Nvidia's Reckoning? AI Bellwether Faces Moment of Truth Amid Tech Correction

This week, Wall Street collectively held its breath to determine if the AI momentum trade still has legs or if we're witnessing the beginning of a significant correction.

Nvidia's Q4 FY2025 earnings have landed at a crossroads for the company and the broader tech market.

Now, the moment of truth has arrived for the undisputed heavyweight champion of the AI revolution.

Before we look at the key points in the earnings and takeaways for investors, let’s understand the run-up to today’s earnings.

Market Context: Paradise Lost?

The stark contrast between broader market metrics and tech-specific performance tells a tale of diverging narratives.

While the S&P 500 and Nasdaq churned to higher-highs for the tail end of 2024, the large divide between the Magnificent 7 and the remainder of the market was clearly closing.

Collectively, the Mag 7 stocks make up 33% of the S&P 500’s market capitalisation and contributed over 50% of the S&P 500's gains last year.

Their dominance hasn’t just been through this AI cycle but has now been a decade-long feature of the markets.

However, a combination of the unknowns surrounding Trump’s tariffs and high consumer prices has sapped much of the fervour on Main St in the US. Consumer confidence just experienced its sharpest monthly decline in four years.

Bond markets and gold’s recent surge show a similar pause on Wall Street at this moment.

Meanwhile, the tech-heavy Nasdaq is touching one-month lows after four consecutive losing sessions, which brought the index down by nearly 5%.

Since last Friday alone we’ve seen:

1. MicroStrategy, $MSTR: -25%
2. Palantir, $PLTR: -28%
3. Tesla, $TSLA: -18%
4. Intel, $INTC: -10.5%
5. Nvidia, $NVDA: -7%
6. Alphabet, $GOOGL: -6.8%
7. Broadcom, $AVGO: -6.50%
8. Meta, $META: -4.4%
9. Microsoft, $MSFT: -4.2%
10. Amazon, $AMZN: -4%

These ten stocks collectively erased nearly $1.5 trillion in market capitalisation in just four trading days — a staggering figure that underscores just how concentrated market wealth has become in these AI-adjacent titans.

Overall, the market's foundation appears strong, but its most valuable pillars — the Magnificent 7 tech giants — are showing signs of fatigue.

Nvidia's Results: The AI Canary in the Coal Mine

Against this backdrop, Nvidia's results take on outsized importance. The company reported quarterly revenue of US$39.3 billion, up 12% from Q3.

That represented 78% growth year-over-year, with earnings per share of US$0.89, above analyst expectations of $0.84.

I won't bore you with all the details, here's what the quarter looked like from a birds-eye view:

Overall, many considered the quarter as 'just good enough'. A collective shrug for results that would be blockbuster for anyone else. These beats are largely meaningless for a company with such high expectations priced in.

What matters most isn't merely whether Nvidia beat estimates — a feat it has accomplished in 16 of the last 17 quarters — but rather what its results tell us about the trajectory of AI spending and the sustainability of its incredible margins.

The Golden Goose

Nvidia reported gross margins of 73% this quarter, falling for the third consecutive quarter and 150 basis points below expectations. A worrying, but not wholly unexpected sign.

There were expectations that these would come down as the company expedited its Blackwell production to meet demand. But that’s not the whole story.

For Nvidia, those margins have been the 'golden goose' of its success, showing the fruits of a near-monopoly position in an incredible growth sector.

For investors, these high gross margins are a helpful metric for assessing Nvidia's position when other metrics like revenues show incredible (and difficult-to-ground) results.

I think these margins give us the clearest indicator of Nvidia's pricing power and competitive moat.

Any significant erosion would signal that the company's near-monopolistic position in AI chips may be facing genuine threats, or costs are running out of control as they try and hold onto their position. 

Either scenario could dramatically alter the investment thesis for the entire sector.

So what do we see?

In the earnings call, Nvidia argued that falling margins were entirely due to the expedited roll-out of its Blackwell chips and the complexity around the many configurations they offered in their servers.

I believe that story is partially true, but there is more at play here. 

During today's earnings call, Nvidia's CFO notably evaded a direct question about whether next quarter would mark the bottom for declining margins before their anticipated return to the promised mid-70s range. I'll explore these gross margin projections in more detail shortly.

Before that point in the call, post-market trading saw Nvidia's stock up around 2.4%. By the end of those margin questions, Nvidia’s stock was down -0.44% in after-hours trading.

The margin story becomes particularly intriguing when viewed alongside emerging fears around DeepSeek, whose recent breakthrough models reportedly achieve comparable performance to those from OpenAI and Anthropic while requiring just a fraction of the computational resources — potentially as little as 1/45th the computing power.

If these efficiency gains prove scalable, some have feared they could eventually undermine the extraordinary demand for Nvidia's premium-priced chips.

I believe this is misguided.

Jevons paradox suggests that as technology becomes more efficient, consumption often increases rather than decreases.

