but a practical driver of profitability and operational efficiency. In his upcoming talk, "𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗜 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝗮𝘁 𝗦𝗰𝗮𝗹𝗲", Wish will share 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝗲𝘅𝗮𝗺𝗽𝗹𝗲𝘀 he has implemented, highlighting: • Why many enterprises face obstacles with AI ROI • How to transition from concept to delivering solutions at scale • Practical lessons and insights from his extensive experience 📅 𝐄𝐯𝐞𝐧𝐭 𝐃𝐚𝐭𝐞𝐬: February 27 to March 1, 2025 📍 𝐋𝐨𝐜𝐚𝐭𝐢𝐨𝐧: BMO Centre, Calgary Secure your spot at YYC DataCon today! Grab your tickets before they sell out: https://lnkd.in/gMB_Sxw6 Don’t miss this chance to learn from a leader who has successfully integrated AI into complex, real-world systems.

    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    Had a wicked time presenting on "AI Applications in Energy" at Plains Midstream Canada. During the session, I shared insights on how to design, deploy, and scale AI systems for Commodities Trading, Midstream assets, and Combined Cycle Power Plants—showcasing real-world examples from systems I've architected. A huge shout-out to Conner Schiissler for flawlessly organizing the event and to Dan Reinbold; you have a rockstar team! Thanks to the attendees for an engaging Q&A that brought out some great discussions! #yycdata #ai #machinelearning #energy #utilities

    • No alternative text description for this image
    • No alternative text description for this image
    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    Is Alberta risking it all for Data Centers? The data will surprise you. AI data centers require highly reliable power sources - specifically, five-9s reliability. Natural gas is likely to be the preferred source for this level of reliability for AI data centers. 𝗧𝗵𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺: Alberta currently has about 13GW of total natural gas generation capacity. 𝗖𝗼𝗻𝘀𝗶𝗱𝗲𝗿 𝘁𝗵𝗶𝘀 𝘀𝗰𝗲𝗻𝗮𝗿𝗶𝗼: 𝗜𝗳 𝘁𝘄𝗼 𝟱𝟬𝟬 𝗠𝗪 𝗣𝗣𝗔𝘀 𝗮𝗿𝗲 𝘀𝗶𝗴𝗻𝗲𝗱 𝗳𝗼𝗿 𝗱𝗮𝘁𝗮 𝗰𝗲𝗻𝘁𝗲𝗿𝘀... ..𝘁𝗵𝗮𝘁 𝗶𝘀 𝟴% 𝗼𝗳 𝗱𝗶𝘀𝗽𝗮𝘁𝗰𝗵𝗮𝗯𝗹𝗲 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗔𝗘𝗦𝗢 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗳𝗼𝗿 𝗱𝗮𝘁𝗮 𝗰𝗲𝗻𝘁𝗲𝗿𝘀. AESO has at least six data-center applications in the early stages of development (that we know about). These would use approximately 5500 megawatts (MW) of power. 