Marvell has signed an agreement to acquire Lumentum in a deal valued at 115 billion dollars (enterprise value, debt included) and around 80 billion in equity value, according to TechCrunch on 2 November 2026. It is one of the largest acquisitions in the history of the sector and sends a clear message: AI infrastructure —the chips, the optics and the network that move data inside data centres— is becoming concentrated in very few hands. The reasonable question if you run a mid-sized company in the Balearics, the Levante region or Madrid is simple: does this affect me at all?
Spoiler: less than it seems in the short term, and more than you think in the medium term. Let’s take it step by step.
What exactly has happened
Marvell, a designer of custom chips, is acquiring Lumentum —a manufacturer of optical components and network equipment— in an all-stock transaction: 4.5 Marvell shares for each Lumentum share. Closing is expected in the first half of 2028, provided it is approved by the shareholders of both companies and by regulators.
The most revealing detail is another one: both companies share a major shareholder, Nvidia, which last March invested 2 billion dollars in each one —in Lumentum, tied to a multi-year agreement to purchase advanced laser components; in Marvell, to strengthen its portfolio of networking and optical chips—. With the deal, Marvell claims it will become a benchmark in connectivity for AI, from the chip to the data centre.
(Source: TechCrunch, "In one of the largest deals in tech history, Marvell buys Lumentum for $115B", 2 November 2026. We do not add here any figures or forecasts that do not appear in the original news item.)
Why AI infrastructure affects you even if you never buy a chip
No Spanish SME is going to negotiate with Marvell. But almost all of them consume, directly or indirectly, what comes out of those factories: when you use Copilot, a conversational assistant, a demand forecasting model or a dashboard that analyses your sales, you are renting computing capacity that depends on this chain.
Three practical consequences:
1. The price of computing is not a constant. Consolidation can lower costs through scale or raise them through lack of competition. Nobody knows yet, and anyone who assures you otherwise is selling smoke. The prudent approach is to design projects whose return does not depend on the cost per query dropping 80% next year.
2. The "AI as a service" model is reinforced. If power becomes concentrated in large operators, the realistic path for a mid-sized company is not to build its own infrastructure, but to consume it from the cloud and concentrate effort where you do have an advantage: your data, your processes and your sector knowledge.
3. The timelines up there are extremely long; yours are not. This deal does not close until 2028. A well-defined AI use case in an SME is designed, tested and measured in weeks. There is no sense in postponing operational decisions while waiting to see how the semiconductor market reorganises itself.
Three takeaways for whoever decides on technology in a private company
Your competitive advantage is not in the infrastructure layer
Nobody is going to beat their competition by having better chips: everyone will have them. The difference will lie in whether your order history is clean, whether your delivery notes can be exploited, whether your item master is normalised and whether your processes are defined. Without ordered and governed data, AI does not deliver: it merely automates disorder at greater speed.
That is why, in most projects that start with us, the first useful piece of work is not "adding AI", but creating order: identifying which data sets are relevant, establishing governance protocols and defining how the result is measured.
Design so that you can change provider
The underlying lesson of any consolidation move is dependency. Applied to your company: separate the use case (classifying supplier invoices, summarising service incidents, anticipating stock shortages) from the specific model that executes it. If the business logic, the integrations and the data live on your side —your ERP, your Dataverse, your APIs—, changing engine in two years’ time is a technical adjustment, not a project from scratch. If not, you are a hostage.
Measure in euros, not in headlines
A good AI use case answers three questions before starting: which specific process it improves, how much that process costs today and how we will know in 90 days whether it has worked. If any of the three has no answer, it is not yet a project: it is a conversation.
What you can do this quarter
- Take inventory of your data. Where it is, who maintains it, what quality it has and who can access it.
- Choose a single painful process. Administrative, repetitive, with volume and with low-value-added human judgement: invoice entry, handling of recurring orders, preparation of monthly reports.
- Prototype with what you already pay for. If your company already works within the Microsoft ecosystem, Power Automate, Power BI and Power Apps allow you to test automations and dashboards without a disproportionate upfront investment.
- Set exit criteria. What minimum result justifies continuing and what result makes us stop. Stopping in time is also profitability.
How we bring this down to earth at Tisa
We have been implementing management solutions for private companies since 1987: commerce and retail, distribution, construction, food industry, hospitality, sports centres and fleets. That sector experience is what allows us to distinguish which automation adds margin and which only adds noise.
We work on three fronts that fit together: Business Central as the management core, custom development and Power Platform to cover what the standard does not resolve (mobile applications, integrations, portals, process optimisation), and AI and data to move from raw data to decisions. In practice, that means starting with a use case adapted to your business, executing it and measuring it, not buying a platform and praying.
You can see real examples by sector at grupotisa.com/casos-de-exito.
Conclusion: AI infrastructure changes, your foundations do not
The fact that Marvell is buying Lumentum for 115 billion is huge news for Wall Street and fairly irrelevant for your bottom line this year. What is relevant is the trend it confirms: AI is becoming a consumption utility, like electricity, and the companies that take advantage of it will be those that have their processes and their data ready when the socket arrives.
Want to know where to start in your specific case? Write to us at info@grupotisa.com or call us at (+34) 971 305 885 and we will assess it with you with no obligation. More information at grupotisa.com.
