AI accelerator manufacturers that acquire software startups gain market share through hardware-software integration

What's this about?

People disagree about whether AI chip makers gain market share by buying software firms.

The key question asks if better chip-and-software teamwork brings more customers.

What supporters say

• Chip firms that join hardware and software work may run their systems more smoothly.
• Teams that build chips, links, and code together can make AI work faster and better.
• Good software tools can make a chip easier to use, so more people may choose it.

What critics say

• Buying a software firm might slow new ideas or remove a rival from the race.
• Chip makers can team up with outside software firms instead of buying them.
• We still lack clear proof that these deals caused any rise in market share.

How to read this

The number of points on each side does not show who is right; check how strong the proof is behind each point.

The bottom line

The idea makes sense: chip makers may gain from owning software skills.

Still, the evidence does not clearly show that buying software firms increases market share.

AI accelerator makers may gain an edge by buying software startups and combining their tools with their chips. But the available evidence supports the technical logic more clearly than it proves that acquisitions actually increase market share.

The case for

Hardware and software designed together can work better in practice. Research on custom AI accelerators shows that coordinating the chip, networking system, compiler and software framework can improve how well the whole system fits large-scale machine-learning work 1. Google’s TPU v4, for example, was designed alongside its optical interconnect and software system. Meta’s MTIA project took a PyTorch-first approach aimed at improving developers’ experience and making the hardware fit existing software workflows.

These studies do not examine acquisitions. They do, however, make the proposed mechanism plausible: bringing software expertise inside a chip company could reduce delays and coordination problems compared with working at arm’s length.

Broader semiconductor research also offers indirect support. A study of 470 semiconductor companies between 1999 and 2018 compared vertically integrated companies with more specialized firms. Its results concern operating efficiency across the industry, not AI-accelerator market share, but they suggest that organizational structure can affect how effectively companies operate 3.

Recent deals show that accelerator manufacturers see software as strategically important. AMD said its purchase of Nod.ai would improve automated compiler tools and speed the deployment of optimized models across its processors and GPUs. NVIDIA completed its acquisition of Run:ai, whose software helps manage and optimize AI-hardware infrastructure. If these tools lower the cost or difficulty of deploying a customer’s models, they could make the manufacturer’s hardware more attractive 2.

The case against

The main problem is a lack of evidence linking acquisitions to market-share gains. TechInsights estimated 2023 data-center AI-chip revenue shares at 65% for NVIDIA, 22% for Intel and 11% for AMD (see Figure 1). But those figures cannot show what caused the companies’ positions. They do not separate the effects of CUDA and other earlier ecosystem investments from product design, supply, pricing, customer relationships or acquisitions 4.

Acquiring software is also not the only way to coordinate hardware and software. Companies can use open-source compilers, framework partnerships, cloud agreements and supplier contracts. Vertically separate rivals may therefore achieve many of the same benefits while retaining more flexibility or access to specialized partners. The available semiconductor research does not show that ownership consistently beats these alternatives 5.

Integration can also produce harms that have little to do with better technology. Research on platform acquisitions has examined whether buying companies affects innovation in acquired patented technologies, using patent citations as a measure. Economic analysis and U.S. antitrust guidance also warn that vertical integration can raise entry barriers, restrict interoperability, help companies monitor suppliers or neutralize potential competitors 6. A manufacturer could gain market share by limiting rivals’ access to software rather than by offering hardware that customers choose because it performs better.

The strongest evidence for the deals comes from the acquiring companies themselves, which creates an obvious attribution problem. There is no credible post-acquisition comparison showing changes in customer switching, hardware use, deployment costs or market share against independent software suppliers.

The bottom line

The evidence supports a narrower claim: tightly integrated hardware and software can improve performance and usability, especially when software is closely tied to compilation, deployment or orchestration. It offers some indirect support for the idea that owning the software team may make that coordination easier.

But it does not establish that acquisitions cause manufacturers to beat vertically separated rivals in market share. Partnerships may deliver similar benefits, while market positions can reflect ecosystems, supply, pricing and customer relationships. The evidence therefore favours the technical case moderately, but the broader market-share claim only weakly: it is plausible as a conditional mechanism, not a demonstrated causal explanation.