𝗧𝗵𝗮𝘁 𝗶𝘀 𝗿𝗼𝘂𝗴𝗵𝗹𝘆 42% 𝗼𝗳 𝗱𝗶𝘀𝗽𝗮𝘁𝗰𝗵𝗮𝗯𝗹𝗲 𝗴𝗲𝗻. Now, let's look at the current challenges already facing Alberta's grid: 𝗥𝗲𝗰𝗲𝗻𝘁 𝗰𝗹𝗼𝘀𝗲 𝗰𝗮𝗹𝗹𝘀 𝟭. 𝗢𝗻 𝗔𝗽𝗿𝗶𝗹 𝟯 𝗮𝗻𝗱 𝗔𝗽𝗿𝗶𝗹 𝟱, 𝟮𝟬𝟮𝟰, 𝗔𝗘𝗦𝗢 𝗶𝘀𝘀𝘂𝗲𝗱 𝘁𝘄𝗼 𝘀𝗲𝗽𝗮𝗿𝗮𝘁𝗲 𝗚𝗿𝗶𝗱 𝗔𝗹𝗲𝗿𝘁𝘀 𝗱𝘂𝗲 𝘁𝗼 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝘀𝗵𝗼𝗿𝘁𝗳𝗮𝗹𝗹𝘀: - High renewable energy during the day - Unexpected outage of thermal generation, leading to tight conditions during evening peak -Wind generation was 900 MW below forecast - A natural gas generator tripped offline, resulting in a 400 MW loss -Rotating outages affected multiple communities across Alberta 𝟮. 𝗔𝗻𝗼𝘁𝗵𝗲𝗿 𝗶𝗻𝗰𝗶𝗱𝗲𝗻𝘁 𝗼𝗰𝗰𝘂𝗿𝗿𝗲𝗱 𝗼𝗻 𝗝𝘂𝗹𝘆 𝟴𝘁𝗵 Calgary Herald: "unplanned thermal generation outage, high temperatures causing a reduction in some generator capability, and a line outage" caused the alert. 𝟯. 𝗧𝗵𝗲𝗻 𝘁𝗵𝗲𝗿𝗲 𝘄𝗮𝘀 𝘁𝗵𝗲 𝗶𝗻𝗳𝗮𝗺𝗼𝘂𝘀 𝗝𝗮𝗻𝘂𝗮𝗿𝘆 𝟭𝟯𝘁𝗵 𝗴𝗿𝗶𝗱 𝗮𝗹𝗲𝗿𝘁 Due to forced outages, low renewable energy generation, low imports, etc. 𝟰. 𝗔𝗹𝗯𝗲𝗿𝘁𝗮'𝘀 𝘃𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝘄𝗵𝗲𝗻 𝗶𝘀𝗹𝗮𝗻𝗱𝗲𝗱: From Sept 23 to Oct 3, AESO logged 27 significant frequency swings with the Alberta-BC intertie down. One dropped to 59.53 Hz when a large unit failed. 𝟱. 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗧𝗵𝗲𝗿𝗺𝗮𝗹 𝗽𝗹𝗮𝗻𝘁𝘀 𝘁𝗮𝗸𝗲𝘀 𝘁𝗶𝗺𝗲 Cascade took 4 years to build. 𝗧𝗵𝗲 𝗧𝗲𝘅𝗮𝘀 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵 𝗮𝘀 𝗮 𝗽𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻: They're telling data centers: bring your own power. ERCOT CEO has proposed data centers to build excess generation capacity and sell surplus to the grid. 𝗞𝗲𝘆 𝗾𝘂𝗼𝘁𝗲 𝗳𝗿𝗼𝗺 Energy Disruptors: TransAlta CEO summarized the current situation in Alberta: "Data centers want 99.9% reliability, and they want speed. At least in Alberta, we can't provide that in a green way right now." AESO It's a miracle that despite all these challenges, we have barely felt the pain. 𝗙𝗼𝗹𝗸𝘀 𝗮𝘁 𝗔𝗘𝗦𝗢 𝗮𝗿𝗲 𝗵𝗲𝗿𝗼𝗲𝘀 𝘄𝗵𝗼 𝗻𝗲𝗲𝗱 𝗺𝗼𝗿𝗲 𝗰𝗿𝗲𝗱𝗶𝘁 𝘁𝗵𝗮𝗻 𝘁𝗵𝗲𝘆 get. What did I miss? #aeso #datacenter #ai Arcus Power Corp

