Editor's Note

This week made the AI buildout feel less like a story and more like a physical supply chain.

Nvidia delivered huge growth, IREN showed how fast AI cloud revenue can scale, and memory demand kept tightening.

But the same week also showed the bill: higher memory prices, power limits, financing needs, and a Fed still focused on inflation.

Nvidia Proved Demand Is Still Enormous

Nvidia gave investors the cleanest demand signal of the week.

Revenue rose 106% year over year to $96.2B. Data center revenue reached $89bn, up 117 per cent year-on-year. Net income rose to $59.7bn, and gross margin was 75 per cent.

Guidance was also strong. Nvidia expects current-quarter sales of $108bn, plus or minus 2 per cent, versus Wall Street expectations of $104bn.

The market reaction matched the numbers. Nvidia added roughly $453B in market value, described as the largest one-day gain recorded in the U.S.

The growth gap inside Big Tech was stark:

• Nvidia: 106%
• Meta: 28%
• Tesla: 26%
• Alphabet: 24%
• Amazon: 20%
• Microsoft: 18%
• Apple: 16%

That is why Nvidia still sits at the center of this market.

But the surprise was not just the growth. It was that the growth did not remove the hard questions.

Nvidia also said memory pricing has become "extreme." Costs are rising faster than expected, and supply could remain constrained through FY28.

So the same report gave investors both sides of the AI story:

Strong demand is here.

The cost of serving that demand is rising.

Memory Became the Week's Key Bottleneck

Micron was the clearest memory signal.

Micron CEO Sanjay Mehrotra said data center customers are asking for roughly 50% more supply than Micron can commit. He also said: "We see no end when supply catches up with demand."

The business evidence matters more than the slogans. Micron has signed more than 16 five-year Strategic Customer Agreements. Customers commit to specified volumes under take-or-pay terms, and they can extend the agreements.

That says customers are trying to lock in supply years ahead.

There were also aggressive Micron claims that went beyond the support in the material:

• $100 billion of guaranteed minimum revenue.
• $22 billion in customer deposits.
• Agreements covering roughly 20% of DRAM volume and one-third of NAND volume.
• Most contracts running through 2030.
• Profit of more than $500 million per day by year-end.
• A $1,500 share price and a $60 billion buyback.

The supplied material did not independently support those larger claims.

The grounded point is enough: AI needs memory, memory supply is tight, and customers are committing earlier and longer.

That tightness is already spreading into system prices. Nvidia customers were reportedly warned that AI server prices will rise more than 15% in many cases, including early 2027 shipments of Vera Rubin and Grace Blackwell systems.

AI Cloud Is Scaling, But The Accounting Is Messy

IREN became one of the best examples of the AI infrastructure trade-off.

On the growth side, the change is large.

Q4 revenue was $137.2M. AI Cloud revenue more than doubled from $33.6M to $70.5M and passed Bitcoin mining revenue for the first time. Full-year AI Cloud revenue reached $128.8M, nearly 8 times the prior year.

IREN also says operating ARR is $1B and capacity is largely sold out.

The company is targeting more than $4B in AI Cloud annual recurring revenue for 2026, with approximately 85% already contracted. Earlier material also cited a Microsoft agreement of five years, $9.7 billion, an NVIDIA deal of $3.4 billion, and multi-year contracts announced in July 2026 of $2.8 billion.

But the cost side is just as visible.

• Q4 adjusted EBITDA fell to $19.2M from $59.5M.
• Q4 net loss was $684.0M, including $450.4M in impairments.
• Full-year net loss was $702.6M, including $638.8M in impairments.

The impairments were mainly tied to decommissioning Bitcoin mining hardware as sites are converted for AI Cloud growth.

IREN has $7.6B in cash. Cash and committed financing total $14B. New GPU financing totals $2.8B and funds 90% of capex. Customer prepayments cover 45 to 55% of GPU capex.

That is not a small pivot. It is a balance-sheet and execution test.

ARR is not GAAP revenue. Contracted demand still has to become delivered capacity, recognized revenue, and acceptable profit.

Power And Financing Are No Longer Side Issues

The AI buildout is running into the physical world.

Elon Musk said AI is constrained by electricity and chips, with power and cooling currently slightly more restrictive.

One speaker cited roughly 2,000 GW of proposed U.S. power generation waiting in interconnection queues, with a median wait of roughly 5 years. Texas is managing a 474 gigawatt queue of large loads. Data centers make up 90%, and some projects face waits of up to 12 years.

That makes power a real input to AI economics, not a side topic.

Financing is another pressure point.

One weekend review said nine technology companies have about $600 billion in capital spending, but $3 trillion when purchase commitments and leases not yet started are included. It also said Google’s off-balance-sheet obligations rose more than 800% year over year and Meta’s rose 700%.

