Hock Tan Says AI Orders Hold as Broadcom Sheds 4.77%

Broadcom CEO Hock Tan said on Sept. 14, 2026 that the company's AI revenue targets have not changed and demand for AI infrastructure remains strong despite Anthropic's push to slow the development of increasingly powerful models, as Broadcom shares closed at 344.72, down 4.77% on the day.

Broadcom chief executive Hock Tan used his first public comments since Anthropic began pressing the industry to slow the development of ever more powerful AI models to deliver a simple message: nothing about the company's AI revenue plan has changed. Demand for AI infrastructure, Tan said, remains strong.

Investors were not entirely persuaded. Broadcom Inc. (AVGO) closed at 344.72 on Monday, Sept. 14, 2026, down 4.77% from the prior close of 361.99 — a drop of 17.27 points — after trading in a range of 343.91 to 350.25. That was a far sharper move than the broad market registered. The Nasdaq 100 ETF (QQQ) finished at $709.18, off 0.80%; the S&P 500 ETF (SPY) closed at $760.88, down 0.45%; and the Dow 30 ETF (DIA) ended at $524.49, down 0.25%. Broadcom's decline was roughly six times the Nasdaq 100 proxy's, which is the clearest signal available that Monday's selling was specific to the AI supply chain rather than a general risk-off day.

The gap between a CEO's guidance and a customer's philosophy

The tension Tan was answering is unusual in semiconductors. Chip demand normally softens because customers run out of money, run out of end-market demand, or find a cheaper supplier. Anthropic's argument is different in kind: that the frontier of model capability should advance more slowly, with more outside testing, because of the risks attached to increasingly powerful systems. That is a statement about pace, not about budgets — but the market treats the two as linked, because the pace of frontier model training is precisely what has justified the scale of compute buildouts.

Tan's response, as reported by CNBC, was to draw a line between the debate and the order book. AI revenue targets have not changed. Demand for AI infrastructure remains strong. For a company whose custom accelerator business is built on multi-quarter design and delivery commitments from a small number of very large customers, that is the most load-bearing thing a chief executive can say.

Why Broadcom is the stock that absorbs this argument

Broadcom sits at an awkward intersection. It is not a merchant GPU vendor selling into a broad market of buyers; its AI business is concentrated in custom silicon designed with and for a handful of hyperscale and frontier-lab customers, plus the networking silicon that ties those clusters together. Concentration cuts both ways. It gives the company visibility that most semiconductor firms would envy — and it means any suggestion that one of those customers might moderate its appetite lands directly on the revenue line rather than being diluted across thousands of accounts.

That is why a philosophical argument about model scaling translated into a 4.77% single-day decline in a company whose management had just publicly reaffirmed its targets. The market was not disputing Tan's statement about current demand. It was repricing the durability of the demand curve beyond the period that current commitments cover.

What the tape actually said on Monday

The broad indices were down modestly. Within that, the Nasdaq 100 proxy's 0.80% decline outpaced both the S&P 500 and the Dow, consistent with a session where large-cap technology carried the weakness. Broadcom's close at 344.72 sat near the bottom of its intraday range of 343.91 to 350.25, meaning the stock did not recover into the close — sellers were still in control when the bell rang.

A stock closing at the low end of its day range after its CEO has publicly defended guidance is a specific kind of signal. It suggests the reassurance was priced as necessary rather than sufficient. Reaffirming a target tells investors the near-term contracted business is intact; it says nothing about orders that have not yet been placed.

The questions the guidance does not answer

Three things are worth tracking from here.

Whether the slowdown push finds institutional support. Anthropic's position only affects Broadcom's economics if it translates into other frontier labs or their funders changing procurement behaviour. A minority view held by one lab is a talking point; a coordinated shift in training cadence across several is a capital expenditure event.
Whether the demand mix shifts from training to inference. If the frontier slows but deployment keeps growing, aggregate compute demand need not fall — it changes shape. That matters enormously for which silicon gets ordered, and in what configuration, even if the headline dollar figure holds.
Whether customer concentration becomes the framing risk. The same structure that gives Broadcom its visibility makes it the most exposed name in the chain to any single large buyer's change of mind. Expect analysts to press harder on that in the coming quarters.

A debate that has now reached the income statement

For most of the current cycle, the safety argument about AI and the revenue argument about AI ran on separate tracks. Monday collapsed them. When a frontier lab campaigns publicly for slower model development, and a chip supplier's chief executive has to stand up the same week and confirm that revenue targets are unchanged, the two conversations have merged.

Tan's answer was the only one available to him, and on the evidence of contracted demand it is presumably accurate. But the stock's 4.77% decline, against a Nasdaq 100 proxy that fell 0.80%, is a reminder that equity markets discount several years forward. The near-term order book is not the variable being argued over. The argument is about what the buildout looks like once the current commitments are delivered — and no chief executive can settle that with a single set of remarks.

For investors, the practical takeaway is narrower than the headlines suggest. Broadcom's management has reaffirmed its AI revenue targets. That is a fact about guidance, not a forecast about 2028. The market's response was to sell the stock anyway, which is a fact about sentiment, not about fundamentals. Both can be true at once, and Monday's session is what that looks like on a screen.

Key facts

AVGO last close: 344.72, down 4.77% as of Mon, Sept. 14, 2026, 20:00 GMT
CEO statement: Hock Tan says AI revenue targets have not changed
Day range: 343.91 – 350.25; prior close 361.99
Benchmark comparison: Nasdaq 100 (QQQ) closed $709.18, -0.80%; S&P 500 (SPY) $760.88, -0.45%

Frequently asked questions

What did Hock Tan actually say?

Broadcom chief executive Hock Tan said that demand for AI infrastructure remains strong and that the company's AI revenue targets have not changed, despite growing concerns about increasingly powerful AI models. His remarks came in response to Anthropic's public push for the industry to slow the development of frontier models. He did not revise any stated target.

How did Broadcom stock react?

Broadcom shares closed at 344.72 on Monday, Sept. 14, 2026, down 4.77% from the prior close of 361.99. The stock traded between 343.91 and 350.25 during the session and finished near the low end of that range, indicating that selling pressure persisted into the close rather than easing.

Did the wider market fall as much?

No. The S&P 500 ETF closed at $760.88, down 0.45%, the Dow 30 ETF at $524.49, down 0.25%, and the Nasdaq 100 ETF at $709.18, down 0.80%. Broadcom's decline of 4.77% was substantially larger than all three, pointing to a company- and sector-specific reaction rather than broad market weakness.

What is Anthropic's slowdown push?

Anthropic has been arguing publicly that the AI industry should slow the pace at which it develops increasingly powerful frontier models, allowing more scrutiny and outside testing. It is an argument about the speed of capability advancement rather than about AI spending levels, but markets treat the two as connected because compute buildouts are justified by that pace.

Why is Broadcom more exposed than other chipmakers?

Broadcom's AI business rests on custom silicon designed with a small number of very large customers, alongside the networking chips that connect those clusters. That concentration gives it unusual forward visibility, but it also means any change of appetite at a single major customer flows straight to revenue instead of being spread across a broad buyer base.

What should investors watch next?

Three things: whether other frontier labs or their funders actually change procurement behaviour rather than just rhetoric; whether demand shifts from model training toward inference, which changes the silicon mix even if dollar totals hold; and how closely analysts start pressing Broadcom on customer concentration in coming quarters.

Sources

Broadcom CEO addresses Anthropic's slowdown push, says AI revenue targets haven't changed — CNBC Top News

Photo: Mcapdevila · BY-SA 3.0 — source