A knockout at the XBOX-ing ring of death
Plus: the anatomy of the AI bubble
Our goal with The Daily Brief is to simplify the biggest stories in the Indian markets and help you understand what they mean. We won’t just tell you what happened; we’ll tell you why and how too. We do this show in both formats: video and audio. This piece curates the stories that we talk about.
You can listen to the podcast on Spotify, Apple Podcasts, or wherever you get your podcasts and watch the videos on YouTube. You can also watch The Daily Brief in Hindi.
In today’s edition of The Daily Brief:
A knockout at the XBOX-ing ring of death
Microsoft’s gaming strategy is under pressure after years of costly acquisitions and its Game Pass push, highlighting how fierce competition, hardware challenges, and changing gaming economics have left Xbox struggling to keep pace with Sony and Nintendo.The anatomy of the AI bubble
A new BIS paper argues that the AI race is encouraging hyperscalers to overbuild data centres and AI infrastructure, raising the risk that intense competition today could create a painful industry-wide bust if demand fails to justify the investment.
Points & Figures by Zerodha
We’re always chasing the day’s biggest stories. But every now and then, we come across a dataset that deserves a closer look than a Daily Brief allows.
That’s what Points & Figures is for.
It’s where we step back from the news cycle and use data visualisations to tell stories about the Indian economy, financial markets, and investing. Our latest edition traces how India’s power landscape is transforming, using electricity generation, capacity, and emissions data to reveal why simply building solar panels isn’t enough to dethrone coal.
Compliance cannot override the investor's interest
Should fund managers always follow regulations, even when doing so hurts investors? Sandeep Parekh examines the Supreme Court's ruling in the Kotak Mutual Fund case and argues why fiduciary duty shouldn't take a back seat to procedural compliance.
A knockout at the XBOX-ing ring of death
We can finally say that we got a gaming industry story on The Daily Brief before GTA 6!
It’s been one hell of a year for the Microsoft XBOX, one of the top three video game consoles in the world. And we’re still half a year away.
Microsoft is letting go of 3,200 employees across the XBOX division, the biggest layoff in XBOX history. The company is also shutting down or selling off multiple loss-making game studios. This decision came only a few months after XBOX got a new CEO, Asha Sharma, who previously led Microsoft’s CoreAI division and was COO of Instacart before that.
This brutal restructuring points to an undeniable truth: Microsoft’s gaming strategy has misfired badly. Over the last few years, the tech giant went on an unprecedented acquisition spree, and introduced Netflix-like subscriptions to gaming. So, why is one of the wealthiest tech firms on earth retreating from a war it just spent billions of dollars to win?
There are plenty of reasons and missteps. We won’t be able to cover all of them, but hopefully, with this story, you get enough of a glimpse into how much their own DNA as a software company mattered to this point in their lifecycle.
Razor-and-blades
Before we look into Microsoft XBOX standalone, it’s worth zooming out into the game console industry as a whole.
In 2001, Charles Herold, a video game journalist, summarized the history of this industry in one succinct, but scary sentence: “Historically, whenever three consoles have competed, one has died.”
Whenever a fourth or fifth major console manufacturer has entered the fray, one has eventually been bled dry and forced out. In the 1990s, for instance, when 3D graphics became popular, there were 5-6 players who tried to cash in on this wave. Out of this, only three remained: Sony, Nintendo, and Sega. Then, a few years later, Microsoft nudged out Sega, which had already lost tons of money in the first 3D graphics war.
But why is this the case? Why does the market reset to this equilibrium so often?
Well, the video game console business is a two-sided market platform. On one side are gamers, and on the other like game developers. Gamers will only buy a console when it offers a solid catalog of games, and developers will only make games for the system when there are enough gamers. More competitors fragment the user base, and also increase costs for developers, since they have to tweak their games for multiple systems.
So, the goal is to start a feedback loop where more consoles sold leads to a larger user base, and therefore more developers come in, and so on and so forth. To kickstart that feedback loop, console manufacturers often price their devices aggressively, even at net operating loss. They offset this by charging game developers high royalty fees on each copy of a game sold.
This model was first truly popularized with the razor and blades, as the competitors of Gillette wanted to steal market share from the pioneer. Much like the razor, the console hardware is sold at a loss. The actual profit comes entirely from the blades: in this case, royalty fees from each copy of a game sold.
A software giant’s hardware gamble
Any company that enters the console business has to ace both software and hardware. Which is where Microsoft has a far more unique story in comparison to its competitors.
