Welcome to Fulgur News. Here's your quick snapshot of what's moving. Bitcoin mining difficulty just posted the largest absolute jump in history, up nearly fifteen percent to 144.4 trillion, after U.S. miners came back online following a brutal winter storm. Abu Dhabi sovereign wealth funds crossed a billion dollars in Bitcoin ETF holdings, buying aggressively into the drawdown. The Supreme Court struck down Trump's IEEPA-based tariffs, potentially unlocking up to 175 billion dollars in refunds that could act as an unintended liquidity injection. And in AI, the open-source versus closed-model debate is basically over — performance has converged so tightly that the real differentiator now is cost and deployment flexibility. Let's get into it.
Let's start with AI models, because something important has happened that's easy to miss in the noise. The performance gap between open-source and closed-source large language models has effectively collapsed. DeepSeek V3 now outscores GPT-4o on the MMLU benchmark, 88.5 to 87.2. And Meta's Llama 5 is scoring 89.2 on MMLU-Pro, which is within spitting distance of GPT-5's 91.1. On coding benchmarks, Llama 5 actually beats GPT-5's initial release with a 92.3 percent pass rate on HumanEval Plus. That is remarkable for an open-source model licensed under Apache 2.0.
The top of the overall leaderboard for February 2026 still has closed models — GPT-5.2 Pro leads, followed by Claude Opus 4.6 and Gemini 3 Pro. But the story isn't who's number one. The story is that the ranking is now so tight that the deciding factor for most real-world deployments is economics, not raw capability.
Llama 5 introduces what Meta calls Mixture-of-Depths architecture, which dynamically allocates compute based on token complexity. Simple stuff gets processed cheaply, hard reasoning problems get more resources. The result is about forty percent less compute during inference compared to brute-force approaches. For an enterprise running millions of queries, that's transformative. J.P. Morgan and Bank of America are already deploying Llama 5 internally, driven partly by performance but mostly by compliance requirements — they want models running on their own infrastructure, not sending sensitive data to third-party APIs.
The framework for choosing models in 2026 has shifted from 'which model is smartest' to a multi-dimensional analysis: reasoning capability, coding performance, knowledge breadth, cost per token, and whether you can actually run the thing on your own hardware. Closed models still have edges in production coding and complex multi-step reasoning. But for a growing number of use cases, the open-source option is not just good enough — it's better, because you control it completely.
Now to Bitcoin mining, where the network just flexed its resilience in a pretty dramatic way. Mining difficulty spiked 14.7 percent to 144.4 trillion at block height 937,440. That's the largest absolute increase ever recorded. What happened? A winter storm across the U.S. knocked roughly 200 exahashes per second offline. Blocks slowed down, difficulty dropped. Then miners reconnected, hashrate surged from 884 to over 1,030 exahashes per second, and the protocol did what it always does — it adjusted. Blocks had been coming in at about eight minutes and 47 seconds, so the network cranked difficulty back up to target that ten-minute average.
But zoom out from the mechanics and the picture gets more nuanced. Miners are in a margin crunch. Post-halving production costs have risen sharply. The average cost to mine a Bitcoin now sits somewhere between 75,000 and 87,000 dollars, though the most efficient operators like MARA and CleanSpark manage around 34,000 to 43,000. With Bitcoin trading below average production cost for many miners, the pressure is real. Historically, this kind of stress has preceded strong returns within about 90 days, but it also forces weaker miners out.
Meanwhile, the convergence of Bitcoin mining and AI infrastructure is accelerating. MARA just acquired a 64 percent stake in Exaion, a French computing infrastructure company, expanding into AI and cloud services. This isn't a side bet — it's a strategic pivot that more miners are making as they compete with AI data centers for cheap energy. The question some are raising is whether mining stays economically viable long-term, or whether these companies gradually become general-purpose compute providers that happen to also mine Bitcoin. By the 2028 halving, projections suggest Bitcoin needs to be somewhere between 90,000 and 160,000 dollars just to maintain current mining operations. The network will survive regardless — it always adjusts — but the composition of who's mining could look very different.
