Microsoft (MSFT) / Open AIs

Let’s be real, the tech industry is not short on ego. We’ve seen behemoths rise, fall, and sometimes stubbornly stick around like a bad rash. This cycle of disruption isn’t just some Silicon Valley soap opera, it’s the reality of capitalism. We’ve got new players vying for dominance, each with their own swagger, strategies and shareholder value projections that look more like sci-fi novels.

The CEO’s Code A Decade of Data

It’s a narrative so outrageous, even Hollywood might deem it a touch improbable. Imagine arriving at a company seemingly stuck in a 2008 time capsule, weighed down by outdated tech and even more outmoded perspectives. That’s not some startup drama but what happened to Microsoft circa 2014. Now, consider a decade of corporate transformation of that size and reach. We are talking a $3 trillion rise in value. How did this turnaround go down?

One central lesson, as revealed during an amazing interview is to stop with the envy-based strategy. Don’t be that desperate dater who tries to mimic the cool person, always following someone else’s trends. When Microsoft hit on the web-browser, that did not guarantee they also would find their way in the search business (although, with some irony, their competitors did — Google used browser user interface conventions as the basis for their search engine!). As if that was not lesson enough: later on the company had another try and did not win on mobile. The answer, according to the guy who actually turned the whole place around was doing things that makes sense given the permission to win the company had already gained with their customer base.

You know the type, a decade spent wrestling with legacy systems and changing the company’s perception, even internal cultural challenges — which reminds me of a painful MBA exercise, but in this case, very, very profitable. Here, we learn how the key was the strategy and focus that emerged during that first pivotal phase shift. One, to be specific, that has enabled an old brand to become one of the main winners of a newly created technological arena! As one executive famously said when asked about how he runs things at work: pattern match for success; discard everything that led to a standstill.

The 10-page Manifesto That Launched a Ten-Fold Increase. Turns out, our subject, while contemplating a potential shift of the corner office’s location wrote a detailed memo. A very long one. It was no haiku, it ran for pages. This memo laid out his strategic compass that would serve as the guiding map during his leadership. Among his ideas was this odd notion that the market for the company should not be described along the traditional “Cloud” divisions and instead should look a strategic layers and business lines, on top of core Cloud infra. Also, he identified his “structural position” and permission from customers. Turns out, most of those customers also wanted MSFT to prevail.

The other idea was a cultural reset. One very memorable analogy in this regard — during Microsoft’s early 2000’s the market praised the firm. A team stroll through the MSFT campus after this positive reception was met with nothing short of hubris by many folks including the later CEO of this story! Now, a growth mindset — one that’s more about being a ‘learn-it-all’ than a ‘know-it-all’ — took over in 2014 when the memo author took the steering wheel. His idea — do things where your existing customers need you. This was very helpful at the beginning as the CEO was well know at the company and could lead cultural change with that knowledge at hand. That cultural strategy has helped this big-tech change into what it is today.

In that context the fact he got on board fast enough to catch “the last train” when cloud emerged may very well make him “the best CEO hire of all time”! So next time, don’t hesitate in crafting long memo outlining what could be the possible future course, because, in very, very rare cases that move can prove to be very, very useful indeed.

Riding the Waves of Tech Disruptions

Now, let’s get real about AI. We’re knee-deep in a technological deluge, the latest phase shift in our collective existence. Forget Y2K; this is the dawn of AI where we try to make sense of language at scale — you can imagine a similar process if your average cat try to “decode” human speech by focusing on single sentences instead of the overarching plot that takes place from a series of them… Our protagonist quickly grasped that there were structural changes coming with an old ambition being fueled by tech at its latest iteration.

The way Microsoft approached AI wasn’t just chasing the shiniest new toy on the playground (Google’s Deep Mind acquisition in early 2010’s) , it was a very old focus in the house but taken through a new lens of computation. MSFT always had invested in Natural Language Understanding since early days. One of the first internal labs in the Microsoft’s MSR research department worked on voice and natural user interface — before “AI” was even a word. So the move for AI was in some sense a bet on the deep capabilities of existing people that had the vision and know how. It had also another bet at it’s heart: what was the single area in that new landscape that will exhibit nonlinear exponential growth? If your company had all the underlying building blocks it might also prove to be a pretty smart business move as well. This approach led MSFT to a small AI startup founded by two crazy fellas who just wanted credits on Azure, namely Open AI! It seemed to be a smart move from MSFT as that also reinforced the basic “value-add” proposition when considering their own customer base. So no, this was not simply about having the most impressive toy, it was more about leveraging the market forces at your disposition and seeing how far that would get you. The real point is, once you “pattern-match” those moves: double down! Which is also the story of the other key turning point that cemented that shift: once it became clear the Github’s CoPilot was indeed an exceptional product the firm committed deeper and deeper on the technology! And also when seeing how the first internal deployments were successful Microsoft got really enthusiastic about it!