Think about it. Microsoft built its empire on the back of software alone. Unlike, say, Sony, it had little to do with manufacturing devices until the XBOX. Microsoft’s entry into the gaming market was motivated by Bill Gates’ fear that home consoles would eventually encroach upon the PC business. So, Gates approved the original Xbox project with the intention of building a “PC-in-a-box“.
Microsoft’s software dominance helped it move fast. The company leveraged DirectX, its dominant programming interface for graphics processing, and built the console around it. This made it incredibly easy for PC game developers to transition to the new platform, allowing them to develop games for the first Xbox much faster. In fact, as per Ed Fries, who was a founding member of the XBOX team, the first model was supposed to be called DirectX-Box.
In contrast, developers for the Sony PlayStation 2, which was part of the same generation as the first Xbox, initially found it notoriously complex to write code for the system.
But this “PC-in-a-box” approach came with fatal structural flaws.
To begin with, both Sony and Nintendo were Japanese hardware-first companies who owned or co-developed much of their electronic stack. For instance, the PS2’s custom CPU and GPU chips were built in partnership with Toshiba. And it was this ownership that allowed Sony to fuse them and shrink the size of the combined chip, which used less power and was cheaper to mass-produce. Nintendo had similar arrangements with IBM and Matsushita.
Microsoft, on the other hand, did not initially own the IP for its hardware. In fact, it took the “PC-in-a-box” approach too seriously — the first XBOX relied on generic PC commodities, such as NVIDIA GPUs and Intel CPUs. Microsoft could not modify its chips to drive down costs or make it suitable for a console.
Because Sony and Nintendo controlled their silicon, both enjoyed a learning curve that steadily reduced manufacturing costs over the lifespan of their consoles. Microsoft, buying off-the-shelf parts, did not.
Later, Microsoft did repair their lack of ownership, with the XBOX 360 indeed having its customized chips. But even then, there was another aspect of their approach that spoiled the release of the XBOX 360.
It was that Microsoft gave hardware the software treatment.
For instance, in 2005-06, desperate to beat the Sony PS3 to market, Microsoft rushed the launch of the XBOX 360. It reportedly shipped out consoles knowing there was a critical flaw that caused the hardware to run too hot too fast. They assumed that, much like a buggy Windows update, they would be able to issue a fix later. But you cannot patch melting silicon with a quick code fix — it takes much, much longer to fix manufacturing issues.
The result was a massive wave of console failures permanently burned into gaming history as the “Red Ring of Death”: each faulty XBOX 360 would show three red quadrants close to the power button. The extended warranty program to repair this cost Microsoft $1.15 billion to fix.
Winning the battle, losing the war
Despite the failures, the Xbox 360 and PS3 fought to a near tie in lifetime sales, with each moving around 85-90 million units throughout their production run.
Both were comfortably beaten by the Nintendo Wii, which had reached 100 million units before either Sony or Microsoft touched 85 million. They played a different game by choosing to not optimize for the highest frontier of gaming technology, and focusing on a younger, more casual-gamer demographic compared to Sony or Microsoft.
However, by the next generation, the gap between Sony and Microsoft would widen significantly. The turning point occurred during the 2013 Electronic Entertainment Expo (E3).
At E3, Microsoft announced that its upcoming Xbox One console would require a daily internet connection. It also detailed a strict digital rights management (DRM) system that prevented the resale or renting of pre-owned discs. Microsoft forcefully bundled the Kinect motion-tracking camera with the XBOX One, making the console too expensive. Gamers violently rejected these proposals.
Sony swooped in and took the exact opposite approach. They launched the PS4 at a price point that undercut the Xbox by $100 and heavily championed physical game ownership. It also made the hardware more friendly for game developers. Furthermore, at launch, gamers noticed that XBOX One’s performance was noticeably weaker than the PS4.
What also helped was that Sony also had its own established game studios that consistently produced highly popular titles which were entirely PS-exclusive. These include Spider-Man, The Last of Us, Ghost of Tsushima and God of War.
Microsoft lacked the internal capability to produce its own beloved games on the same scale. Many of the games that were meant to be XBOX exclusives, like Fable Legends, were canceled. Microsoft did have the beloved first-person shooter franchise, Halo, but its later editions were so buggy that the developers had to issue a public apology.
With this generation of consoles, Sony had decisively won the field against Microsoft. By 2022, the PS4 sold over 117 million units, while the XBOX One had less than half that number.
