Talking points: Bitcoin's meltdown and the AI 'chat wars'

What makes this downturn different from previous crypto winters is that the industry got everything it wanted. Institutional adoption arrived. Bitcoin ETFs launched. The regulatory environment improved.

Traditional finance embraced digital assets. None of it mattered. The expected wave of new retail investors never showed up, and without fresh capital entering the system, prices have drifted steadily lower since October.

The crypto market runs on narrative and faith. Right now, it has neither.

The product problem nobody wants to talk about

Beyond the price charts, the industry faces a deeper issue: after years of building, almost nothing has achieved mainstream adoption. Decentralised social media platforms failed to attract users. Blockchain gaming never broke through. The promise of a new internet-native financial system remains largely theoretical.

The one exception is stablecoins. Companies issuing and building infrastructure around dollar-pegged tokens continue to raise capital and attract acquisition interest. Stablecoin adoption has reached genuine scale, and the businesses behind them appear largely insulated from the broader crypto downturn.

Everything else is struggling. Smaller tokens and speculative projects face an especially grim outlook. Crypto exchanges are cutting staff, with Gemini planning to lay off roughly 25% of its workforce. Some companies are quietly pivoting away from crypto entirely, diversifying into tokenised stocks or gold to keep user capital on their platforms. When your own industry participants are hedging against your asset class, the signal is hard to ignore.

Bitcoin's underperformance relative to gold and silver raises uncomfortable questions about its "digital gold" thesis. If the primary value proposition is as a store of value, the comparison with actual precious metals, which are rallying, is not flattering.

AWS is winning the cloud war by the numbers

Buried beneath the capex panic, AWS posted 24% growth in the quarter, a four-point acceleration from the prior period. More striking: AWS generated more quarter-over-quarter incremental revenue in dollar terms than Microsoft's Azure for the second consecutive quarter. Its dollar growth was comparable to Google Cloud Platform's entire expansion.

Street estimates project 21% growth for AWS this year and next, but those numbers look conservative. Back-of-the-envelope calculations based on revenue per gigawatt suggest a much higher trajectory. AWS brought 1.2 gigawatts of capacity online in the fourth quarter and plans to double total capacity over two years. If utilisation rates hold, the maths points to growth well above current consensus.

Analysts are expected to revise their models upward over the coming weeks as they work through the numbers with Amazon's team. The disconnect between the capex sticker shock and the underlying growth trajectory may prove temporary, but for now, uncertainty is dominating.

NVIDIA's chips didn't work the way customers expected

A separate story is developing around NVIDIA that has received less attention than it deserves. The company's Blackwell chip, designed to be connected in clusters of 70 to 72 units, caused significant deployment headaches for some of its largest customers.

Major AI companies, including OpenAI, along with cloud providers like Oracle and Meta, struggled to get Blackwell systems operational quickly. Initial delays surfaced in late 2024, followed by reports in early 2025 that chips were not performing as expected in testing. Some customers reduced their orders or reverted to older chip generations.

NVIDIA has since released the GB300, an updated chip that improved performance and eased the growing pains. But the episode raised a broader concern: NVIDIA is releasing chips at a faster cadence than ever before, and customers are being asked to deploy current-generation hardware while next-generation products are already in the pipeline. The Vera Rubin architecture is approaching, and some buyers are questioning whether to invest heavily in Blackwell systems that may soon be superseded.

None of this threatens NVIDIA's dominance. The company has no real competitor for high-end AI training chips, and customers have limited alternatives. But the gap between shipping a product and having that product work reliably at scale is a friction that the market has largely ignored.

The AI model wars are heating up, and nobody is loyal

OpenAI and Anthropic both released major updates to their coding tools this week, with OpenAI's GPT 5.3 Codex focused specifically on code generation and Anthropic's Claude Opus 4.6 offering broader reasoning capabilities. The two products are competitive but serve slightly different use cases, and engineers are increasingly running both simultaneously.

The more interesting dynamic is the absence of brand loyalty. Companies building on top of these models report no allegiance to either lab. The decision of which model to use comes down to performance on specific tasks, cost, and latency. For long-form writing and complex reasoning, Opus 4.6 is currently preferred. For certain code generation workflows, Codex has advantages.

Both companies are deploying "forward-deployed engineers," a term borrowed from military usage via Palantir, to embed technical staff inside enterprise customers. The race for the enterprise wallet requires bodies on the ground, not just better models, and both labs are investing heavily in this approach.

Model capability tends to advance in step functions, with one or two inflection points per year where performance jumps materially. Each leap resets the competitive landscape and forces both companies to respond. The result is a market where the best model changes regularly and customers benefit from the rivalry but struggle to build long-term preferences.

Is AI killing enterprise software, or is this just panic?

The broader question hanging over the market is whether AI will destroy the enterprise software industry. Traders have been selling software stocks aggressively, and the comparison to cord-cutting in entertainment is being made frequently.

Bloomberg's editors offered a more measured take. The cord-cutting analogy is instructive precisely because it took 10 to 15 years to fully play out. The disruption of enterprise software by AI is likely to follow a similarly extended timeline. Some companies will be hurt. Others will adapt. Mergers will happen, particularly as stocks trade at depressed multiples that make recurring-revenue businesses attractive to private equity.

The irony is that when AI companies eventually go public, investors will get a closer look at their actual financials, and those numbers may make the "boring" software companies look appealing by comparison. Many AI startups are burning cash at extraordinary rates with concentrated customer bases and uncertain paths to profitability. NVIDIA remains the only company unambiguously making money from AI.

For now, the market is in a phase of paralysis. Valuations have reset in public markets but are still adjusting in private markets. Boards and private equity firms may see opportunities but remain uneasy about the ultimate direction of the industry. That indecision could stall dealmaking for months.

The software industry is not dying. But the market is pricing in the possibility that it might, and until that fear recedes, the selling is unlikely to stop.