    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    Energy Transition: How Florida won "The Price is Right". The holy grail of energy transition has always been access to reliable, affordable, sustainable energy. Florida could be the first state to achieve all 3. 𝗦𝘂𝗻𝘀𝗵𝗶𝗻𝗲 𝗦𝘁𝗮𝘁𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝟭𝟬𝟭: - Florida is the third-largest electricity producer in the U.S., tied with California California ISO and only behind Texas ERCOT. - 𝟳𝟱% 𝗼𝗳 𝗶𝘁𝘀 𝗲𝗹𝗲𝗰𝘁𝗿𝗶𝗰𝗶𝘁𝘆 𝗰𝗼𝗺𝗲𝘀 𝗳𝗿𝗼𝗺 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗴𝗮𝘀 - 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗮𝗻𝘆 𝗼𝘁𝗵𝗲𝗿 𝗹𝗮𝗿𝗴𝗲 𝘀𝘁𝗮𝘁𝗲. - 𝗜𝘁'𝘀 𝗮𝗹𝘀𝗼 𝘁𝗵𝗲 𝟯𝗿𝗱 𝗳𝗮𝘀𝘁𝗲𝘀𝘁-𝗴𝗿𝗼𝘄𝗶𝗻𝗴 𝗺𝗮𝗿𝗸𝗲𝘁 𝗳𝗼𝗿 𝘀𝗼𝗹𝗮𝗿, 𝗷𝘂𝗺𝗽𝗶𝗻𝗴 𝗳𝗿𝗼𝗺 𝟭,𝟰𝟯𝟮 𝗠𝗪 𝗶𝗻 𝟮𝟬𝟭𝟴 𝘁𝗼 ~ 𝟭𝟳,𝟬𝟬𝟬+ 𝗠𝗪 𝗶𝗻 𝟮𝟬𝟮𝟰. 𝗧𝗵𝗲 𝗛𝗶𝘀𝘁𝗼𝗿𝘆: When the shale boom hit, Florida saw an opportunity. They prioritized natural gas. 𝗧𝗵𝗶𝘀 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻 𝘀𝗵𝗮𝗽𝗲𝗱 𝘁𝗵𝗲𝗶𝗿 𝗲𝗻𝗲𝗿𝗴𝘆 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲: - 𝟴 𝗼𝘂𝘁 𝗼𝗳 𝘁𝗵𝗲 𝘀𝘁𝗮𝘁𝗲'𝘀 𝟭𝟬 𝗹𝗮𝗿𝗴𝗲𝘀𝘁 𝗽𝗼𝘄𝗲𝗿 𝗽𝗹𝗮𝗻𝘁𝘀 (𝗯𝘆 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗮𝗻𝗱 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻) 𝗮𝗿𝗲 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗴𝗮𝘀-𝗳𝗶𝗿𝗲𝗱. - Florida leads the nation in generators that can switch between natural gas and fuel oil. - Nat gas plants provide consistent baseload power, which is crucial for grid stability and resilience (Spinning Reserves). 𝗪𝗵𝗲𝗿𝗲 𝗱𝗼𝗲𝘀 𝗙𝗹𝗼𝗿𝗶𝗱𝗮 𝘀𝘁𝗮𝗰𝗸 𝘂𝗽 𝗶𝗻 𝗽𝗿𝗶𝗰𝗲𝘀: - Residential electricity in Florida is 27% less expensive than in Pennsylvania - 60% less than in New York - 137% less than in California 𝗙𝗹𝗼𝗿𝗶𝗱𝗮'𝘀 𝗺𝗮𝗿𝗸𝗲𝘁 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝟭𝟬𝟭: - No ISO or RTO participation - Self-contained transmission system - 5 major Independently Operated Utilities (IOUs): Duke Energy Corporation, Florida Power & Light, Gulf Power, Tampa Electric, Florida Public Utilities Company - Several co-ops and public power entities - Some utilities participate in the Florida Municipal Power Agency (FMPP) 𝗦𝗼 𝗵𝗼𝘄 𝗱𝗶𝗱 𝘁𝗵𝗲𝘆 𝗱𝗼 𝗶𝘁? -Focus on "cost-effective and affordable energy supply" over "Net-Zero Targets" -Maintain natural gas power plants for reliable base load -Calculated renewable energy adoption and roll-out 𝗥𝗲𝘀𝘂𝗹𝘁: - Solar power will reach 32% of supply, by 2030 (CF excluded). - Duke Energy Corporation plans to build 14 new solar plants between 2025 and 2027, adding 1,050 megawatts of clean energy to Florida's grid. 𝗧𝗵𝗲 𝗳𝗶𝗻𝗮𝗻𝗰𝗶𝗮𝗹 𝗶𝗺𝗽𝗮𝗰𝘁 𝗼𝗳 𝘁𝗵𝗲𝘀𝗲 𝗶𝗻𝘃𝗲𝘀𝘁𝗺𝗲𝗻𝘁𝘀 𝗶𝘀 𝘀𝗶𝗴𝗻𝗶𝗳𝗶𝗰𝗮𝗻𝘁: 𝗜𝘁'𝘀 𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝘁𝗼 𝘀𝗮𝘃𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀 $𝟭𝟱𝟬 𝗺𝗶𝗹𝗹𝗶𝗼𝗻 𝘁𝗼 $𝟮𝟬𝟬 𝗺𝗶𝗹𝗹𝗶𝗼𝗻 𝗽𝗲𝗿 𝘆𝗲𝗮𝗿 𝗶𝗻 𝗿𝗲𝗱𝘂𝗰𝗲𝗱 𝗳𝘂𝗲𝗹 𝗰𝗼𝘀𝘁𝘀. What did I miss? #energytransition #ai #naturalgas #solar Quant AiQ Arcus Power Corp

