The Power Behind the Chips

Gas turbine prices have climbed roughly 300% over the past three years, according to CNBC, as hyperscalers scramble to lock in on-site generation for AI data centers and many AI campuses cannot wait years for grid interconnection . The spike tells you everything about where the AI infrastructure race stands: the bottleneck is no longer chips or capital. It's power.

GE Vernova's order book is full until 2029, with orders going as far out as 2031 , CNBC reported from the company's Greenville, South Carolina plant. Microsoft ordered seven turbines for a roughly 2.7 GW Texas data center, with Meta, Google, OpenAI, and Anthropic also reportedly queuing up . Each unit stands roughly 30 feet tall and can power more than 500,000 homes. But the real story isn't the hardware—it's the economics. When you can't get grid power for three to six years, you buy your own power plant. That shift is rewriting the infrastructure stack beneath AI.

Can Google's Chips Actually Challenge Nvidia?

Google announced its eighth generation of custom AI chips, the TPU 8t and TPU 8i, at Google Cloud Next in April , TechCrunch reported. For the first time, one chip, the TPU 8t, is geared for model training and another, the TPU 8i, is aimed at inference . The company touts up to 3x faster AI model training, 80% better performance per dollar, and the ability to get 1 million+ TPUs to work together in a single cluster compared to previous generations.

The competitive framing matters. Nvidia controls 81% of the data center AI chip market with $193.7 billion in revenue for the fiscal year that closed in January , according to industry analysis. But Google recently struck a $5 billion deal with Blackstone to create a new cloud-services business and has decided to sell chips directly to customers rather than only through its cloud , TechSpot reported. The company has said it plans to raise $85 billion in equity, largely to support AI infrastructure .

Jensen Huang isn't worried. "I would love to hear them demonstrate the cost advantage of TPUs," he said in an April podcast appearance, adding "It makes no sense in my mind." Yet Citadel Securities, a longtime Google Cloud client, recently began using TPUs for some of its research software , according to TechSpot. The question isn't whether Google will dethrone Nvidia this year. It's whether the AI chip market is finally becoming competitive enough that hyperscalers have real alternatives.

What Happens When Chip Prices Become a Consumer Problem?

The semiconductor selloff that began in early June has turned brutal. The PHLX chip index dropped 10% on June 5, its worst single-day loss since March 2020, wiping $1.3 trillion in sector market value , according to Yahoo Finance. The selloff was triggered by Broadcom's fiscal Q2 2026 earnings report, where AI networking revenue of $4.1 billion missed analyst expectations of $4.8 billion by 14%, and CEO Hock Tan's decision to maintain rather than raise the 2027 AI semiconductor outlook disappointed investors .

SoftBank Group plunged, leading a broad sell-off in Asian technology stocks amid mounting concerns over the rising cost of AI infrastructure, with SoftBank sinking more than 5% , CNBC reported Friday. The pain spread globally. South Korea's SK Hynix fell more than 8%, while Samsung Electronics lost around 9% .

Then Apple and Microsoft made it real. Apple raised prices on MacBooks and iPads by up to $300 on June 25, citing an "unprecedented" surge in memory and storage chip costs driven by AI data center demand , according to BeInCrypto. Microsoft followed hours later, announcing Xbox console price increases of $100 to $150 per model, effective August 1, and Microsoft stock fell 3.5% . When the world's most valuable tech companies start passing chip costs to consumers, the AI infrastructure boom stops being an industry story and becomes a margin story.

Why Are Data Centers Buying Fuel Cells Instead of Waiting for the Grid?

Rystad Energy research projects a tenfold increase in fuel cell market revenues by 2030, rising from around $2.8 billion in 2025 to roughly $30 billion, with a contracted order book of approximately 9 gigawatts (GW), including framework agreements with Oracle, AEP, Equinix, and Brookfield . The catalyst is simple: US grid interconnection timelines have tripled since 2015, now stretching to three to six years for large loads , Rystad reported.

Bloom Energy has expanded its partnership with Oracle under a master services agreement for up to 2.8 GW of fuel cell systems , Yahoo Finance reported in April. Bloom's fuel cell technology will power what is expected to be one of the largest data center microgrids operating in the United States at the time of completion , according to Oracle. The economics are compelling: CATL's new sodium-ion battery for energy storage boasts a capacity exceeding 300Ah, an efficiency of 97%, and a cycle life of over 15,000 cycles, covering 2-8 hour utility-scale storage and AI data center scenarios , with commercial rollout slated for 2026.

CATL is pivoting hard. In June 2026, CATL told Reuters it expects energy storage to account for half of global sales by 2030, up from roughly a quarter today . The company has invested aggressively: In May, a CATL-affiliated fund agreed to pay up to $942m for as much as 38.1% of VNET Group, China's first US-listed internet data center operator, making the fund VNET's largest shareholder . Five years ago, 2% of CATL's sales came from battery storage. The AI boom is rewriting that mix.

Where Does Physical AI Fit?

ON Semiconductor agreed to buy Synaptics in a nearly $7 billion all-stock deal to bolster its push into physical artificial intelligence technology , CNBC reported Wednesday. The deal will give ON Semi's total addressable market a $30 billion boost to $243 billion by 2030 . The stock didn't like it— shares notched their worst day since March 2020 after the announcement.

"Physical AI is the next frontier, and this combination gives us the full stack — power, sense, compute and control — to lead it," ON Semiconductor CEO Hassane El-Khoury said . Examples include robots, self-driving cars, drones, and automated factory systems , according to TechTimes. Unlike cloud-based AI, physical AI must process information locally and respond in real time. That requires a different chip architecture—and ON Semi is betting $7 billion that the market is big enough to justify vertical integration.

The deal underscores a broader trend: consolidation in semiconductors as companies seek scale in specialized AI applications beyond the data center . Nvidia dominates training and inference chips for cloud AI, but the physical AI market — chips that process sensor data in real time for machines operating in the physical world — remains fragmented across dozens of suppliers .

What Changed This Week

The AI infrastructure trade cracked. Chip stocks erased $1.3 trillion in a single session, Apple and Microsoft raised consumer prices citing chip costs, and SoftBank lost $38 billion in market cap on reports that OpenAI may delay its IPO to 2027. Meanwhile, GE Vernova's gas turbine order book stretches to 2031, fuel cell revenues are projected to grow tenfold by 2030, and Google is spending $85 billion to challenge Nvidia with custom silicon. The bottleneck has shifted from compute to power—and the companies solving that problem are printing money.

What to Watch

ON Semiconductor CEO Hassane El-Khoury will defend the Synaptics deal as the stock digests its worst day in four years. Nvidia, AMD, and Intel report quarterly results in July—watch for any softening in data center revenue guidance, which could extend the chip selloff. Google's TPU 8t and 8i chips become generally available later this year; early customer adoption will signal whether the company can finally dent Nvidia's 81% market share. And keep an eye on grid interconnection timelines: if they stretch beyond six years, expect more hyperscalers to buy their own power plants.