Two radical pivots
Realising it had lost ground in the console war with its missteps, Microsoft made two major strategic changes.
First, in order to create a line of exclusive XBOX games, Microsoft went on a spree of game studio acquisitions. One such acquisition was Bethesda, the makers of The Elder Scrolls and Fallout franchises, for $7.5 billion. Then, in one of the biggest acquisitions in its history, Microsoft took over Activision Blizzard, the makers of Call of Duty, World of Warcraft, and owners of Candy Crush Saga, for ~$69 billion.
Secondly, Microsoft introduced a strategy that changed the industry forever: a subscription service for gaming. Perhaps, it was best suited to do so, as a firm that earned its bread-and-butter through recurring software and cloud revenue. In fact, Microsoft leveraged its world-class Azure infrastructure to build the cloud capabilities of the Game Pass.
This even forced Sony to revamp its own subscription offerings. For once, Microsoft had some semblance of an upper hand.
Self-cannibalization
However, these key changes in Microsoft were, sadly, in vain.
In the short term, the subscription strategy appeared successful. Console sales surged, Game Pass subscribers increased, and the game catalog got better.
However, Game Pass ultimately cannibalised Microsoft’s own buyers.
Given the choice, consumers obviously preferred a cheap monthly subscription over buying expensive individual games. Over the long term, this crunched XBOX’s operating margins down to a meagre 3%. And while Game Pass exploded initially, subscriber growth has slowed dramatically, and is now stuck at ~30 million subscribers — a far cry from XBOX’s 2026 projection of 77 million.
The Game Pass was being torn apart by a single tension: its incentives and that of the game studios began to clash.
See, developing a modern video game is highly expensive, so its physical copies would also be expensive. But launching it on a subscription service that’s far cheaper than the physical copy would simply depress potential revenue. Third-party game developers were strongly against launching their games on the Game Pass on the first day of the game release.
This, as per tech analyst Ben Thompson, is also partly why XBOX bought so many game studios at once. It would help them bypass this problem, and promise subscribers a “day one” release. However, these acquisitions bloated their balance sheets, most likely at a hefty premium. With subscriptions, there was no way XBOX could reap their acquisition value anytime soon.
To make the finances work, Microsoft was forced into a corner. They had to start launching their games on Sony and Nintendo’s platforms, thereby destroying the entire point of game exclusivity.
Game over?
In hindsight, a software-first firm having a hold this long over a hardware-heavy business is truly impressive. Microsoft changed the market like no other with subscriptions. It is forcing console makers to do away with physical sales.
Yet, the economics of producing games still haven’t aligned with offering them at a flat monthly rate. Meanwhile, the Sony PS5 and the Nintendo Switch are among the best-selling consoles of all time. Sony’s own subscription service reached record profitability last year.
There is still no clarity whether their restructuring exercise will help them survive. They’ve effectively given up on game production capacity by laying off developers. The PS5 already outsold the Series S and X combined 2 years ago. Plus, consoles will only get more expensive with memory chip prices skyrocketing.
Historically, whenever three consoles have competed, one has died. That might just still be true today.
Anatomy of an AI bubble
This year, the major hyperscalers — the American giants that dominate cloud computing — will spend more than $700 billion on capital projects. Much of that will go into data centres and AI chips.
Beyond a point, big numbers stop making sense entirely. But let us explain what that means. Think of every single rupee the Union Government spends in an entire year. Think of every highway it builds, every salary and pension it credits, every bondholder it pays, every subsidy it grants, every agency it runs, every rocket it launches, every military base it maintains, every school or hospital it runs, everything. All of that costs a hundred billion dollars less than what a handful of American companies will spend this year.
How do you even begin to conceive the effects of such an incredible wave of spending? If its scale is so far beyond the limits of human imagination, how can one even imagine how things could go wrong?
A recent working paper from the Bank for International Settlements — the central bank for central banks — is a careful, formal attempt at conceiving what is happening. Its author, the economist Phurichai Rungcharoenkitkul, argues that the industry is building far more than it should, not because of genuine need, but because the incentives push each company to try and one-up its competitors. This dynamic, however, could turn into a time bomb for the entire sector.
To be clear, the paper isn’t predicting that the AI bubble will collapse anytime soon. It simply points to why the race to build AI infrastructure is creating conditions that would make a bust, if it ever comes, much worse.