    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    The most underrated challenge in large AI models isn't algorithms - it's data center reliability. Here is how Facebook managed 16,384 Nvidia H100 GPUs to train Llama 3 for 54 days straight. Key takeaway: 𝗦𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝗯𝗿𝗼𝗸𝗲 𝗱𝗼𝘄𝗻 𝗲𝘃𝗲𝗿𝘆 𝘁𝗵𝗿𝗲𝗲 𝗵𝗼𝘂𝗿𝘀. 𝗦𝗼 𝗶𝗳 𝗠𝗲𝘁𝗮'𝘀 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲𝗱 𝗮 𝗳𝗮𝗶𝗹𝘂𝗿𝗲 𝗲𝘃𝗲𝗿𝘆 𝟯 𝗵𝗼𝘂𝗿𝘀 , 𝘄𝗵𝗮𝘁 𝗸𝗶𝗻𝗱 𝗼𝗳 𝗰𝗵𝗮𝗼𝘀 𝗮𝘄𝗮𝗶𝘁𝘀 𝗮 𝘀𝘆𝘀𝘁𝗲𝗺 𝟮𝟬 𝘁𝗶𝗺𝗲𝘀 𝗹𝗮𝗿𝗴𝗲𝗿? 𝟭. 𝗚𝗣𝗨 𝗙𝗿𝗮𝗴𝗶𝗹𝗶𝘁𝘆 A single faulty GPU caused 148 interruptions during their training run. That's 30.1% of all recorded interruptions. 𝟮. 𝗧𝗵𝗲 𝗗𝗼𝗺𝗶𝗻𝗼 𝗘𝗳𝗳𝗲𝗰𝘁 One tiny GPU hiccup can derail the entire training job. It's like a massive Jenga tower - pull out one block, and the whole thing comes crashing down. 𝟯. 𝗘𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝗮𝗹 𝗖𝘂𝗿𝘃𝗲𝗯𝗮𝗹𝗹𝘀 Even slight temperature changes throughout the day caused 1-2% throughput variations. Sounds small, but at this scale, it adds up fast. 𝟰. 𝗣𝗼𝘄𝗲𝗿 𝗚𝗿𝗶𝗱 𝗦𝘁𝗿𝗲𝘀𝘀 Imagine tens of thousands of GPUs suddenly changing their power consumption. We're talking megawatts of fluctuation. It's pushing data center power grids to their absolute limits. 𝟱. 𝗧𝗵𝗲 𝗦𝗰𝗮𝗹𝗲-𝗨𝗽 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 By the end of next year, Meta's aiming to have 350,000 Nvidia H100s 𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗯𝗼𝘁𝘁𝗼𝗺 𝗹𝗶𝗻𝗲: Training massive LLMs isn't just about throwing more GPUs at the problem. It's a delicate dance of hardware management, failure mitigation, and environmental control. Meta's team pulled off a miracle by maintaining 90% effective training time despite these challenges. This is the kind of under-the-radar expertise that separates the big-league players from the wannabes in the AI race. So next time you hear about the latest breakthrough in AI, remember: behind every flashy headline is an army of engineers fighting an endless battle against hardware failures, power fluctuations, and the laws of thermodynamics. This is the real frontier of AI development, and it's only going to get more challenging as models grow larger. Arcus Power Corp #ai #llama #renewableenergy #iot #datacenter #nvidia #meta GIF created by Wish Bakshi

