AI Growth Is Running Into Physical Limits

About 15 GW of AI compute produced in 2027 cannot be turned on that year, according to the consensus estimate.

The chips may exist. The working data centers may not.

AI also needs power, transformers, wiring, liquid cooling, massive chillers, complex networking, and working data centers.

Power Is Becoming the Constraint

SpaceX plans to manufacture critical gas-turbine blades and vanes in-house.

Elon Musk said this could bring new natural-gas turbines online up to 18 months faster for SpaceXAI’s expanding data centers.

SpaceX and Tesla are also each building 100 GW per year of solar production capacity. Musk said natural gas will still be needed for several years to supplement and bootstrap solar.

Demand for AI compute is growing. Producing more chips solves only one part of the problem.

Memory Stocks Are Falling Despite Shortage Claims

Memory stocks keep selling off. Memory producers and buyers are describing something very different.

SK Hynix says it is sold out past 2030. Micron has customers locked through the end of the decade.

Apple called the shortage a “hundred-year flood” and raised iPad and Mac prices. NVIDIA raised server prices about 15% and increased memory-heavy commitments from $119 billion to $279 billion.

SK Hynix CEO Kwak Noh-jung told Nikkei that the crunch will run through the end of 2030. He sees no signal of oversupply or a classic downturn while AI demand remains strong.

These statements do not explain why memory stocks are falling. They show how far the stock action has moved from what the companies are saying.

Vistra Shows Both Sides of the Power Trade

Vistra has approximately 44,000 MW of generation capacity. Its 20-year power agreements with Meta and Amazon total 3,800 MW.

Q1 2026 revenue reached $5.64 billion, up 43.4% year over year. Net income was $1.029 billion, compared with a loss in 2025. Operating margin was 26.6%.

Vistra plans to acquire a 5,500 MW Cogentrix Energy natural-gas portfolio for approximately $4.7 billion.

One brief placed Vistra near 13 times forward earnings. But that case depends on future earnings holding up.

The stated risks include regulation, debt after the acquisition, and commodity-price volatility. Vistra mostly sells power at market prices, primarily in Texas, where demand increased this year while prices declined.

The reported need is 50 to 100 GW of new U.S. electricity capacity by 2030 for AI data centers alone.

Electricity is part of the AI investment debate. That does not remove the risks of owning an electricity producer.

SpaceX Launches NASA’s Roman Space Telescope

SpaceX successfully launched NASA’s $4.3 billion Roman Space Telescope.

The launch begins a five-year mission.

Venezuela’s Oil Agreement Has Large Limits

Venezuela confirmed an agreement covering 65 billion barrels across 17 fields, with the U.S. receiving “majority control.”

At roughly $83 WTI, the reported gross market value is approximately $5.4 trillion.

Reports describe 100-year development rights, 55% effective U.S. output, an ownership stake, rights to buy oil at cost, and $100 billion of investment.

But gross value is not economic value.

Venezuela mainly produces heavy, sour crude. It trades below benchmark prices and costs more to produce, blend, and refine.

Using an illustrative net value of $25 to $35 per barrel, one estimate reached up to approximately $1.25 trillion over the lives of the fields.

Shallow Investments said preparing the fields for extraction would take years. Bloomberg reported that direct U.S. control is prompting warnings of long-term risks for participating oil companies.

Calm Stocks Sit Beside Higher Rate Expectations

The S&P 500 is about 1.5% below its record after adding nearly $12 trillion since the March 2026 bottom.

It has gone 22 consecutive sessions without falling at least 1.0%. The VIX has closed at or below 16 for 18 straight trading days.

That calm sits beside Deutsche Bank’s expectation that the Fed will raise rates in September and December.

Kalshi Finance reported that gold overtook the U.S. dollar as the world’s largest global reserve asset. A separate thesis argued that gold has “serious catching up to do” after global money supply and gold diverged earlier this year.