There are other angles to it though! The entire competition is playing this AI race as one open world battle. We have other key players — Meta with its LLAMA project, Elon with xAI. This ain’t a lonely run: everyone’s aware of what’s at stake, so you’ll have other big brands as Google also vying to seize control. Even though you may have competitors at all levels they all need core infrastructure, access to GPUs and training models! This, as history reveals, might even have the consequence of producing multiple winners across the tech value-chain as we can foresee how this would play out if multiple “big players” battle with their own infrastructures! The other, second conclusion: network effects still matter so, whatever happens on the foundational levels the application layer will continue to bring immense benefits for those who can best cater their client needs, so Microsoft should look out for the startups that manage to master the space as there might be new, fresh winners that could appear! The third point — there is value even in what can be called “failed tech experiments” (remember “Microsoft Browser” lost to Google’s Chrome): that loss creates an opportunity for fresh starts on old challenges and also is a very good way to make sure a company never takes anything for granted ever again! As an executive in the story stated: sometimes, you need a loss to “win back some share”. So embrace some of your former failures, that is part of your learning trajectory, but also, remember: this game, in the words of that very person, has no rules: “watch for the one who adds to it”.

This may have seemed simple when that was stated; But the AI wars won’t just be a matter of deep learning algorithm capabilities and infrastructure dominance! We can even forecast where consumer behaviors will gravitate towards “agents” instead of classic interfaces; chat as answers will most probably overtake traditional search. It will shift even commercial and advertising spaces to new levels. This new technology will create a shift on business behavior: every query could lead to action so many apps would potentially disappear completely as interfaces as everything would potentially get unified via these agent systems. The agents could operate by a novel “license system” similar to what happens in many Enterprise environments or might move to other business models we haven’t seen yet. So, in brief — we may see agents orchestrating every system in an unseen fashion. And the main focus of every player will then shift on who’s allowed to use the other players software as a core challenge that will have repercussions across everything in the market. To stay ahead in this environment each firm needs to double down on both consumer and enterprise apps but must also realize that any AI system that acts on real data needs trust built into it at core, and trust for other businesses using it is absolutely primordial, both for your own product offerings and other market’s opportunities as well.

The ROI on Exponential Ambition

Let’s address the trillion-dollar elephant in the room, Return On Investment! It turns out memory may be almost unlimited according to one source (one of the interviewers referred to as Mustafa, that may have very well meant Mustafa Suleyman, co-founder of Inflection AI). That would be game-changer indeed, as the biggest challenge is still that AI tends to “forget” everything at every session it has. Memory could change absolutely everything by providing that persistent aspect needed to leverage these “agents” further than it’s currently feasible.

It was also a common sense move when thinking in terms of real life application; people don’t only seek “knowledge” as abstract objects they want actions performed via these tools and for actions, persistent memory matters tremendously. There are new challenges with that though: not all memory is the same as its form also matters, we are shifting to AI native schemas to get the best use of every information chunk we are adding into this new type of storage. Even if you may have many capabilities there, the other biggest thing in front is computer use where one should think who decides what will be permissible as agent actions, will that be the system user, or system itself? It also turns out the AI may be used to make business operation a breeze via very useful AI implementations on M365 or for even sales management as you might find that “you need agents, all over, in every department” as we seem to be heading into an “agent first business world”. Microsoft is, in this very moment trying to “collapse all the apps” with their own agent tools so that every department on the organization has its own co-pilot using excel for number crunching, using legal writing as specialized “pages” app or anything they could have previously done manually via all of these automated capabilities that we didn’t had access yet!

So where is all that taking us? Is MSFT spending a lot on Capex? Yes. More like an industrial company. Like steel firms with software as an extra topping! Do I expect those returns on capital employed? According to that conversation, it is actually going well in that new terrain. But it’s very expensive to keep up with tech as, in addition, they may well get all of their core supply lines with GPU’s as “a little short” on time! All those things taken into consideration is hard to manage and might be quite complex in relation to forecasting cycles, inventory etc.