    • Gif Generated by Wish Bakshi
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    We were wrong. 5 years ago; the narrative was that natural gas power plants were a dying breed. Today, the case for nat gas power is stronger than ever. 𝗦𝗼 𝘄𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝗲𝗱? In 2019, the narrative around natural gas in the US power sector seemed clear. 𝗪𝗲 𝘄𝗲𝗿𝗲 𝗼𝘃𝗲𝗿𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴. 𝗗𝗲𝗺𝗮𝗻𝗱 𝘄𝗮𝘀 𝗳𝗹𝗮𝘁. 𝗔 𝗴𝗹𝘂𝘁 𝘄𝗮𝘀 𝗳𝗼𝗿𝗺𝗶𝗻𝗴. According to S&P Global, between 2008 and 2019, the US added 120 GW of gas-fired capacity. At least 200 new gas plants were planned or in development, totaling nearly 70 GW of additional capacity. 𝗖𝗮𝗽𝗲𝘅 𝗿𝗼𝘀𝗲 𝗳𝗿𝗼𝗺 $𝟲𝟬 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 𝗶𝗻 𝟮𝟬𝟬𝟴 𝘁𝗼 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 $𝟭𝟭𝟬 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 𝗶𝗻 𝟮𝟬𝟭𝟴. 𝟮𝟬𝟭𝟵 𝘄𝗮𝘀 𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝘁𝗼 𝘀𝘂𝗿𝗽𝗮𝘀𝘀 $𝟭𝟮𝟬 𝗯𝗶𝗹𝗹𝗶𝗼𝗻. Several factors were driving this overbuilding: - Historically low natural gas prices due to the shale gas revolution - Utility business models that rewarded new infrastructure - Outdated demand forecasts - High reserve margins - Slower-than-expected growth in renewable energy 𝗠𝗮𝗻𝘆 𝗲𝘅𝗽𝗲𝗿𝘁𝘀 𝗯𝗲𝗹𝗶𝗲𝘃𝗲𝗱 𝘁𝗵𝗲𝘀𝗲 𝗽𝗹𝗮𝗻𝘁𝘀 𝘄𝗼𝘂𝗹𝗱 𝗯𝗲𝗰𝗼𝗺𝗲 𝘀𝘁𝗿𝗮𝗻𝗱𝗲𝗱 𝗮𝘀𝘀𝗲𝘁𝘀 𝘄𝗲𝗹𝗹 𝗯𝗲𝗳𝗼𝗿𝗲 𝘁𝗵𝗲𝗶𝗿 𝗽𝗹𝗮𝗻𝗻𝗲𝗱 𝗹𝗶𝗳𝗲𝘁𝗶𝗺𝗲𝘀 𝘄𝗲𝗿𝗲 𝗼𝘃𝗲𝗿. 𝗙𝗮𝘀𝘁 𝗳𝗼𝗿𝘄𝗮𝗿𝗱 𝘁𝗼 𝟮𝟬𝟮𝟰: The narrative has shifted dramatically. According to Sierra Club research: 𝗧𝗵𝗲 𝗳𝗶𝗿𝘀𝘁 𝘀𝗶𝘅 𝗺𝗼𝗻𝘁𝗵𝘀 𝗼𝗳 𝟮𝟬𝟮𝟰 𝗮𝗹𝗼𝗻𝗲, 𝗰𝗼𝗺𝗽𝗮𝗻𝗶𝗲𝘀 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗱 𝗽𝗹𝗮𝗻𝘀 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱 𝗺𝗼𝗿𝗲 𝗻𝗲𝘄 𝗴𝗮𝘀 𝗽𝗼𝘄𝗲𝗿 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗮𝗰𝗿𝗼𝘀𝘀 𝘁𝗵𝗲 𝗨𝗦 𝘁𝗵𝗮𝗻 𝘁𝗵𝗲𝘆 𝗱𝗶𝗱 𝗶𝗻 𝗮𝗹𝗹 𝗼𝗳 𝟮𝟬𝟮𝟬. If this trend continues, 2024 will mark the most new gas-power generation announced since at least 2017. 𝗕𝗮𝘀𝗲𝗱 𝗼𝗻 BloombergNEF 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵: 𝗪𝗵𝗮𝘁'𝘀 𝗱𝗿𝗶𝘃𝗶𝗻𝗴 𝘁𝗵𝗶𝘀 𝗿𝗲𝘀𝘂𝗿𝗴𝗲𝗻𝗰𝗲? - Surge in demand from AI data centers - Increased power needs for manufacturing facilities - Growing adoption of electric vehicles ERCOT 𝗵𝗮𝘀 𝘁𝗵𝗲 𝗺𝗼𝘀𝘁 𝗻𝗲𝘄 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗱 𝗶𝗻 𝟮𝟬𝟮𝟰 𝘀𝗼 𝗳𝗮𝗿. Looking at all planned additions, the US Southeast is the leader. This shift