First observed in the 19th-century coal industry, the paradox shows that efficiency improvements make technology more economically viable, thereby expanding its adoption and, ultimately, its consumption.

Applied to AI, this paradox indicates that more efficient models like those from DeepSeek are more likely to expand the overall AI market dramatically by lowering barriers to entry.

This enables more companies to deploy AI solutions and make entirely new applications possible.

Far from reducing the total addressable market for AI chips, these efficiency gains could accelerate adoption across enterprises.

All of this is to say that this cheaper-to-run AI is likely going to benefit established players like Nvidia through sheer volume growth.

But of course, this all sounds very academic. Have we seen this in principle recently? Yes.

Chinese tech giants Tencent, Alibaba, and ByteDance have all ‘significantly increased’ orders of Nvidia’s hobbled H20 — a chip specific to China due to US export controls — since DeepSeek shocked the market.

That uptick may be due to some distortion from the export controls, but their CEO’s messages have been one of aggressive expansion on the back of DeepSeek.

The rumour mill in Washington is that these H20 chips are likely to face the chopping block as the Trump administration considers the next wave of chip bans on China.

This has been a steady case of the US increasing the intensity of its chip bans, only for Chinese domestic hardware and software companies to again take the US by surprise.

As Huawei’s CEO Ren Zhengfei summed it up recently:

‘Our previous worries about lack of core technology have eased, I firmly believe that a greater China will rise faster’

So, the West’s concerns around China and AI have pushed many of our hopes on the next wave of chips from Nvidia.

Blackwell: The Next Chapter

The second key focus for investors in this earnings call was around the ramp-up of Nvidia's next-generation Blackwell architecture, the successor to its wildly successful Hopper H100 chips that have powered the AI revolution thus far.

Jensen Huang was at pains in the call to repeat that ‘demand for Blackwell was extraordinary. '

Figures for the quarter showed Blackwell sales at US$11 billion — their fastest product ramp in company history

However, talk of a production scale-up was vague at best. The best we got was an admission that the task to scale these higher ‘was very difficult’.

That’s a massive understatement.

Nvidia has around 350 plants making 1.5 million components for its latest Grace-Blackwell servers, which come from innumerable suppliers worldwide.

They are spending aggressively to expand that footprint, but they are reaching scales where things as simple as skilled labour could become an issue.

That will exacerbate if the West continues to split with China.

That issue becomes existential if Taiwan is disrupted in any way, as TSMC remains the primary manufacturer of Nvidia’s chips.

The sensitivity of those supply chains cannot be overstated here.

Geopolitics aside, the ramp-up of Blackwell production is an important one to watch in the future to gauge their expectations of demand and success in executing the market's dream.

Over time, Nvidia expects the job of building Blackwell to get simpler as supply chains improve.

Server complexity and configuration mixes should also eventually streamline as customers find clear winning mixes in their servers.

At least, that’s what Nvidia is banking on to get its margins back to the mid-70s.

I am far from convinced at this stage. Nvidia’s second-half ramp-up of Blackwell would have to hit double digits to reach its margins of old.

So, can investors trust the guidance on this one?

Forward Guidance: Crystal Ball or Smoke Screen?

For Q1 FY2026, Nvidia projects revenue of US$43 billion +/- 2%.

The company expects gross margins to fall to 71%, plus or minus 50 basis points, but painted a sunny expectation of improvements by January next year.

I remain sceptical that this margin turnaround here will be achievable in the short term and think Nvidia’s next few quarters will remain a tense proving ground for investors.

Investors are understandably skittish about what growth looks like in calendar year 2026-27 and how far the hyperscalers' capex will go. Currently, the mood is akin to a game of musical chairs.

But that’s not the only potential iceberg ahead. As I highlighted in my June 2024 article, ‘Nvidia needs more to go right than you think,' there were major issues with the narrative that Nvidia was going to be the only winner in the AI race. 

To save you from reading another article, those were:

• Inevitable competition coming with such high market share/margins
• The changing needs of AI (seen in inference-time compute focus in reasoning models)
• The shifting market for AI chips (seen in the beginning split of inference vs training chips)

In addition, Trump's tariffs and their impact on Nvidia's Chinese chips remain a huge unknown.

Revenue from China as a percentage of total income was flat this quarter, but overall, it has halved since the introduction of the first export controls.

The question remains if the West can pick up the slack from other geographies.

The US$500 billion Stargate project could be that great demand driver, but for now, I remain sceptical.

Trump is fairly well known for hyperbole in his figures — often calling Forbes to tell them how their estimates of his wealth are off by orders of magnitude.

While the final investment figures remain debated and the short-term impact likely overstated, the longer-term trajectory points to continued massive data centre expansions as the US competes with China.

US aside, we have seen similar promises by France and the EU, with data centre projects put forward this month totalling around US$324 billion.

But for individual investors, these figures are so large that they almost become meaningless —it's like trying to decide on a stock based on a country's GDP.

So, where should investors look today?

