755% Power Surge, AMD's $8.2B World Labs Bet & Nvidia's Record Buyback

Jamie Cole (0:00)

AI Hardware & Chips, Daily News.

I'm Jamie Cole. Thanks for joining me.

Today, the power wall, why AI's real bottleneck isn't Nvidia chips.

The single most important shift in AI infrastructure right now isn't about chips. It's about electricity. A new analysis projects US South data center power demand will surge 755% by 2029, rising from 9 gigawatts today to nearly 79 gigawatts. The projected supply gap is 57 gigawatts. That's not a rounding error. That's a structural constraint that reframes the entire AI investment landscape. The key implication is this. For the past two years, the bottleneck conversation has been about Nvidia's supply chain, TSMC's capacity and export controls. Those constraints are real. But the analysis now points downstream to transformers, cooling systems and raw megawatts. Investors tracking the AI infrastructure layer should be watching companies like GE Vernova, Eaton and Vertiv just as closely as any chip stock. The response from hyperscalers is already taking shape and it's not waiting for the grid.

Behind the meter gas generation, meaning power produced on-site rather than drawn from the utility grid, is projected to hit 29.6 gigawatts by 2030

88% of that new US. South industrial gas capacity is going directly to datacenters. The reason is simple. On-site generation deploys faster than traditional grid infrastructure. Hyperscalers can't wait years for utility upgrades when they're racing to commission GPU clusters. The trade-off is regulatory exposure. Gas-fueled datacenters face real ESG and carbon policy risk. And if climate regulations tighten faster than expected, the economic shift toward nuclear or renewables on a compressed timeline. That's an unresolved proof point, not a settled question. While the power story was developing, AMD made its larger strategic move in years. The company acquired World Labs in an all-stock deal valued at $8.2 billion.

Stock rose 3.61% on the announcement.

The signal here is AMD's deliberate pivot away from competing purely on GPU market share, where Nvidia holds over 80% toward embodied AI and robotics. World Labs works on spatial intelligence, the kind of AI that lets machines understand and navigate physical environments. AMD's earlier hiring of Fei-Fei Li as an advisor pointed in this direction. The acquisition confirms it. Whether this move creates durable differentiation or simply diversifies AMD's risk depends on execution and on how fast the robotics market actually scales. That's not proven yet. Here's the thing. Nvidia's management sent its own signal this week. The company announced a $150 billion share buyback authorization, bringing total cumulative authorization to $235 billion. The timing matters. The stock is trading at 14.5 times forward earnings for fiscal year 2028, the lowest multiple in roughly a decade. Even his forward sales growth is estimated near 70 percent. A buyback at that scale. At that valuation, is a clear management statement on where they think the stock should be priced. Whether the market agrees depends on whether their earnings acceleration actually materializes. Compressed multiples on high growth names reflects skepticism, not just opportunity. Samsung's Foundry division is targeting 66 percent of revenue from high performance computing by 2029, up from 28 percent today. The company's S44 process node at 1.4 manometers is slated for mass production that same year. Samsung also expects its customer base to quadruple from its 2017 baseline by then. The important distinction is that this is a roadmap, not a result. TSMC's near-monopoly on leading-edge AI chip production gives it enormous structural advantage. Samsung's yield challenges at advanced nodes are well documented. The 2029 targets are credible ambitions. Whether they translate into revenue mix depends on execution that hasn't happened yet. Consider this. On the model and application side, OpenAI's DevDay this week launched DOTS, its personal AI agent product, running on the GPT 6.1 sole model, with a safety framework called OpenShell.

DOTS is live for pro and business users in eligible markets. The launch matters because it moves agentic AI from concept to commercial product. Though the real-world efficacy of OpenShell's guardrails at scale remains unproven. Separately, OpenAI is reportedly in early discussions about a new funding round that could reach $30 billion.

That figure isn't finalized. The context is a March round that raised $122 billion at an $852 billion valuation. A new raise this soon signals either accelerating capital needs or opportunistic investor interest. Given that an IPO remains on hold pending safety resolution, the strategic rationale for another private round is worth watching.

The clearest near-term signals to track whether the power supply gap starts attracting serious policy response, whether AMD's World Labs integration shows a coherent product roadmap, and whether Nvidia's earnings trajectory justifies management's confidence. Those are the real metrics right now. Thanks for listening. This podcast was built using AI technology, a Yes We production.