is causing utilities to revise their decarbonization goals: PacifiCorp, 𝗵𝗮𝘀 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗱 𝗽𝗹𝗮𝗻𝘀 𝘁𝗼 𝗮𝗱𝗱 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗳𝗶𝘃𝗲 𝗴𝗶𝗴𝗮𝘄𝗮𝘁𝘁𝘀 𝗼𝗳 𝗻𝗲𝘄 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗴𝗮𝘀 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 They cancelled 7 GW of renewable energy projects over the next two decades. 𝗧𝗵𝗲 U.S. Energy Information Administration 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝘁𝗵𝗮𝘁 𝗴𝗮𝘀 𝘄𝗶𝗹𝗹 𝗮𝗰𝗰𝗼𝘂𝗻𝘁 𝗳𝗼𝗿 𝗻𝗲𝗮𝗿𝗹𝘆 𝟰𝟬% 𝗼𝗳 𝗨𝗦 𝗽𝗼𝘄𝗲𝗿 𝗯𝘆 𝗺𝗶𝗱𝗰𝗲𝗻𝘁𝘂𝗿𝘆, 𝘄𝗶𝘁𝗵 𝗿𝗲𝗻𝗲𝘄𝗮𝗯𝗹𝗲𝘀 𝘀𝘁𝗶𝗹𝗹 𝗯𝗲𝗹𝗼𝘄 𝗼𝗻𝗲-𝘁𝗵𝗶𝗿𝗱. Natural gas is essential for grid reliability and deploying more renewables to backstop the grid. #energytransition #datacenter #naturalgas #cleantech #ai Arcus Power Corp

    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    I enjoyed discussing AI's role in power markets and renewable energy with electrical engineers, data scientists, and energy enthusiasts. We are amid a generational grid transformation. The power grid is evolving to meet the two-fold challenges of surging electricity demand and the integration of renewable energy sources. AI navigates these challenges, from assisting grid stabilization against the variability of wind and solar power to optimizing battery storage/utilization and managing the complex balance of supply and demand. Machine learning can help enhance grid resiliency and pave the way for a smarter, more sustainable energy future. In this story, AI is our Sam to Frodo. Thank you for having me #yycdatasociety Benjamin Reeves Karen Zhang Stay tuned, more to come! #platformcalgary #AI #renewableenergy #trading #datascience #electricalengineering #energystorage #windturbines #grid #powergeneration

    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    "I'm going to make him an offer he can't refuse." (The Godfather) Whether we like it or not, Data Centers might become Nat Gas's new best friend. Here is why... 𝗧𝗵𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺: These data centers need 24/7 baseload power. Solar goes to bed at night. Wind is unreliable. Coal is dead/dying. Batteries aren't ready (believe me, I want this). Nuclear? Too slow to build. 𝗙𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀: 1. AI queries are energy hogs. We're talking 10x more than your average Google search. 