Now: about scaling those new large language models (or LLM’s as people refer to them). Do you follow the old “training” route, keep spending fortunes, doubling down even more money to simply double the performance on hardware each and every year, or something entirely new like, 01 architecture — the one openAI is currently developing? Well, according to MSFT leadership “scaling laws will be hard to sustain” from purely exponential perspective due to physical constrains — however a new technique emerged by which one makes the “inference itself another scalability tool by reusing what is already inferred, using these data points on test time compute to improve future rounds”. That could well change, or accelerate the new models by a factor “billion x”, maybe something more if all went really well. So: “do not bet against scaling laws” seems to be good advice while one is aware that these same scale models won’t win network effects: application and customer focus is much more important in the game than just a great model on it’s own, and many clients might have different requirements along their processes that can’t be addressed with simply one model that would run perfectly in one environment but would simply become cumbersome in another, and all the tech is going, with that notion of specific apps that caters diverse organizational business flows, on their own distinct niche, while at the same time acting in synchronicity across every function they could integrate with. This type of diversity in “models” may be a “good problem” to solve, as those who master that terrain may very well end up having very profitable returns while also leading other “value chain opportunities across many lines of software application!” In a way we need “Multiple providers” instead of “one king” type of architecture where it has been made clear that that model (the only big winner) doesn’t always mean that market opportunity is better, faster, or delivers the most return.

Microsoft may have “outsourced some of the core capabilities because no plan could have been implemented to manage the boom that GPT created when it was launched” so that explains why it went a bit short of infrastructure in the last few months due to what is called a “short term catchup phase”, that is currently taking place as of the time the interview went online. They’ve decided they don’t want “two unnecessary training runs” because “they will fully leverage the power of AI” along multiple levels but will try to improve at “inference” due to some core value they might unlock in that specific approach. As of now Microsoft are NOT supply constrained for computer chip components but yes to a “certain amount of power that may still create challenges for the entire production capacity”, or for, what’s being referred to as “that level 2 type thinking” — “using those inference derived tokens and cycling them into your original models for better refinement.” In order to really unlock “AI” one needs, from Microsoft standpoint the app, data base and computer core hardware at very different levels of engagement in the stack — all acting together at the very same moment, or at a very close proximity in real life, in order to have these workflows as efficiently designed as possible with all the tech one could have under their hood! In that setting it has become clear that this whole thing with Ai is much more about “AI at the core of Cloud Computing as a new standard.” and therefore, they believe, that the real goal here is the “full AI applications over a diverse environment for various workloads in a heterogeneous world.”

This creates new interesting questions! Is the whole relationship between MSFT and OpenAI actually that unique? Will it persist in the future, given their specific set of dynamics in relationship with another large player in the market (that very “apple search deal”!) — a “coopertition” situation where everyone may benefit by having each other (one for access, another for product value-chain)? Do they operate more like customer and supplier, an investor relationship, or even an “IP owner/operator combo”? Is it important to see a clear demarcation of the two “companies” (a more obvious move now considering one of them — OpeanAI- also seeks funding, restructuring itself) so they get to grow in their own specific fashion with unique values in all senses of the term? Well, even that the other part did answer quite a bit about “their positions on supporting their choices” as “at core both parties are working towards the long-term perspective!” And for one, they still hold firmly at their view: one is “the most relevant company in the new shift”. So “they support whatever choice is more interesting from that specific point of view”.

Let’s think ahead about the concept of openness of AI, versus models that remain more enclosed on “one central command post”. Meta created Llama models so anyone can get in. Are open approaches better or worse? Well, as far as our main player seems to see it — it is not necessarily about religion; “open vs close is not more than different business strategies at core in the realm of new products! Both strategies are meant to amplify “Network effects”. So if Meta believes to create one shared architecture akin to how “Linux revolutionized operation systems” so they commodify its own tools to capture “everyone, in one side” this, from a strictly tactical level is, a smart thing to do, even when your company owns “one model!” Both open and closed strategies can be leveraged if one believes that it fits better for a certain purpose at hand. At last: the issues around “AI Safety” and regulatory needs will be applicable for each tech firm, so there isn’t one superior option in that aspect as this is what States and regulators need to get on their work. And all models are able to get to an equivalent level of “safety controls” regardless if they’re closed or open source and the market might have the advantage of not depending “only on a single one” (i.e., having diversity is, usually, better than lack of diversification!) which may also increase, again, “Network effects all across every sector.”

The big thing: even the smallest model can easily “reverse engineered” or simply copied so, in that sense we must also accept this is the game for all who decide to engage with it! All in all; Microsoft wants to make sure every part of that new “AI shift” is addressed in all ways! The real game plan: leverage technology to its fullest and bring to market all of that.

A Call to Action That Is Not All Tech

This all seems complex but what it truly calls to action in every possible dimension we engage in is: act with more responsibility. Not solely based on technology but rather from what I see as “all possibilities given what is truly possible”. It is our common duty and purpose as humanity not only to ask but more so, to demand the absolute best from ourselves and everything around us while always holding high a sense of human awareness, ethics and also with all its limitations that can very well give direction to that “best path” one should consider following and thus lead. After all it’s no one’s, or “the company’s” technology but, “ours and for all the world” by giving what only humans could deliver and all the tech will only, merely, enhance. Let’s get to it!

More shiny stuff on platformeconomies.com!