Shifting technology landscape among global fault-lines

By Raju Vegesna, Chief Evangelist, Zoho

When the ground underneath keeps shifting, we cannot build a stable house. Similarly, when the technology landscape keeps shifting weekly, we cannot build a stable product, business model or business. This is the current state of the market. No vendor, including the ones actively changing the ground, has any real idea how things will pan out in the coming weeks and months.

During these shifts, I am noticing everyone is gunning for everyone else.

At the hardware layer, Nvidia is impartial about who wins the model layer. As a hardware provider, they want lots of models deployed on their hardware, including several open source models, which expands their deployment base instead of being restricted to a small set of players. So they themselves are open sourcing several models. Google has good hardware in TPU and doesn’t want to miss the hardware game either, so they are now selling TPUs externally and going after Nvidia. Apple is coming in as an efficiency player, taking this to the edge, and benefits when the model layer gets commoditized. So do other hardware players like Samsung and several Chinese device manufacturers. Their interests align well with commoditizing models, which is partly why several Chinese models are open sourced and competitive with the frontier ones.

At the model layer, model companies are gunning for the application (software) business and are inching towards services because the services pie is much bigger. Software licensing is roughly a $400 Billion market. Services is roughly a $2 Trillion market. Model companies need bigger markets to address to justify their massive investments, and the aggregate AI capex funding this race is ~$400B/year. I won’t be surprised if they zoom into markets that occupy a bigger percentage of GDP, like healthcare ($9.5 Trillion globally), pharmaceuticals ($1.7 Trillion market), finance, etc.

At the software layer, currently I see software companies playing into the hands of model companies by handing them the keys to their software and data through MCP. This is a pathway to software commoditization. When the LLM becomes the front door, the user relationship migrates to the LLM provider, and your software becomes a backend tool that gets called by name.

What I’d like to see is the reverse, where software companies commoditize model companies. The path is straightforward in theory. Own the workflow, own the data, own the user relationship, and treat the model as a swappable backend. Apple is already planning this with Apple Intelligence. I haven’t seen many companies execute on this yet, partly due to the massive hardware investments needed and partly because most software companies don’t own their cloud. I think this could be a differentiation for us. I suspect other application companies like Microsoft, Adobe and more will try to commoditize model companies the same way, as model companies try to encroach into their territory.

Another point on this. When models commoditize, proprietary workflow data and the knowledge layer (more on this later) becomes more valuable. Software companies sit on the data, the workflows and the user relationships. A great advantage, particularly the when we own the communication layer (with Mail, Cliq, etc).

The services layer is now in play as well. If AI is the one writing the code, it doesn’t matter where the developer is sitting. On the other hand, if code can be written quickly, it is better for the developer to be closer to the customer. Shortened implementation cycles are a byproduct of efficiency. They could nullify the price arbitrage services companies have leveraged over the years. The rise of FDEs (Forward Deployed Engineers) is a result of recognizing these gains. With model companies themselves wanting to do services, there is a battle coming between them and services companies who have built vertical expertise over the last few decades. That vertical expertise remains the services companies’ strength, and they could use the models themselves to challenge product companies by building bespoke applications for each business. They have the customer relationship, which is a big plus for them.

Another dimension to services is, the marginal cost of adding an engineer is approaching zero. Companies are spinning up thousands of agents to write code in parallel. The economics of Brooks’ law (mythical man month) are flipping. But the law itself isn’t dead. Coordination overhead doesn’t vanish, it just moves up a level. Silicon Valley is operating under the assumption that the mythical man month is dead along with historical software moats.

What I listed above in the hardware, models, apps and services layers is just a portion of observations. There are more dimensions to this than any individual can comprehend. Customers themselves are also frozen, running pilots that go nowhere and delaying decisions. That feedback loop extends the instability.

So, how will this play out?

I have no idea. No one does.

Not even Anthropic, OpenAI, Nvidia, Microsoft, Google. Given the number of variables at play, everyone is scrambling to figure the market out.

All I know is this. Ground stability is needed to build a house or a business. That stability doesn’t exist today.

In the past, such stability was delivered when bubbles crashed. The dot com crash in 2000 and the financial crash in 2007 stabilized the dot com era craze and the Web 2.0 craze respectively. Is market-induced stability coming? Possible. The faultlines are visible. A couple of visible fault lines: the war (which is about energy, a core ingredient of AI), and investments on hold (Middle East sovereign investors are among the largest AI funders).

At this very moment, no one is winning, and everyone is trying to figure this out.