Writing and Building Intelligence Economy | Grew Startups from zero to $1M+ | Fighting Cyber Crime | Join https://xitter.cf/t.co/mMbDcbJxbp
👇 Intelligence Economy
Joined April 2008
- Tweets9.1K
- Following12.4K
- Followers13.1K
- Likes1.2K
Pinned Tweet
What Everyone Is Missing About The SpaceX IPO xitter.cf/i/broadcasts/1XxyggLED…
Bloomberg reported that Alphabet, Meta, Microsoft and Amazon hold nearly $2.4 trillion in commitments for data center leases, buildings, energy and equipment, some running for decades. Which of your AI commitments could you exit within two years?
Sameer Khan retweeted
The $60 billion per gigawatt figure works out to roughly US GDP divided by total US electricity generation, an average across the economy. It says little about what the next AI gigawatt will produce. Ask for that number before approving the power.
Sameer Khan retweeted
Gartner predicts over 80% of neocloud providers will stop operating independently by 2030. If you rent AI compute from one, your contract could outlast the company that signed it. What does yours say about an acquisition?
Sameer Khan retweeted
Everyone treats the gigawatt as the unit of AI progress. I think it only measures the size of the bet. What matters is what the next gigawatt produces. intelligenceeconomy.co/p/mus… Who in your company owns that number?
Sameer Khan retweeted
A headline AI number is mostly a statement of intent. The $60 billion per gigawatt figure says little about when anything switches on. Capital can be announced in a day. Electricity arrives on the utility's schedule, and no budget line moves that schedule.
Most plant AI conversations start with the model. They should start with the record. In a July Manufacturing Leadership Council survey, 60% used half or less of their data and 42% could verify its accuracy. An agent is only as right as its record.
Sameer Khan retweeted
Common advice says wait for AI costs to fall before committing big money. That weakens when the scarce input is a grid connection, since the queue keeps moving. intelligenceeconomy.co/p/mus… Is waiting cheaper for your plan?
Sameer Khan retweeted
Most AI plans name the chips they depend on. Few name the substation. I made the case for asking about power first in my latest piece: intelligenceeconomy.co/p/mus… Who owns the answer to when your AI sites get power?
Sameer Khan retweeted
An Adipec survey of 1,832 energy decision-makers in 78 countries found 81% expect AI power demand to exceed projections. The energy industry itself thinks the forecast looks low. Does your AI plan use a power number your utility would sign?
Common advice says wait for AI costs to fall before committing big money. That weakens when the scarce input is a grid connection, since the queue keeps moving. intelligenceeconomy.co/p/mus… Is waiting cheaper for your plan?
Sameer Khan retweeted
Reuters: CenterPoint expects peak electric load up about 50% to nearly 31 gigawatts by 2031. Most AI plans are built around chip supply. Power may be the slower queue. Does your capex model carry your utility's timeline?
An Adipec survey of 1,832 energy decision-makers in 78 countries found 81% expect AI power demand to exceed projections. The energy industry itself thinks the forecast looks low. Does your AI plan use a power number your utility would sign?
Sameer Khan retweeted
Most AI project reviews start with the budget and end with the vendor. The first question should be when the power arrives, since a site without an energy date has no output date either. Approval should wait until the power timeline is written down.
A headline AI number is mostly a statement of intent. The $60 billion per gigawatt figure says little about when anything switches on. Capital can be announced in a day. Electricity arrives on the utility's schedule, and no budget line moves that schedule.
Sameer Khan retweeted
Most executives judge an AI build by its size, because size is easy to announce. The harder question is what the capacity will produce and when you will know. I worked through the math here: intelligenceeconomy.co/p/mus…
Most AI plans name the chips they depend on. Few name the substation. I made the case for asking about power first in my latest piece: intelligenceeconomy.co/p/mus… Who owns the answer to when your AI sites get power?
Sameer Khan retweeted
1/4 Every AI use case can hit its KPI and the enterprise can still lose.
Sales optimizes conversion. Risk optimizes loss avoidance. Operations optimizes throughput. Finance optimizes cost.
The conflict appears where those systems meet.
ALT Executive infographic titled The Local Optimum Trap. Four lanes—Sales optimizing conversion, Risk optimizing loss avoidance, Operations optimizing throughput, and Finance optimizing cost—converge on one customer and one workflow with conflicting objectives. Takeaway: Govern the intersections, not only the models.
Sameer Khan retweeted
Reuters: the five big AI hyperscalers issued $121 billion in US corporate bonds in 2025, against an average of $28 billion a year from 2020 to 2024. If your AI vendors are borrowing to build, how does that change the risk in your contract?
Reuters: CenterPoint expects peak electric load up about 50% to nearly 31 gigawatts by 2031. Most AI plans are built around chip supply. Power may be the slower queue. Does your capex model carry your utility's timeline?