Earlier this month, Nvidia said it was working with a Wall Street consortium on a $500bn funding package. The group included Apollo Global, KKR, Brookfield, BlackRock and Goldman Sachs.

That suggests strong demand is not funding itself cleanly. More capacity needs more outside capital.

Good Results Were Not Always Rewarded

Marvell showed the difference between good results and good enough.

Q2 revenue was $2.74B, slightly above the $2.71B estimate. Adjusted EPS matched the $0.94 estimate. Data center revenue reached $2.2B, up 46% year over year. Q3 revenue guidance was $3.15B, above the $3.02B estimate.

The stock still fell. It was down 5% after recovering from a 10% overnight decline. It had already risen more than 180% this year.

That was one of the cleanest market lessons of the week: AI exposure alone was not enough. The stock price mattered.

Cybersecurity had a better reaction.

Salesforce gained more than 22% after a strong quarter, higher outlook, and its Anthropic "Claudeforce" partnership. CrowdStrike rose more than 20% after beating estimates and raising guidance.

CrowdStrike’s ARR reached $5.84B, up 25%. Record net new ARR was $333M, up 51%.

Palo Alto Networks reported next-generation security ARR of $8.1B, up 60%, and remaining performance obligations of $18.4B, up 36%. CEO Nikesh Arora said advances at the AI frontier have increased the urgency around cybersecurity.

AI is not only a chip story. It is also changing demand in software and security.

Rates, Tariffs, And Inflation Kept The Bar High

The week did not give growth stocks a free pass.

The U.S. national debt crossed $40 trillion. The 30-year Treasury yield reached 5.34% last week, up from 4.82% in late June.

Treasury plans to increase the maximum size of its bond buybacks from $2 billion to at least $4 billion. Reports also said it may use its roughly $950B to $1T General Account to support larger purchases of long-dated bonds.

Gold and Bitcoin moved with the debt concern. Gold touched a three-month high after gaining more than 5% the previous week. Bitcoin gained 22% the previous week and touched $80,000 overnight Tuesday.

Then the Fed pushed back.

Fed Chair Kevin Warsh said high inflation is "concerning" and called price stability the Fed’s "predominant focus." July PCE inflation was 3.7% year over year, slightly above the 3.6% forecast. Core inflation remained at 3.3%.

Markets raised the implied chance of a September rate increase to roughly 58%, from about 35% the prior day. The 2-year Treasury yield rose about 12 basis points.

Canada added another cost risk. It announced retaliatory tariffs on $19.94 billion of U.S. goods, with tariff rates from 15% to 50%, taking effect September 8.

Higher costs and higher rates make the AI spending test harder. The revenue is growing fast. The required return is also rising.

Counter-Thesis and Risk Watch

The strongest risk is that AI demand stays large, but Nvidia captures less of the future profit.

A Jalapeño risk-watch source said OpenAI’s Broadcom-built chip performs 1.5 to 1.9x more AI work per watt and produces responses up to 3.6x faster. Analysts cited by the source expect Nvidia’s inference share to fall from more than 90% toward 20 to 30% by 2028.

Wall Street Millennial argued that circular financing has played a major role in the AI boom, pointing to financial relationships among Nvidia, OpenAI, and cloud providers.

George Gammon compared Nvidia with Cisco during the internet boom. He cited Nvidia’s customer concentration, with four companies generating 50% of revenue, and argued that Nvidia is taking equity and balance-sheet exposure across the AI ecosystem. His near-term base case is for Nvidia to rise well above 300 and probably above 350. His base case for the following 2 to 3 years is a fall to around 40 or 50.

Cathie Wood was described as bearish on memory stocks like Micron and SK Hynix because she still views HBM as cyclical and commoditized. She is described as bullish on CBRS and Nvidia via Groq because those inference architectures can reduce or avoid reliance on HBM.

Wall Street Millennial also warned that robot success depends heavily on the environment. Amazon ended Scout in October 2022 after sidewalk delivery proved slow and inconvenient, while its controlled warehouse system worked much better, with more than 700,000 Kiva robots operating in spaces designed around them.

Meta risk also widened. Reuters said Meta’s up to $18 billion settlement with nearly all U.S. states over social media harm to teenagers opened a new front. Wall Street Millennial’s risk-watch item said that on August 7th, 2026, Meadow was forced to pay $942 million in civil penalties to New Mexico.

Looking Ahead

The open questions are now simple.

Can AI revenue grow fast enough to cover rising memory, power, and financing costs?

Can companies like IREN turn contracted ARR into recognized revenue without losing too much money on the way?

Can Nvidia keep its economics if custom chips from large AI customers keep improving?

And if the Fed keeps price stability first, how much will investors pay for long-duration AI growth?