This week the market ran the same test on four different companies, and mostly failed all of them.

Nvidia posted $96 billion in quarterly revenue on Wednesday, and the stock still had to fight for every point of its after-hours gain.

Salesforce beat earnings by 80% the same night, and investors spent the next day pointing at the decelerating growth hiding underneath the beat.

Oracle's backlog hit $638 billion this year, and the stock is still down more than half from its September peak.

The market has stopped treating a big number as proof of anything on its own.

Broadcom just walked into that same test with a $164.6 billion AI backlog, and today I want to look at exactly what it left out.

Missed Wednesday's deep dive on SanDisk?

Nvidia gave the market exactly what it wanted: proof that AI demand is still running hot. Revenue came in at $96.2 billion, ahead of expectations, adjusted EPS was $2.22, and the company guided for about $108 billion in Q3 revenue. Shares rose roughly 7% as investors focused on Nvidia’s message that AI spending is expanding beyond hyperscalers into AI labs, neo-cloud providers, and deeper AWS partnerships. In plain English, the AI buildout still looks very much alive.

Peter Thiel has been steadily selling Palantir shares for years, but he still remains the company’s largest individual shareholder. That makes the signal more nuanced: it may reflect diversification and long-planned monetization rather than a direct loss of confidence. Still, when a founder sells into a stock with huge expectations and a 2026 revenue forecast around $8.15 billion, investors should at least ask whether valuation has run ahead of fundamentals.

Berkshire’s Alphabet bet has become too big to treat as a side position. Its Q2 filing showed Berkshire boosted its Alphabet stake by 83% to about $37.8 billion, making it a top-three holding, while also adding Lennar and trimming names like Capital One, Nucor, Kroger, and Constellation Brands. The bigger message is that Greg Abel’s Berkshire is leaning much harder into AI-linked cloud infrastructure, but doing it through a highly profitable hyperscaler rather than a speculative AI pure play.

Connect your customer data, teams, and tools without the hassle of complex integrations or lengthy setups. One easy platform gives marketing, sales, and service teams a unified customer view and the tools to turn it into growth.

Why HubSpot and what's new
• Generate leads and automate marketing with Marketing Hub
• Build your pipeline and close more deals with Sales Hub
• Scale customer support and drive retention with Service Hub
• Keep customer data clean, connected, and actionable with one, unified platform

Join 306,000+ in over 135 countries using HubSpot to grow their businesses.

See what a more connected approach to growth can do for you and your team. Get setup quickly and start checking off your hardest tasks.

Get started free

This week gave us three versions of the same headline. Massive numbers, muted stock reaction.

Nvidia's guide, Salesforce's beat, Oracle's backlog, and now Broadcom's $164.6 billion in committed AI orders.

The easy read is simple. Biggest backlog in the chip business, six major AI customers locked in, $100 billion in AI revenue coming by 2027 according to the CEO.

Pull on that thread and it gets less simple fast.

Broadcom will not say which customers make up that number, how concentrated it is, or how much of it is being financed rather than paid for outright.

That is the actual story here.

Broadcom disclosed $164.6 billion in firmly committed AI orders this week.

Roughly 30% of that converts to revenue within the next twelve months.

Some of those contracts carry no termination rights at all. That is not a sales forecast. That is signed paper.

Six major AI customers are now locked into Broadcom's custom accelerator business.

CEO Hock Tan says $100 billion in AI chip revenue is coming in 2027 alone.

Compare that to Oracle's $638 billion backlog, which proved this week that hyperscalers will sign contracts at almost any size.

Broadcom's version carries less delivery risk. Custom AI silicon has been its core business for years, not a pivot chasing the AI trade.

Ask Broadcom which customers make up that $164.6 billion and you get silence.

Ask how concentrated it is, or how much runs through debt and escrow rather than cash, and you get the same silence.

That matters because this isn't an abstract risk.

Meta and OpenAI's AI spending is widely reported to run through a mix of debt, equity partnerships, and circular vendor deals.

Nobody outside those companies fully knows how much leverage sits underneath the buildout right now.

If even a fraction of Broadcom's backlog rests on two or three buyers stretching their balance sheets, "biggest backlog in chips" stops being reassuring.