Why the industry overbuilds
The paper argues that there’s a peculiar dynamic to the current AI build-out: the major hyperscalers aren’t chasing demand, as much as they’re trying to outdo their rivals.
This isn’t vanity alone. AI, the paper argues, is being played as a ‘winner-take-most’ contest. Everyone is betting that the firm that reaches the frontier first shall capture most of the users, the data and the revenue. The rest will have to divide what remains. An early lead can lock in users, generate more data, and attract developers who want to build the best models.
In a contest like this, companies spend dollars on compute for two reasons. For one, it buys them real capacity — better models, and more bandwidth for inference. But more perversely, it also makes it likely that the improvements in performance shall pry customers away from their rivals. The industry is entering conditions of structural overcapacity.
Beyond a point, what’s good for a company might be bad for the industry as a whole. It might make sense for Lab A to try and steal a customer from Lab B, but the revenues flowing into the industry as a whole, will remain the same. Effectively, the industry keeps spending an ever-larger sum of money to wrest the same pool of revenue away from each other.
The paper tries to size this waste. In its model, even by conservative estimates, if five labs were to keep competing this way, they would end up building one and a half times the capacity the market could actually use. By less conservative estimates, they would end up building thrice the supply the market could absorb.
This gap — between what the industry spends, and what the market needs — is what the paper sees as over-investment.
That said, it’s important to remember that this is just a model, based on the author’s estimates. It assumes a certain trajectory for how the industry’s revenues grow. Those assumptions may or may not be right.
The trap of building early
All businesses put money into building assets before they know if they’ll ever find the revenues to match. Even your local shopkeeper spends on putting up a shop before it’s completely clear if it will pay for itself. Any business is a gamble.
What, then, sets AI apart?
To the paper, the biggest issue is that the AI race doesn’t let individual companies wait uncertainty out. For most ordinary businesses, this is a key lever. You can choose to only act once you have a degree of certainty. If you aren’t there, you can wait for the fog to clear.
To an AI lab, however, waiting looks unusually costly, because the first firm to reach a certain threshold wins the entire market. Any firm that waits to deploy its cash risks entering a market where the users, the data and the developers are all committed elsewhere. While patience is usually a virtue, here, it carries a competitive cost.
So firms must commit before the payoff from AI is known. And worse still, much of that commitment is financed with debt. As a result, the industry is sitting on a large stock of debt-funded capital, sunk into machinery that can be used for few other things.
The bust
This creates a board that looks something like this:
One, there’s machinery. To the paper, AI infrastructure is highly specialised — valuable in serving AI demand, but hard to redeploy anywhere else. Companies could reuse the land, buildings and power connections, but the chips themselves have few other uses.
Two, all of this is financed by debt — and that cash that is already committed.
On top of this, as we’ve written before, these labs are tied to each other in a massive web. A weird sort of arrangement is flourishing across the industry: companies with data centre capacity are “investing” in AI labs, but those investments are going straight back into buying the investor’s own computing power. Some $46 billion of equity capital has already been deployed this way. Alongside it sit commitments to buy $879 billion more of compute in the coming years — money that hasn’t been spent yet, but is promised on the books.
The more money that is committed like this, the better returns AI must deliver tomorrow. If the industry’s revenues justify the spending being carried out today, you will see more rounds of investment, and the boom shall continue. But it doesn’t, there will be a problem that cascades across the industry.
One, the debts will come due. New infrastructure will no longer be built. These companies will be forced to sell their machinery to raise the money they need to repay their dues.
Only, the only companies that buy AI equipment will be other AI companies. If these companies are all currently over-investing at the same time, when the crash comes, everyone will be in the same bind. Just as companies will be forced to sell AI equipment, all the other potential buyers will leave the market. According to the paper’s models, if this is to happen, that capital will lose half its value at once, with losses rising as high as 80% as more equipment keeps hitting the market. The more the industry borrows now, the worse it shall be.
This is a pattern you see in many market busts.
The problem, however, is that the entire industry is interconnected through financing networks. When a lab fails, the stakes of its investors shall lose value. Those stakes, however, would have been collateral for loans. So, as labs fail, their investors lose borrowing power. They, in turn, would have to liquidate their investments elsewhere in the industry — which will pull compute off the table for other labs. Because the industry is currently rigged together, in case of a crisis, the contagion will travel all across the industry at once.
To be fair, this isn’t a forecast. It’s simply one stylised map of how a crisis could happen. According to the author’s models, there is roughly a 50% chance that the industry will see such a bust. But that’s far from certain.