2. Goldman Sachs estimates data centers could guzzle 8% of U.S. power by 2030. That's nearly triple their current 3% appetite. 3. MAGMA companies (Meta, Apple, Google, Microsoft, Amazon) are cranking up their capex by a whopping 37% next year. And they're just getting started. 𝗧𝗵𝗲 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻: Nat gas: flexible and abundant. Forecasts for data center gas demand in 2030 range from 1 to 20 Bcf/d, with a median of 6.5 Bcf/d. 𝗙𝗼𝗿 𝗰𝗼𝗻𝘁𝗲𝘅𝘁, 𝘁𝗵𝗮𝘁 𝗺𝗲𝗱𝗶𝗮𝗻 𝗶𝘀 𝗮𝗯𝗼𝘂𝘁 𝟭𝟴% 𝗼𝗳 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗨.𝗦. 𝗴𝗮𝘀 𝗱𝗲𝗺𝗮𝗻𝗱 𝗳𝗼𝗿 𝗽𝗼𝘄𝗲𝗿 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻. 𝗧𝗵𝗲 𝗵𝗶𝗴𝗵-𝗲𝗻𝗱 𝗲𝘀𝘁𝗶𝗺𝗮𝘁𝗲? 𝗜𝘁'𝘀 𝗼𝘃𝗲𝗿 𝗵𝗮𝗹𝗳. And it's not just speculation. It's happening now: 𝗡𝗮𝘁𝘂𝗿𝗮𝗹 𝗴𝗮𝘀 𝗻𝗼𝘄 𝗳𝘂𝗲𝗹𝘀 𝟰𝟮.𝟰𝟭% 𝗼𝗳 𝗨.𝗦. 𝗲𝗹𝗲𝗰𝘁𝗿𝗶𝗰𝗶𝘁𝘆 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻, 𝘂𝗽 𝗳𝗿𝗼𝗺 𝟯𝟱% 𝗶𝗻 𝟮𝟬𝟭𝟴 𝗮𝗻𝗱 𝗷𝘂𝘀𝘁 𝟮𝟰% 𝗶𝗻 𝟮𝟬𝟭𝟬. While gas demand for power is surging, other sectors are stagnating. Industrial use? Up just 3.1%. Residential use? Down 2.5%. Commercial use? Down 1.2%. 𝗧𝗵𝗶𝘀 𝗱𝗶𝘃𝗲𝗿𝗴𝗲𝗻𝗰𝗲 𝗶𝘀 𝗰𝗿𝘂𝗰𝗶𝗮𝗹. 𝗜𝘁 𝘀𝗵𝗼𝘄𝘀 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝘀𝗲𝗰𝘁𝗼𝗿 𝗶𝘀 𝗱𝗿𝗶𝘃𝗶𝗻𝗴 𝗴𝗮𝘀 𝗱𝗲𝗺𝗮𝗻𝗱 𝗴𝗿𝗼𝘄𝘁𝗵. Now, let's talk about supply and infrastructure: The U.S. is swimming in natural gas. We've got decades of supply. Here's why this matters: 𝗦𝗽𝗲𝗲𝗱: Gas-fired plants can be built in 3-4 years. Try that with a nuke plant (Vogtle was 7 years late and $17 billion over budget). 𝗙𝗹𝗲𝘅𝗶𝗯𝗶𝗹𝗶𝘁𝘆:  Gas plants can ramp up and down quickly, perfect for balancing those pesky intermittent renewables. 𝗔𝗯𝘂𝗻𝗱𝗮𝗻𝗰𝗲:  We're not running out of gas anytime soon. This isn't a short-term play. TC Energy is reinforcing pipelines. Energy Transfer is in talks with chip manufacturers. Energy transition, AI, and data centers are about to reshape energy demand in a way we haven't seen since the Industrial Revolution. What did I miss? Arcus Power Corp #naturalgas #ai #datacenter #energytransition

    • No alternative text description for this image
  • Quant AiQ reposted this

    View profile for Wish Bakshi, graphic

    I build Data & AI Systems | Specialist in Energy Trading & Operations (OT) | Power, SCADA, LNG, Renewables