Alibaba’s Spending and Shrinking Cash Change the Valuation

Alibaba reported 9% growth, including 45% in cloud and 16% in AI.

Operating activities provided 3 billion. Capital expenditures were 9 billion, producing 6 and 1/2 billion negative. Repurchases were cut, while cash declined from 60 billion to 30 billion.

The valuation reviewer reduced his estimate of e-commerce cash creation from 20 billion, then 15 billion, to 10 billion. He also lowered his earnings-per-share input from roughly seven to four.

With his other assumptions unchanged, his intrinsic value came out at half the stock price.

His conclusion was to reprice Alibaba, find better situations, and not bet on AI.

Revenue is growing. But the spending and shrinking cash changed his valuation.

AI Chip Competition Is Getting Broader

One article described OpenAI’s planned Jalapeno inference chip as a direct threat to NVIDIA following the $100 billion OpenAI-NVIDIA deal in 2025.

The article said NVIDIA is expanding through Hugging Face and Poolside acquisitions, self-driving technology, open-source efforts, and plans to make its entire AI stack available next year.

ARK Invest bought almost 250,000 NVIDIA shares as the stock fell nearly 5%.

Marvell drew opposite views. One post called it an AI-connectivity opportunity as moving data between chips, racks, and data centers gets harder.

EquityExplorer put NVIDIA and Alphabet in its “strong buy” group, but placed Marvell and D-Wave Quantum in its “do not buy” group.

Meta’s AMD Stake Depends on Future Deployment

Meta can earn warrants for up to approximately 10% of AMD by deploying up to 6 GW of GPUs.

One author estimated that the stake could be worth approximately $200 billion if AMD reaches a $2 trillion valuation, or approximately $300 billion at $3 trillion.

Those are that author’s future scenarios. They are not current value.

Apple Faces a Reported CEO Change

Bloomberg’s Mark Gurman reported that John Ternus will become Apple CEO this week and take over from Tim Cook on September 1.

Gurman connected the transition to two challenges: rebuilding Apple’s long-standing management team and extending its success into the AI era.

Software Earnings Carry Large Implied Moves

The largest listed implied moves are MongoDB at ±20.6%, Yext at ±19.9%, and Asana at ±19.2%.

Stock Market Nerd identified PANW, MDB, SNOW, and ZS as the software reports to watch. It said ZS needed to do better.

Technology Leads the Revenue-Growth List

The reported one-year revenue-growth leaders were:

  • SpaceX at 92%
  • NVIDIA at 69%
  • Palantir at 56%
  • AMD at 50%
  • Broadcom at 48%

Other listed rates were Netflix at 30%, Meta at 28%, Tesla at 26%, Alphabet and ServiceNow at 24%, Amazon at 20%, Microsoft at 18%, and Apple at 16%.

Technology dominates the top of that reported list.

Healthcare AI and Company Predictions Remain Unsettled

One healthcare AI list covered ten stocks across sequencing, diagnostics, robotic treatment and surgery, software, drug discovery, synthetic biology, telehealth, clinical data, and treatment decisions.

Daniel predicted that NBIS will become the fourth hyperscaler, IREN will monetize its 5.8 GW portfolio and exceed a $100 billion value, LMND will become the largest insurer, HIMS will build the most valuable subscription, and Amazon will become a $10 trillion company by 2030.

These are Daniel’s predictions, not settled outcomes.

Tesla-SpaceX Merger Claims Remain Speculation

One post assigned a 60% probability of a Tesla-SpaceX merger before 2028. Another claimed an announcement would come “any day now.”

One shareholder said SpaceX would reinvest all revenue in its Mars mission.

A skeptical view said SpaceX’s work as a U.S. military contractor could create ITAR-related consequences for Tesla’s China operations. Another view said a merger would re-rate the entire space sector overnight.

For now, these remain speculation, claims, and objections.

More compute production does not automatically mean more usable compute.