#GoldmanSees1.2TAICapex
About GoldmanSees1.2TAICapex
Goldman Sachs expects combined 2027 capital spending by Meta, Microsoft, Alphabet, Amazon and Oracle to reach about $1.2 trillion, up from roughly $800 billion in 2026, largely driven by AI infrastructure. The buildout may support demand for chips, memory, data centers, power and cloud services, but monetization remains the key test. Can AI applications generate enough revenue and cash flow to justify the rising investment?
Populaire
Récents
GoldmanSees1.2TAICapex Publications populaires
NVIDIA’s $150B buyback increase is a signal of management confidence.
Its new share repurchase authorization lifts its remaining authorized repurchase capacity to $235B, expected to run through fiscal 2028. NVIDIA calls it the largest share repurchase authorization increase in its history.
Key figures
· Q2 FY2027 revenue: $96.2B, up 106% YoY
· Data Center revenue: $89.0B, up 117% YoY
· Q3 guidance: about $108B, plus or minus 2%
· Q1 returns: ~$20B, including repurchases and dividends
· Q2 capital returned to shareholders: around $26B, including repurchases and cash dividends
But a buyback authorization is capacity, not a fixed purchase schedule.
The impact depends on execution, share price, future cash generation, margins, and how much capital NVIDIA needs to sustain technology leadership across the AI stack.
That is the bigger market question.
AI spending is no longer just about buying more GPUs. It is about data centers, networking, power, AI factories, and whether customers can translate compute investment into durable economic returns.
If demand stays strong, the remaining $235B authorization may support the view that NVIDIA sees a multi-year cash-flow opportunity. If monetization slows or margins tighten, the trade-off between reinvestment and shareholder returns could become more visible.
Earlier in May, NVIDIA added $80B to its authorization and raised its quarterly dividend from $0.01 to $0.25 per share. The September move is a step-up in a broader capital-return policy, not a standalone event.
Two details sharpen the thesis. Q3 guidance assumes no Data Center compute revenue from China, making the $108B target a cleaner test of demand elsewhere while keeping policy risk in view. Q2 gross margin was 75.0%; Q3 guidance is 74.0%, plus or minus 50 basis points. Margin durability will help determine how much AI revenue can be reinvested in R&D and returned to shareholders.
For market participants, what matters more here: NVIDIA’s buyback size, AI demand, or whether AI customers can translate compute spending into durable returns?
#NVIDIA150BBuyback
The hard part of the alternative-chip market was never the chips. It was the balance sheet.
General Compute just announced a multi-year agreement with Cerebras to deploy wafer-scale inference. It’s the first deployment drawn against the $400M debt facility from Upper90, and that facility might be the first deal to put inference chips up as collateral instead of GPUs.
Every other neocloud buys NVIDIA, because NVIDIA underwrites the purchases. Alternative accelerators had no equivalent channel to get financed, deployed, and in front of customers at scale. Debt for alternative chips just went from a thesis to a signed contract.
Agentic coding is the first workload on the new capacity, where hundreds of sequential calls per task turn decode speed into wall-clock time. Capacity opens in Q1 2027.
So which chip gets financed next? 🤔
General Compute
🚨 BREAKING: Super excited to announce we're deploying the world's fastest inference with @Cerebras.
Talk to any developer and they're excited to build with 20x faster AI... the problem is there's almost no compute available.
We're here to solve that.
Using GPUs for prefill - it's now more affordable than ever too.
Thank you to the whole Cerebras team and excited to grow this partnership.
First tokens live Q127 🚀
The AI boom is becoming a capital-spending story.
Building AI at scale requires much more than powerful models.
It needs chips, data centers, networking, electricity, storage and enormous investment.
That's why the AI trade is spreading far beyond the companies developing AI software.
The infrastructure behind AI may become one of the biggest investment stories of this cycle.
#AI #ArtificialIntelligence #Tech #NVIDIA
NVIDIA’s enlarged authorization is less a forecast than an allocation signal: management sees enough balance-sheet flexibility to fund AI investment while returning capital. With the program running through fiscal 2028, the real test is whether rising AI capex continues converting into free cash flow rather than simply raising spending needs.
#NVIDIA150BBuyback
🔥 AI CHIP SHORTAGE… OR MEMORY SHORTAGE?
Micron’s latest earnings could reveal a hidden bottleneck in the AI boom. HBM, the high-bandwidth memory powering advanced AI GPUs, consumes far more wafer capacity than traditional DRAM. Industry data cited by S&P Global puts the capacity ratio of HBM3E to conventional DDR at roughly 3:1.
As AI demand surges, more production is shifting toward HBM—potentially squeezing regular memory supply.
🤖 AI needs GPUs.
🧠 GPUs need HBM.
🏭 HBM needs capacity.
.
Memory prices ATH.
Chip Exports ATH. Korea up 25% in last 3m alone.
I expect Micron earnings will see a major reprice up tomorrow. As well as AI supply chain in general soon.
All demand pointing up and at growth rates higher than anyone expected. This needs a major blow to reverse course (eg Hyperscaler AI CAPEX cuts) which are currently not on the horizon.
Why is Burry still short?
🚨 THIS IS SOMETHING VERY SERIOUS!!!
I just read the math behind the AI boom and it doesn't add up.