    The Grid: What's happening in PJM Interconnection is ugly. It's a dress rehearsal for what capacity and energy-only markets might encounter in the coming years. 𝗧𝗵𝗲 𝘁𝗼𝘁𝗮𝗹 "𝗰𝗼𝘀𝘁 𝘁𝗼 𝗹𝗼𝗮𝗱" 𝗿𝗼𝘀𝗲 𝘁𝗼 $𝟭𝟰.𝟳 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 𝗳𝗿𝗼𝗺 $𝟮.𝟮 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 𝘀𝗶𝗻𝗰𝗲 𝘁𝗵𝗲 𝗹𝗮𝘀𝘁 𝗮𝘂𝗰𝘁𝗶𝗼𝗻. Consumers will have to cover the "Total Cost to Load" if post-auction adjustments are minor. 𝗪𝗲'𝗿𝗲 𝘀𝗲𝗲𝗶𝗻𝗴 𝗿𝗲𝘁𝗶𝗿𝗶𝗻𝗴 𝘁𝗵𝗲𝗿𝗺𝗮𝗹 𝗽𝗹𝗮𝗻𝘁𝘀, 𝗯𝗼𝗼𝗺𝗶𝗻𝗴 𝗱𝗮𝘁𝗮 𝗰𝗲𝗻𝘁𝗲𝗿𝘀, 𝗶𝗻𝗰𝗿𝗲𝗮𝘀𝗶𝗻𝗴 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗴𝗮𝘀 𝗲𝘅𝗽𝗼𝗿𝘁 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆, 𝗮𝗻𝗱 𝗮 𝗴𝗿𝗶𝗱 𝘁𝗿𝘆𝗶𝗻𝗴 𝘁𝗼 𝗮𝗱𝗮𝗽𝘁. This post will discuss the basics and the impact of data centers. A previous post discussed why natural gas will play a critical role (see comments). Most of the region's latest auctions saw prices jump to $270/MW-day. Some zones hit $465/MW-day. This is up from $29/MW-day in the previous auction. 𝗪𝗵𝘆 𝘁𝗵𝗲 𝘀𝗽𝗶𝗸𝗲? • 6,600 MW of generation retired • Only 864 MW of new generation added • 1,311 MW less demand response offered • Peak load forecast up by 3,243 MW • Target installed reserve margin increased from 14.7% to 17.8% • Dominion zone re-entered the auction 𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗺𝗶𝘅: • Gas-fired: 48% • Nuclear: 21% • Coal: 18% • Demand response: 5% • Hydroelectric: 4% • Solar and wind: 1% each • Other: 2% 𝗡𝗼𝘄 𝗹𝗲𝘁’𝘀 𝗮𝗱𝗱 𝗱𝗮𝘁𝗮 𝗰𝗲𝗻𝘁𝗲𝗿𝘀 𝘁𝗼 𝘁𝗵𝗲 𝗺𝗶𝘅. Aurora Energy Research - North America report suggests data center load growth uncertainty from 9GW to 20GW by 2030. 𝗧𝗵𝗮𝘁'𝘀 𝘂𝗽 𝘁𝗼 𝟰𝟱% 𝗼𝗳 𝘁𝗵𝗲 𝘀𝘁𝗮𝘁𝗲'𝘀 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗽𝗲𝗮𝗸 𝗱𝗲𝗺𝗮𝗻𝗱. Dominion Energy, Virginia's main utility, expects a 50% increase in peak power demand in just six years. This rapid growth raises some concerns: • Reliability: Can the grid keep up? • Price volatility: Supply constraints could lead to more price spikes • Interconnection delays: A backlog is slowing down new projects 𝗟𝗮𝘀𝘁 𝘆𝗲𝗮𝗿, 𝗣𝗝𝗠 𝗳𝗼𝘂𝗻𝗱 𝟮𝟲 𝗚𝗪 𝗼𝗳 𝗴𝗮𝘀 𝗮𝗻𝗱 𝗰𝗼𝗮𝗹 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀 𝗮𝘀 𝘂𝗻𝗿𝗲𝗹𝗶𝗮𝗯𝗹𝗲. That's a significant chunk of capacity that can't be counted on. Watch this market closely. The outcome will provide us with a glimpse into the future. Bonus question: which energy source do you think can sustain this type of load growth? #datacenter #naturalgas #ai #pjm Arcus Power Corp

    • No alternative text description for this image

Similar pages