The Shifting AI Landscape

While market attention remains fixated on data centre buildouts by hyperscalers like Microsoft, Amazon, and Google, the next frontier for AI investment lies in enterprise adoption.

That’s easier said than done.

Transforming AI's theoretical capabilities into practical business applications that generate tangible returns on investment isn’t like flicking a switch.

This migration from infrastructure to consumption represents the true test of the AI story.

Microsoft inadvertently highlighted this challenge when an analyst report from TD Cowan this week reverberated through the tech market helping push the sell off.

The report revealed a ‘a couple hundred Megawatts’ reduction of data centre capacity in Microsoft’s lease program.

That was enough to put the fear of God into an already shaky tech sector — sending many of the major power suppliers and data centre buildout companies down double digits.

That move now appears to be an overreaction to prudent and necessary moves by Microsoft.

I would argue those adjustments were even inevitable — proceeding with every contract would have pushed Microsoft's capital expenditures well beyond its projected $85 billion for 2025.

However, in tandem with other signals, it raised questions about the pace of AI deployment.

In a recent podcast interview, Microsoft CEO Satya Nadella hinted that the company may be slightly getting ahead with its buildout.

This is a candid admission when considering the imbalance between Microsoft's expected $85 billion in 2025 capital expenditures against just $13 billion in related revenue.

I'm cautiously optimistic that the next phase of AI will begin to capture more revenues as the use cases blossom along with AI's general intelligence. But I could also be wrong.

I would argue that the capabilities seen in the latest models that I have tested make a strong case for functional and valuable improvements in AI assistants in the coming years.

I foresee a near future where functioning in daily life without an AI assistant seems as curious as not having a smartphone.

But who will serve that AI of the future, and what chips it will use, remains an open question.

Competition Looms on the Horizon

The extraordinary margins that have propelled Nvidia to trillion-dollar status may prove unsustainable as competition intensifies across multiple fronts.

While software ecosystem disadvantages have thus far muted AMD's competitive threat, several emerging players present more fundamental challenges if markets continue to shift

Things like:

• Extreme processing units (XPUs) like Cerebras and Groq have developed innovative chip architectures that sidestep Nvidia's interconnect advantage, with Cerebras utilising ‘wafer-scale’ technology that places orders of magnitude more transistors on a single chip.

Groq's deterministic processing approach has demonstrated inference speeds up to 1,320 tokens per second — dramatically faster than traditional GPU setups.

These chips are still very small-scale, so it's not going to change anything in the near term. But these specialist chips could be a larger factor if AI needs shift again, similar to the changes seen in the newer reasoning models .

• Custom silicon from hyperscalers like Google (TPUv6), Amazon (Trainium2), and Microsoft threatens to reduce dependency on Nvidia chips for internal workloads, potentially removing significant revenue streams. Even Apple has been developing its own AI chips for edge applications. 

The mix of internal versus external workloads for these hyperscalers is important to watch to determine the extent of Nvidia’s ability to capture revenues from the Enterprise AI push. If players like Meta shift more into their future chips and focus them towards these external workloads to serve businesses, their demand for Nvidia chips could also plummet.

• Software frameworks like MLX, Triton, and JAX are creating higher-level abstractions that could diminish the importance of Nvidia's proprietary CUDA ecosystem — potentially following the historical pattern where assembly language eventually gave way to higher-level programming languages.

The Investment Takeaway: Finding Value Amid Volatility

For individual investors navigating this bearish landscape, the key insight isn't whether to abandon technology investments but rather how to identify value opportunities amid the correction.

The era of indiscriminately pouring capital into any company with ‘AI’ in its press releases is giving way to a more discerning phase where fundamental business models and competitive positioning will matter enormously.

While hyperscaler data centre buildouts represented the first wave of AI investment, the more durable and potentially more lucrative opportunities may lie in companies that can harness these capabilities to solve real business problems — the picks and shovels of AI application rather than just the infrastructure.

As Warren Buffett famously observed, when the tide goes out, you discover who's been swimming naked.

The current market correction is revealing which companies have sustainable competitive advantages and which have merely been riding the AI hype wave.

For investors this is a fantastic time to embrace active rather than passive investing and find the companies with a head start in the next phase of the AI rollout.

That’s companies supporting enterprise AI rollouts and boosting productivity.

For my own bets, data remains the clear roadblock for company adoption. Pulling workflows and siloed data out of individual departments will be the first step for many companies that wish to embrace AI.

For those that don’t, expect to be left behind as more sectors face disruption.

This transformation remains in its early stages despite the extraordinary valuations already assigned to the current beneficiaries.

Whether Nvidia maintains its crown or gradually cedes ground to emerging competitors, the AI revolution marches forward — creating opportunities for discerning investors who can separate enduring value from temporary market infatuation.

If you would like to know what we think the next generation of AI winners will be CLICK HERE to subscribe to our FREE daily insights at Fat Tail Investment Research.

We cover market-leading topics like AI, gold, commodities, and macro trends.