What the model measures, and what it doesn’t
The paper is interesting as a stylised description of what could happen. It’s based on assumptions, though, some of which are worth questioning.
Here’s one example. The paper projects that over the course of its simulation, the industry will process anywhere from seven to twenty-fold as many tokens as it currently does, while revenues shall only grow by three to four times. Even assuming that’s true, though, the paper doesn’t account for how much cheaper those tokens could get. After all, you can process more tokens by running machines harder, or improving the quality of the underlying software, without increasing how much you invest. The paper’s main assumption — that the industry is building excess capacity — sits on these assumptions.
There’s also a deeper difference. The paper’s models of the ideal amount the industry should build are built around its revenues. It’s completely possible that this spending creates tremendous value for the users of AI, without showing up in the industry’s revenues. If that’s the case, a financial crisis could still play out as it suggests, but the build-out could still have longer term benefits.
This is a pattern we’ve seen before.
Take the railway bubble in Britain. Investors spent too much money on building Britain’s railway infrastructure, and suffered heavy losses. But the greatly expanded rail network it created carried its economy for generations. Or consider the dotcom crash. When it happened, it wiped out vast amounts of shareholder wealth, and saw company after company go bankrupt. But the fibre lines it left behind ultimately let millions of people access the internet for much less than it would otherwise have cost.
History is filled with examples of some people being ruined financially, only for their investments to be hugely valuable to those who come later.
Will that be the case with AI? We don’t know. Over the last year, the cost of a token has fallen to a tenth, even as its usage has soared. Does this mean too much has been built, and firms are selling under cost? Does it mean technology has improved, and the advantages have passed on to consumers? These are hard questions to answer.
And it’s harder, still, to predict how all of this will eventually end.
Tidbits
1. The District Consumer Panel of Raipur has ordered Maruti Suzuki to replace a customer’s car in the first-ever legal fight regarding E20 fuel. The car owner claimed the fuel caused issues, but Maruti Suzuki plans to appeal the decision to protect its reputation.
Source: ET
2. Chinese LLM lab Moonshot released Kimi-K3, which has hit the number one spot on the Frontend Code Arena test, beating Anthropic’s Claude Fable 5 on certain benchmarks. It also scored incredibly high on a separate technology test, ranking just behind the top-tier GPT-5.6 Sol model.
Source: Techmeme
3. The Serum Institute of India (SII) signed a deal with the Gates Medical Research Institute to manufacture a highly promising new vaccine for Tuberculosis (TB). If final testing goes well, this could become the first new vaccine to protect adults from lung TB in over 100 years.
Source: BS
4. Wipro reported a nearly flat consolidated net profit of ₹3,352 crore for Q1 FY27, up just 0.6% year-on-year, even as revenue jumped 10.6% to ₹24,479 crore. The outlook for Q2 remains muted — the company guided for constant-currency IT services revenue between -1.5% and +0.5%, reflecting continued caution in discretionary tech spending by clients.
Source: The Hindu
5. SBI successfully raised $1.9 billion in foreign money using a special financial scheme, leaving other competing banks far behind. This move gives SBI a huge amount of extra cash to fund its large projects and boost its lending power.
Source: ET
- This edition of the newsletter was written by Maine & Pranav.
What we’re reading
Our team at Markets is always reading, often much more than what might be considered healthy. So, we thought it would be nice to have an outlet to put out what we’re reading that isn’t part of our normal cycle of content.
So we’re kickstarting “What We’re Reading”, where every weekend, our team outlines the interesting things we’ve read in the past week. This will include articles and even books that really gave us food for thought.
The Chatter by Zerodha
Our team at Markets spends a lot of time reading earnings call transcripts and listening to management interviews. Along the way, we come across plenty of interesting insights that are worth sharing.
That’s what The Chatter is for.
It’s a weekly newsletter where we dig through what India’s biggest companies are saying and bring you the most interesting insights into businesses, industries, and the wider economy.
Join us on WhatsApp, where we share interesting soundbites from concalls, articles, and everything else we come across throughout the day. You’ll also get notified the moment a new video or article drops so that you can read or watch it right away.
Thank you for reading. Do share this with your friends and make them as smart as you are 😉
Thanks for the insights !
Great post!
Slight correction - in one of the tidbits, you attribute the Claude Fable 5 model to OpenAI. The Fable 5 model is Anthropic's.