It starts being a concentration bet wearing a good headline.

Broadcom didn't fall alone on August 24th.

The Philadelphia Semiconductor Index dropped 2.42% that day, and Micron fell 5.83% in the same session.

Broadcom actually had the second-smallest decline among major chip names.

The move traces back to the same worry showing up everywhere this week. AI financing fears, not AI demand fears.

Investors are not doubting whether the compute gets used.

They are starting to doubt how much of the spending behind it is real cash versus leverage stacked on leverage.

Broadcom closed Wednesday at $355.59, down slightly on the day and about 1.9% lower over the past week.

That puts the stock roughly 28% below its 52-week high of $495, and that gap hasn't meaningfully closed even as the backlog headlines kept getting bigger.

Every time good news lands, it mostly buys back a day or two of selling before the stock drifts again.

That is not what conviction looks like.

The zone worth watching heading into next week sits around $340 to $345, close to where buyers stepped in during the last leg of the semis selloff. [VERIFY: support level approximate, confirm against latest chart before publishing]

If that holds, the market still believes the backlog even without the customer detail.

If it breaks, the concentration question stops being theoretical.

If Broadcom or any of its named AI partners discloses order size specifics next week, that's the single biggest catalyst either direction. Bull case: disclosure shows genuine diversification across the likes of OpenAI, Meta, Google, and Anthropic, and the concentration fear evaporates. Bear case: disclosure shows two customers make up the majority of the backlog, and the market reprices the multiple down hard. Wildcard: nobody discloses anything, and the stock keeps drifting on uncertainty alone, which is its own kind of answer.

This week's semis selloff was about financing structure, not demand, and that conversation isn't going away. Bull case: hyperscaler balance sheets prove strong enough to fund this without stress, and the leverage fear fades. Bear case: one major AI buyer shows real strain servicing debt tied to compute spending, and every backlog number in the sector gets discounted overnight.

Oracle already showed the market what happens when backlog outpaces cash flow: negative $23.7 billion in free cash flow this year. Bull case: Broadcom's custom chip model needs far less capex than Oracle's data center buildout, so the comparison doesn't hold. Bear case: if Broadcom's cash flow starts looking anything like Oracle's, the multiple compresses fast regardless of backlog size. I covered this exact pattern with AMD three weeks ago in Data Center Revenue Doubled. AMD Fell Anyway. It hasn't broken since.

Broadcom's smaller decline relative to peers during the August 24th selloff could mean real relative strength. Bull case: institutional money used the pullback to add exposure to the most diversified custom silicon name in the sector. Bear case: the whole group is still repricing lower, and Broadcom just has more room to fall before it catches up to the rest.

I want to like this backlog number more than I actually do.

$164.6 billion is a real, enormous, mostly non-cancellable pile of committed spending, and pretending otherwise would be dishonest.

But I spent this whole week watching companies report genuinely good numbers and getting punished or barely rewarded for them anyway.

Nvidia beat and had to fight for its gain. Salesforce beat by 80% and the market immediately found the deceleration underneath it. Oracle's backlog is nearly five times last year's and the stock is still down 56% from its peak.

Three times in one week is not a coincidence.

Investors stopped taking headline numbers at face value, and Broadcom's silence on customer concentration is exactly the kind of gap that pattern goes looking for.

Not because the backlog is fake. Because a real number can still be an incomplete one.

That is the lesson this week actually taught me, and it's worth carrying past Friday. The biggest number in the release is rarely the one that decides whether the stock works from here.

Ava is the first AI BDR to run outbound end to end, finding leads or ingesting your CRM accounts, sending personalized emails on your reps' behalf, and booking meetings, autonomously or on copilot. She runs outbound for DoorDash and Grammarly. She's SOC 2 Type II audited, SSO and GDPR ready.

Book a demo

This week tried to teach every investor watching earnings the same thing.

A big number by itself is not an edge.

The edge is in what companies choose not to say next to it.

Backlog, guidance, beats. They all sound like certainty.

Most of them are really just the start of a question, not the end of one.

Go into the weekend asking what's missing from the number, not just marveling at the number.

That habit will serve you longer than any single earnings beat this year.

Know someone who'd rather think than chase? Forward them this edition.