Broadcom’s September 2 results show custom AI hardware turning into substantial sales. On the accompanying call, management said it expected to accelerate Ironwood shipments to Anthropic in the fourth quarter. That puts Google’s TPU technology at the center of a relationship whose economics differ for Broadcom Inc. (NASDAQ:AVGO) and Alphabet Inc. (NASDAQ:GOOGL).
Broadcom can sell custom accelerators and networking components. Alphabet can strengthen an alternative computing platform and its cloud offering. However, demand for Google-designed chips does not automatically establish how much revenue or profit Alphabet receives from every deployment.
Follow the disclosed agreements
Broadcom’s April 6 filing described agreements to develop and supply Google’s future TPUs, plus networking and other rack components through up to 2031. Separately, Anthropic was to access approximately 3.5 gigawatts of next-generation TPU capacity through Broadcom beginning in 2027, with consumption dependent on Anthropic’s continued commercial success.
Broadcom provided a more aggressive deployment outlook on September 2. Management said it is deploying 1 gigawatt of Ironwood for Anthropic in 2026, expects another 5 gigawatts of TPU v8i deployment in 2027 and sees an incremental 10 gigawatts in 2028. Those figures are management expectations and should not be treated as equivalent to the 3.5-gigawatt contractual disclosure in the April filing.
That earlier disclosure is important because it identifies both an opportunity and a condition. It does not justify treating every future gigawatt as operating capacity today, or assigning identical revenue to Broadcom and Google.
The near-term data is already pretty significant. Broadcom reported $16.7 billion of AI semiconductor revenue for its fiscal third quarter ended August 2, up 221% year over year. Companywide free cash flow was $13.7 billion. Management forecast $21.7 billion of AI semiconductor revenue for the fourth quarter; that remains guidance.
For Broadcom Inc., the bull case is sustained customer adoption of custom systems. The bear case is that a concentrated set of expensive deployments depends on customers’ funding, technical execution and commercial growth. Strong aggregate results cannot establish Anthropic-specific margins.
For Alphabet Inc., broader TPU adoption could reinforce its technology and attract workloads. Its June-quarter cloud revenue grew 82%, with operating income of $8.814 billion. Yet Alphabet’s overall quarterly capital expenditure exceeded operating cash flow. Expansion still has to earn a return, and Google Cloud results do not isolate the profitability of the Broadcom-Anthropic arrangement.
Insider Monkey’s tracked worksheet sample recorded 170 Broadcom holders in Q2 2026 versus 173 in Q1, while Alphabet’s Class A count rose to 275 from 265. The samples comprised 1,006 and 1,022 managers. Ken Fisher’s firm reported 15,131,673 Broadcom shares and 39,989,840 Alphabet Class A shares at June 30, historical exposure rather than proof of a TPU thesis.
August 14 short interest was 1.20% of Broadcom’s float and 0.67% of Alphabet’s, according to MarketBeat. Investors now need to distinguish shipped hardware, consumed capacity and the profit each partner retains.
READ NEXT: The Trillion-Dollar AI Capex Gap: Why Too Much Hardware Could Be Nvidia’s Trap and Microsoft’s Opportunity and Redditors Are Skeptical of SpaceX’s Orbital Data Centers. Are They Right?
Hedge Fund Resource Center