Abu Dhabi just made a statement that's hard to ignore. Two of its sovereign wealth funds — Mubadala Investment Company and Al Warda Investments — have collectively crossed one billion dollars in Bitcoin ETF holdings, specifically in BlackRock's iShares Bitcoin Trust. Mubadala increased its stake by 46 percent through the end of 2025, holding roughly 631 million dollars worth of shares. Al Warda climbed to about 408 million. They were buying while Bitcoin was falling 23 percent in the fourth quarter. That's conviction, not momentum chasing.
Contrast that with Harvard Management Company, which actually reduced its Bitcoin ETF exposure during the same period. And U.S. spot Bitcoin ETFs have now logged five straight weeks of net outflows, totaling 3.8 billion dollars. Institutional investors in the West are de-risking amid macro uncertainty — tariffs, the yen carry trade unwinding, stablecoin liquidity contracting. Small wallets have been accumulating, increasing holdings by 2.5 percent since the October all-time high, but large holders have trimmed by 0.8 percent.
There's a smart contrarian piece making the rounds arguing that the supposed institutional floor under Bitcoin is partly an illusion — that a lot of ETF inflows were driven by basis trade arbitrage rather than genuine long-term allocation. As those yields compressed, the money left. Leveraged funds are heavily short. The Fear and Greed Index is deep in fear territory.
But here's what's interesting: K33 Research says current conditions echo the late 2022 bear market bottom. Retail optimism fading, those loud 150K price calls drying up — Santiment actually flags that as healthy, a reset of sentiment to neutral. And then you have the Supreme Court striking down Trump's IEEPA tariffs, which could trigger up to 175 billion in refunds flowing back into the economy. That's an accidental liquidity event nobody was pricing in. Meanwhile, the Blue Owl Capital liquidity crisis — forced to liquidate 1.4 billion in assets — has some analysts drawing parallels to 2008-style private credit stress, which historically has preceded unconventional monetary responses that tend to benefit hard assets. The setup is messy, but the contrarian case is building.
Stablecoins are quietly becoming the most consequential infrastructure story in crypto, and maybe in payments broadly. Circle's USDC grew 73 percent in 2025, hitting roughly 75 billion in market cap, outpacing Tether's USDT growth of 36 percent for the second consecutive year. Together they still control over 80 percent of the stablecoin market, but the gap in growth rates tells a clear story: institutions are choosing regulated stablecoins.
USDC is fully backed by cash and short-term Treasuries, operates under U.S. state licenses, and is compliant with Europe's MiCA framework. Tether lacks comparable regulatory standing in those jurisdictions. This matters because Visa has integrated USDC into its settlement processes. Stripe, PayPal, and Circle itself are all pushing stablecoin payment use cases for 2026. The SEC just made a quiet but potentially significant move, allowing broker-dealers to treat stablecoins as capital. That's a regulatory shift that could dramatically expand how stablecoins function within traditional finance.
Circle's vision for 2026 is ambitious. They're building Arc1, an open Layer-1 blockchain designed as what they're calling an Economic OS for the internet, plus the Circle Payments Network and StableFX for enterprise foreign exchange. They're positioning themselves not as a crypto company but as financial infrastructure — The Motley Fool actually compared Circle to a utility stock, which tells you something about how the narrative around stablecoins has shifted.
The bigger picture is a bifurcation. You have regulated onshore stablecoins like USDC embedding themselves into institutional workflows — treasury management, B2B payments, settlement. And you have offshore liquidity stablecoins like Tether serving markets where regulatory clarity doesn't exist or isn't wanted. Both have enormous roles to play. But if you're tracking where mainstream adoption is heading, it's following the compliance trail. Banks are entering the stablecoin space now, which introduces competition but also legitimacy. The transition from crypto plumbing to actual payments infrastructure is happening faster than most people expected.
Here's the takeaway to sit with. The two biggest stories right now — AI models converging in capability and stablecoins converging with traditional finance — share the same underlying dynamic. The technology is no longer the bottleneck. Distribution, regulation, and economics are. The best AI model doesn't win; the one you can run on your own terms does. The best stablecoin doesn't win; the one your bank and your regulator will accept does. If you're building, investing, or just paying attention, that's the lens worth applying to everything you're evaluating right now. I'm out. See you next time on Fulgur News.