AI is $4.2 trillion short of paying for itself. And Big Tech is keeping the debt off its books.
Bain says AI has to make $6 trillion a year by 2031 to cover what's being built right now.
Every ChatGPT subscription and every AI tool companies pay for adds up to $1.8 trillion. At best.
Meta's new data center in Louisiana is a $30 billion project. Meta and Blue Owl put $27 billion of debt for it into a separate company, so it never shows up on Meta's balance sheet.
Microsoft and Google now say their servers last 6 years. Nvidia drops a new flagship chip every year.
Longer life on paper means smaller yearly costs, which means bigger profits.
Michael Burry estimates that hides $176 billion of Big Tech's costs by 2028.
I'm bullish on AI. I use it every day.
But I was in crypto in 2022. I know what it looks like when the debt shows up before the money does.
It looks like this.
AI needs $6 TRILLION in annual revenue by 2031 to justify the data centers being built, per Bain & Company.
Anthropic made just $4.6B in 2025.
The entire AI industry needs to generate roughly 1,300x that every single year just to justify the infrastructure being built.
Bull Theory
🚨 THE $2 TRILLION ANTHROPIC IPO MIGHT EXPOSE THE BIGGEST PROBLEM WITH THE ENTIRE AI BOOM.
The same money keeps moving between the same companies.
Anthropic could soon ask public investors to value it at more than $2 trillion.
But behind that valuation is a financial loop that almost nobody is talking about.
It looks something like this:
Amazon/Google → invest billions into Anthropic → Anthropic spends billions on Amazon/Google cloud infrastructure → Amazon/Google earn cloud revenue → Anthropic raises more money at a higher valuation → Amazon/Google's investments become more valuable
This is basically how the loop works:
Amazon and Google invest billions into Anthropic. Anthropic needs massive amounts of computing power to build and run Claude, so it signs huge cloud contracts with Amazon and Google.
Money goes into Anthropic as investment capital, then billions flow back toward the infrastructure businesses of the companies funding it.
Look at Amazon.
It has invested $33 billion into Anthropic, while Anthropic has committed to spend more than $100 billion on AWS over the next decade.
Amazon isn't just betting on Anthropic becoming valuable. It is also positioning itself to collect enormous cloud revenue as Anthropic grows.
Google has an even bigger relationship.
It has committed up to $40 billion to Anthropic in a deal involving both cash and compute, on top of an earlier $3 billion, for $43 billion total. Anthropic has committed to spend $200 billion on Google Cloud over the next 5 years.
Google can participate in Anthropic's rising valuation while also supplying the expensive infrastructure Anthropic needs to operate.
That becomes much more important when you look at Anthropic's actual financials.
According to details from Anthropic's IPO prospectus reviewed by Reuters, the company generated only $4.6 billion of revenue in 2025, but spent $7.33 billion on compute and infrastructure alone.
Total operating expenses reached $12.65 billion, and its operating loss widened to $8.06 billion.
And now comes the number that makes everything else look small.
Anthropic has $518 billion in future cloud, computing and infrastructure obligations, while it ended 2025 with only $20.28 billion in cash, cash equivalents and short-term investments.
Its future infrastructure commitments are more than 100 times its 2025 revenue and roughly 25 times its year-end liquidity.
Yet the company could IPO at more than $2 trillion.
At that valuation, Anthropic would be worth roughly 435 times its 2025 revenue.
Even crazier, the company was valued at around $965 billion only four months ago, meaning its proposed valuation has increased by more than $1 trillion in that time.
Anthropic is growing insanely fast, and that's the argument investors will use to justify it. Revenue grew roughly 12x in 2025.
But that's exactly where the bet becomes enormous.
Anthropic doesn't just need revenue to keep growing. It needs revenue to grow fast enough to eventually support hundreds of billions in infrastructure commitments, while turning a business that currently spends far more than it earns into an extremely profitable one.
And even its current revenue isn't as secure as a $2 trillion valuation might suggest.
Nearly one-quarter of Anthropic's 2025 revenue came from just two customers, and many of its largest customers aren't locked into long-term contracts. They can reduce spending whenever they choose.
There's another layer to this loop.
When private AI companies raise funding at dramatically higher valuations, the Big Tech companies that already hold stakes in them get to record paper gains on those stakes.
So the same companies can earn cloud revenue from AI spending while also benefiting financially when the AI companies doing that spending get revalued higher.
This doesn't mean the cloud revenue is fake, or that anything illegal is happening. The compute is real and it's actually being used.
But it raises an uncomfortable question: how much of this demand is independent, and how much of it is the same capital moving in a circle?
Big Tech funds AI labs. Those labs spend enormous amounts back with Big Tech. That spending drives data-center expansion and cloud growth. Higher AI valuations increase the value of Big Tech's original investments, freeing up even more capital to expand the system further.
Eventually, someone outside that loop has to generate enough real cash flow to justify everything being built.
That's why this IPO matters beyond Anthropic itself.
Public investors may soon be asked to put a $2 trillion price tag on a company sitting in the middle of this exact system, one carrying hundreds of billions in future obligations, billions in operating losses, and an enormous amount of future growth already priced in.
If end-user demand grows large enough, this could be one of the greatest infrastructure bets ever made.
If it doesn't, the AI boom may discover that investing billions into your own customers so they can spend billions back with you created a lot more reported growth than actual profit.
